News

Preferred Networks develops a custom deep learning processor MN-Core for use in MN-3, a new large-scale cluster, in spring 2020

Dec. 12, 2018, Tokyo Japan – Preferred Networks, Inc. (“PFN”, Head Office: Tokyo, President & CEO: Toru Nishikawa) announces that it is developing MN-Core (TM), a processor dedicated to deep learning and will exhibit this independently developed hardware for deep learning, including the MN-Core chip, board, and server, at the SEMICON Japan 2018, held at Tokyo Big Site.  


With the aim of applying deep learning in the real world, PFN has developed the Chainer (TM) open source deep learning framework and built powerful GPU clusters MN-1 and MN-1b, which support its research and development activities. By using these clusters with the innovative software to conduct large-scale distributed deep learning, PFN is accelerating R&D in various areas, such as autonomous driving, intelligent robots, and cancer diagnosis and increasing efforts to put these R&D results to practical use.

To speed up the training phase in deep learning, PFN is currently developing the MN-Core chip, which is dedicated and optimized for performing matrix operations, a process characteristic of deep learning. MN-Core is expected to achieve a world top-class performance per watt of 1 TFLOPS/W (half precision). Today, floating-point operations per second per watt is one of the most important benchmarks to consider when developing a chip. By focusing on minimal functionalities, the dedicated chip can boost effective performance in deep learning as well as bringing down costs.

  • Specifications of the MN-Core chip
    • Fabrication Process : TSMC 12nm
    • Estimated power consumption (W) : 500
    • Peak performance (TFLOPS) :   32.8(DP) / 131(SP) / 524 (HP)
    • Estimated performance per watt (TFLOPS / W) : 0.066 (DP)/ 0.26(SP) / 1.0(HP)

(Notes) DP: double precision, SP: single precision, HP: half precision

https://projects.preferred.jp/mn-core/en/

 

Further improvement in the accuracy and computation speed of pre-trained deep learning models is an essential prerequisite for PFN to work on more complex problems that remain unsolved. It is therefore important to make continued efforts to increase computing resources and make them more efficient. PFN plans to build a new large-scale cluster loaded with MN-Cores, named MN-3, with plans to operate it in the spring of 2020. MN-3 comprises more than 1,000 dedicated server nodes, and PFN intends to increase its computation speed to a target of 2 EFLOPS eventually.

For MN-3 and subsequent clusters, PFN aims to build more efficient computing environments by making use of MN-Core and GPGPU (general-purpose computing on GPU) according to their respective fields of specialty.   

Furthermore, PFN will advance the development of the Chainer deep learning framework so that MN-Core can be selected as a backend, thus utilizing both software and hardware approaches to drive innovations based on deep learning.

 

PFN’s self-developed hardware for deep learning, including MN-Core, will be showcased at its exhibition booth at the SEMICON Japan 2018.

  • PFN exhibition booth at SEMICON Japan 2018
    • Dates/Time: 10:00 to 17:00 Dec. 12 – 14, 2018
    • Venue: Booth #3538, Smart Applications Zone, East Hall 3 at Tokyo Big Site
    • Exhibits:
      (1)  Deep Learning Processor MN-Core, Board, Server
      (2) Preferred Networks Visual Inspection
      (3) Preferred Networks plug&pick robot

 

*MN-Core (TM) and Chainer (TM) are the trademarks or the registered trademarks of Preferred Networks, Inc. in Japan and elsewhere.

Preferred Networks releases ChainerX, a C++ implementation of automatic differentiation of N-dimensional arrays, integrated into Chainer v6 (beta version) for higher computing performance

Dec. 3, 2018, Tokyo Japan – Preferred Networks, Inc. (“PFN”, Head Office: Tokyo, President & CEO: Toru Nishikawa) releases ChainerX, a C++ implementation of automatic differentiation of N-dimensional arrays for the Chainer™ v6 open source deep learning framework. Chainer v6 will run without the need to change most of the code used in previous versions.

Since the release of its source code in 2015, the development of Chainer, known as a pioneer of flexible and intuitive deep learning frameworks, has been very active and attracted many users. Many other deep learning frameworks have followed suit in adopting Chainer’s Define-by-Run method, demonstrating the foresight of Chainer. Chainer’s pure Python implementation policy has, on the one hand, contributed to the legibility and simplicity of code, but on the other, it was becoming a bottleneck due to increased overhead of the Python execution system relative to the overall runtime as its performance improved.

Therefore, the release of ChainerX, which is written in C++ and integrated into the main Chainer, is a first step in achieving higher performance without losing much of Chainer’s flexibility and backward compatibility for many users.

 

 

Main features of ChainerX are:

  • C++ implementation in close connection with Python – NumPy, CuPy™, and automatic differentiation (autograd), all of which are mostly written in Python, have been implemented in C++

The logic of matrix calculation, convolution operations, and error backpropagation has all been implemented in C++ to reduce CPU overhead by Python by up to 87% (comparison of overhead measurements only)

 

  • Easy to work with CPU, GPU, and other hardware backends

Replaceable backends have increased portability between devices

 

Figure:In addition to the multidimensional array implementation which corresponds to NumPy/CuPy, the Define-by-Run style automatic differentiation function is covered by ChainerX.

 

 

As well as improving ChainerX performance and expanding the backend, PFN plans to enable models written in ChainerX to be called from non-Python environments.

For more details on Chainer X, developer Seiya Tokui is scheduled to give a presentation at NeurIPS, a top conference in machine learning (formerly called NIPS), in Montreal, Canada this month.

Dec. 7, 12:50-02:55 Open Source Software Showcase:

http://learningsys.org/nips18/schedule.html

 

Chainer has adopted a number of development proposals from external contributors. PFN will continue to quickly adopt the results of the latest deep learning research and promote the development and popularization of Chainer in collaboration with supporting companies and the OSS community.

 

  • About the Chainer™ Open Source Deep Learning Framework

Chainer is a Python-based deep learning framework developed and provided by PFN, which has unique features and powerful performance that allow for designing complex neural networks easily and intuitively, thanks to its “Define-by-Run” approach. Since it was open-sourced in June 2015, as one of the most popular frameworks, Chainer has attracted not only the academic community but also many industrial users who need a flexible framework to harness the power of deep learning in their research and real-world applications.

Chainer quickly incorporates the results of the latest deep learning research. With additional packages such as ChainerRL (reinforcement learning), ChainerCV (computer vision), and Chainer Chemistry(a deep learning library for chemistry and biology)and through the support of Chainer development partner companies, PFN aims to promote the most advanced research and development activities of researchers and practitioners in each field. (http://chainer.org/

Preferred Networks releases the beta version of Optuna, an automatic hyperparameter optimization framework for machine learning, as open-source software

Dec. 3, 2018, Tokyo Japan – Preferred Networks, Inc. (“PFN”, Head Office: Tokyo, President & CEO: Toru Nishikawa) has released the beta version of Optuna™, an open-source automatic hyperparameter optimization framework.

In deep learning and machine learning, it is essential to tune hyperparameters since they control how an algorithm behaves. The precision of a model largely depends on tuning the hyperparameters. The number of hyperparameters tends to be high especially in deep learning. They include the numbers of training iterations, neural network layers and channels, learning rate, batch size, and others. Nevertheless, many deep learning researchers and engineers manually tune these hyperparameters and spend a significant amount of their time doing so.

Optuna automates the trial-and-error process of optimizing the hyperparameters. It automatically finds optimal hyperparameter values that enable the algorithm to give excellent performance. Optuna can be used not only with the Chainer™ open-source deep learning framework, but also with other machine learning software.

 

Main features of Optuna are:

  • Define-by-Run style API

Optuna can optimize complex hyperparameters while maintaining high modularity.

  • Pruning of trials based on learning curves

Optuna predicts the result of training with an iterative algorithm based on a learning curve. It halts unpromising trials to enable an efficient optimization process.

  • Parallel distributed optimization

Optuna supports asynchronous distributed optimization and simultaneously performs multiple trials using multiple nodes.

 

Optuna is used in PFN projects and with good results. One example is the second place award in the Google AI Open Images 2018– Object Detection Track competition. PFN will continue to develop Optuna, while prototyping and implementing advanced functionalities.

 

 

* Chainer™ and Optuna™ are the trademarks or the registered trademarks of Preferred Networks, Inc. in Japan and elsewhere.

Mitsui and Preferred Networks to Establish Joint Venture to Provide Biomedical/Healthcare Solutions, Including Cancer Diagnostic Service, Based on Deep Learning Technology

Nov. 15, 2018, Tokyo Japan – Mitsui & Co., Ltd. (“Mitsui”, Head Office: Tokyo, President and Chief Executive Officer: Tatsuo Yasunaga) and Preferred Networks, Inc. (“PFN”, Head Office: Tokyo, President & CEO: Toru Nishikawa) entered into an agreement on November 15, 2018 to establish a new joint venture in the United States to commercialize deep learning based biomedical/healthcare solutions. The CEO of the new joint venture will be Nobuyuki Ota, COO of Preferred Networks America, Inc.

 

In recent years, deep learning technology has made significant progress in providing innovation across various industries. These innovations have been notably recognized in the fields of biomedical/healthcare, through applications in drug discovery, diagnostics and treatment, and other related areas.

The global market for deep learning based biomedical/healthcare solutions is expected to grow continuously. PFN has set cancer diagnostics as one of its key focus areas, and by combining PFN’s technology in cancer diagnostics with Mitsui’s network in the healthcare field, including hospital groups, related assets, as well as business partners, the joint venture aims to accelerate the development of the business, and implementation of the technology into society.

Upon the establishment of the joint venture, Mitsui and PFN will accelerate research and development of deep learning based biomedical/healthcare solutions for the early detection of cancer, aiming to resolve societal issues.

Comments from Satoshi Tanaka,
Representative Director and Executive Vice President of Mitsui & Co., Ltd.

Healthcare related businesses have been positioned as a growth area in our Medium-term Management Plan. We are delighted to collaborate with PFN, a leading deep learning technology company, to develop innovative technologies and expand its business overseas.

 

Comments from Toru Nishikawa,
President and CEO of Preferred Networks, Inc.

Since 2014, PFN has been continuously engaged in research and development to apply deep learning technology to the field of medicine. We are extremely pleased to be able to extend our partnership with Mitsui in launching collaborative projects that utilize the results of these R&D efforts, including those in a cancer diagnosis, in the United States.

 

Comments from Nobuyuki Ota,
COO of Preferred Networks America, Inc.

We aim to develop an early cancer diagnostics service based on deep learning technology. Our goal is to save people who potentially suffer from cancer through early detection, and ultimately establish a preventive healthcare platform.

 

 

Profile of Joint venture (Tentative)

Name of Company Preferred Medicine, Inc.
Head Office 330 Primrose Road., Burlingame, CA, USA, 94010
Establishment November 2018
CEO Nobuyuki Ota
Capital US$1 million (Mitsui 50%, PFN 50%)
Business activities Development and operation of biomedical/healthcare solutions, initially focusing on cancer diagnostics based on deep learning as well as other types of machine learning technology

Preferred Networks and PFDeNA launch joint research project to develop a deep learning-based system to detect 14 types of cancers with a small amount of blood

Aim to bring to market by 2021 to extend healthy life expectancy with early cancer detection

Oct. 29, 2018, Tokyo Japan – Preferred Networks, Inc. and PFDeNA Inc. will start research and development to create a blood test system that utilizes deep learning technology to detect 14 types of cancers*1 in their early stages.

In this R&D initiative, PFN and PFDeNA will use blood samples (DNA repository samples) and clinical information, both collected by the National Cancer Center Japan (NCC) for research purposes with donor consent. PFDeNA will measure the expression levels of ExRNA*2 in the DNA repository samples by using a next-generation sequencer*3 in a manner that does not identify individuals. PFN will apply deep learning technology to learn, evaluate, and analyze the measurements together with clinical data. The aim is to put the resulting system to practical use, which will be able to accurately determine the presence or absence and kind of cancer based on ExRNA expression levels in the blood.

 

Social background

Cancer is the leading cause of death among Japanese people, with about one in two developing cancer in their lifetimes. The number of Japanese who died from cancer is more than 370,000 a year and continuing to rise. This amounts to one out of every 3.6 deaths being caused by cancer *.

Even though it is critical to detect cancer at an early stage, screening rates for various types of cancers remain at roughly 30%, one of the lowest among developed countries. Each type of cancer has its own screening methods and requires different areas and organs in our bodies to be tested. The level of accuracy differs from one test to another. The burden of taking these tests need to be reduced both physically and financially in order to improve the screening rates.

Against this backdrop, many studies have been reported recently on gene expression of ExRNAs which include miRNAs*4, bringing to light miRNA expressions that are unique biomarkers of cancer in each organ. Because the types or numbers of miRNAs expressed in bodily fluids will change once a person has cancer, researchers have high expectations that it will become easier to diagnose cancers using easily-collectible bodily fluids, such as blood.

 

Going forward

After PMDA’s*5 review and approval, PFN and PFDeNA aim to develop the results of this research into a business by 2021 and promote its widespread use in Japan.

The high-precision, low-impact screening system will require only a small amount of blood to detect 14 types of cancers in their early stages and is expected to become a common cancer test in the future. Through early cancer detection, PFN and PFDeNA will contribute to efforts to decrease the mortality rate, to reduce medical costs, to extending healthy life expectancy and increasing cancer screening rates in Japan.

 

*1 The 14 types of cancers covered in this research are stomach cancer, colon cancer, esophageal cancer, pancreatic cancer, liver cancer, bile duct cancer, lung cancer, breast cancer, ovarian cancer, cervical cancer, uterine cancer, prostate cancer, bladder cancer, and kidney cancer.

*2 ExRNA is an RNA existent in the blood and other bodily fluids, mainly miRNA (microRNA) in this research. miRNA helps regulate a variety of biological activities and is expected to be used as a diagnostic biomarker.

*3 A next-generation sequencer is a piece of equipment used to sequence the base pairs of human genes in parallel at high speed.

*4 miRNA is a ribonucleic acid that is about 20 bases long and plays a role in regulating gene expression.

*5 PMDA is an acronym for the Pharmaceuticals and Medical Devices Agency of Japan, which is an organization that conducts the scientific review for quality, efficacy, and safety of pharmaceuticals and medical equipment. https://www.pmda.go.jp/english/about-pmda/outline/0005.html

*Source: “Summary of Vital Statistics for 2017” (Ministry of Health, Labour and Welfare)

Preferred Networks releases version 5 of both the open source deep learning framework, Chainer and the general-purpose array calculation library, CuPy.

Preferred Networks, Inc. (PFN, President and CEO: Toru Nishikawa) has released Chainer(TM) v5 and CuPy(TM) v5, major updates of PFN’s open source deep learning framework and general-purpose array calculation library, respectively.

In this major upgrade after six months, Chainer has become easier to use after integrating with ChainerMN, which has been provided as a distributed deep learning package to Chainer. The latest v5 will run as-is on most of the code used in previous versions.

 

Main features of Chainer v5 and CuPy v5 are:

  • Integrated with the ChainerMN distributed deep learning package

・With ChainerMN incorporated in Chainer, fast distributed deep learning on multiple GPUs can be conducted more easily.

  • Support for data augmentation library NVIDIA(R)

・Chainer v5 performs faster data preprocessing by decoding and resizing of JPEG images on GPUs.

  • Support for FP16

・Changing to half-precision floating-point (FP16) format is possible with minimal code changes.

・Reduced memory consumption, which allows larger batch sizes.

・Further speed increases with the use of NVIDIA(R) Volta GPU Tensor Cores.

  • Latest Intel(R) Architecture compatibility

・Chainer v5 supports the latest version 2 of Chainer Backend for Intel(R) Architecture (previously, iDeep, which was added to Chainer v4) for faster training and inference on Intel(R) Processors.

  • High-speed computing and memory saving for static graphs

・Chainer v5 optimizes computation and memory usage by caching static graphs that do not change throughout training. This speeds up training by 20-60%.

  • Enhanced cooperation with Anaconda Numba and PyTorch, enabling the mutual exchange of parallel data

・Added ability to pass a CuPy array directly to a JIT-compiled function by Anaconda Numba.

・DLpack:Array data can be exchanged with PyTorch and other frameworks.

  • CuPy basic operations are 50% faster

・Performance of basic operations such as memory allocation and array initialization has improved.

 

Chainer and CuPy have incorporated a number of development results from external contributors. PFN will continue to quickly adopt the results of the latest deep learning research and promote the development and popularization of Chainer and CuPy in collaboration with supporting companies and the OSS community.

 

◆ About the Chainer(TM) Open Source Deep Learning Framework

Chainer is a Python-based deep learning framework developed and provided by PFN, which has unique features and powerful performance that allow for designing complex neural networks easily and intuitively, thanks to its “Define-by-Run” approach. Since it was open-sourced in June 2015, as one of the most popular frameworks, Chainer has attracted not only the academic community but also many industrial users who need a flexible framework to harness the power of deep learning in their research and real-world applications.

Chainer quickly incorporates the results of the latest deep learning research. With additional packages such as ChainerRL (reinforcement learning), ChainerCV (computer vision), and Chainer Chemistry(a deep learning library for chemistry and biology)and through the support of Chainer development partner companies, PFN aims to promote the most advanced research and development activities of researchers and practitioners in each field. (http://chainer.org/

Preferred Networks unveils a personal robot system at CEATEC Japan 2018, exhibiting fully-autonomous tidying-up robots

Oct. 15, 2018, Tokyo Japan – Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) will unveil a fully-autonomous tidying-up robot system, which is currently under development, at the CEATEC Japan 2018 exhibition held in Makuhari Messe near Tokyo. A technical demonstration of the system will be given at the event.

PFN is developing technology to create a society where robots can actively support our daily living activities. Unlike in controlled and regulated environments like factories, robots in the home need to respond flexibly to dynamic and complex situations and communicate naturally with humans.

At its exhibition booth (A060), PFN will demonstrate the new robot system using HSRs (Human Support Robots) developed by Toyota Motor Corporation and showcase their ability to keep a cluttered room neat and tidy. This has been difficult to achieve using conventional technologies based on object recognition and robot control. The deep learning-based robots can recognize various scattered household items like clothes, toys, and stationery, grasp and place them in their designated locations. In the demonstration, PFN will also show that these cleaning robots can be controlled intuitively through verbal and gestural instructions.

For technical details, please visit the website.
https://projects.preferred.jp/tidying-up-robot/en

 


The fully-autonomous tidying-up robot system has been awarded the Semi-Grand Prix in the Industry/Market category at CEATEC Award 2018, which recognizes innovative technologies, products, and services from among a large number of exhibits at CEATEC JAPAN 2018.

 

  • PFN exhibition booth

・Dates/time:10:00–17:00 Oct. 16–19, 2018

・Location:Booth A060, Total Solutions Area, International Exhibition Hall 2

・Exhibit:Technical demonstration of personal robots “fully-autonomous, tidying-up robot system” (first public exhibition)

 

In addition, PFN President and CEO Toru Nishikawa will make a keynote speech entitled “Robots for Everyone” on the opening day of the CEATEC exhibition. As well as introducing the outlook on future technologies, he will explain how PFN is applying cutting-edge technologies of machine learning, deep learning, and robotics to solve real-world problems.

  • CEATEC Keynote Future
  • ・Date/Time: 12:30~13:15 Tuesday on Oct. 16, 2018・Location:Convention Hall, International Conference Hall, Makuhari Messe・Speaker:Preferred Networks President and CEO Toru Nishikawa・Title/outline:”Robots for Everyone”The possibilities of robots are rapidly expanding thanks to the advancement of machine learning technology. Fusing the machine learning technology with robotics is essential for making robots that can flexibly respond to unexpected situations and execute various tasks like a human. Soon, we will begin to see an increasing number of robots helping to perform tasks in many places, working alongside humans. As well as introducing the current technology and PFN’s new initiative, President Nishikawa will provide insight into how we can leverage technology in the new era of these kinds of robots and the outlook on future technologies.

Preferred Networks hired Professor Takeo Igarashi of The University of Tokyo as a Technical Advisor

On August 1st, 2018, Preferred Networks, Inc. (PFN, HQ: Chiyoda-ku, Tokyo, President & CEO: Toru Nishikawa) hired Takeo Igarashi (Professor at The University of Tokyo) as a technical advisor.

Professor Igarashi is a pioneer in human-computer interaction (HCI), user interface and computer graphics research. He has published many research papers at international conferences, notably Teddy, a ground-breaking technique to create 3D models from 2D sketches. Currently, he is the project leader of the JST CREST research project “HCI for Machine Learning”, which aims to make machine learning techniques more accessible and easier to use.

In this role as a technical advisor, Professor Igarashi will provide technical advice and guidance on HCI and HRI (Human-Robot Interaction) research at PFN, with a view to accelerating the development and adoption of PFN’s technology.

For more information, please refer to the research blog.

Prof. Takeo Igarashi

Takeo Igarashi

  • Biography

Takeo Igarashi is a Professor at the Computer Science Department of The University of Tokyo. He received a Ph.D from the Department of Information Engineering at The University of Tokyo in 2000. He then worked as a post doctoral research associate at Brown University (2000 – 2002). He joined the University of Tokyo as an Assistant Professor in 2002, and became a Full Professor in 2011. He also served as a director for the JST ERATO Igarashi Design Interface project (2007 – 2013). His research interest is in user interfaces and interactive computer graphics. He is known for the development of a sketch-based 3D modeling system (Teddy) and a performance-driven animation authoring system (MovingSketch). He has received several awards including the IBM Science Prize, the JSPS Prize, the ACM SIGGRAPH 2006 Significant New Researcher Award, and the Katayanagi Prize in Computer Science. He served as conference co-chair for ACM UIST 2016, program co-chair for ACM UIST 2013, program chair for ACM SIGGRAPH ASIA 2018 technical papers, associate editor for ACM Transactions on Graphics and program committee member for various international conferences including ACM CHI, UIST, and SIGGRAPH.

Preferred Networks releases deep learning-based, high-precision, visual inspection software

 

The software will make it possible to construct an inspection system quickly and inexpensively with minimal training datasets

 

Oct. 11, 2018, Tokyo Japan – Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) has developed Preferred Networks Visual Inspection,  high-precision, visual inspection software based on deep learning technology. PFN will start licensing the software to partner companies in December 2018. We will announce the new software at a new product seminar (N3-5) on Thursday Oct.18 during the CEATEC Japan 2018 exhibition held in Makuhari Messe near Tokyo.

The use of machine learning and deep learning technologies is spreading rapidly in many areas including the manufacturing floor. However, existing visual inspection systems based on deep learning require as many as several thousand images for training, as well as engineers to annotate the considerable number of images to facilitate the training process. Poorly explained inspection results are also among other issues that have been tackled.

In order to solve these problems, PFN has utilized its technical know-how acquired through the development of the deep learning framework Chainer(TM) and applications of deep learning to our main business domains – transportation systems, manufacturing, and bio-healthcare – to develop the Preferred Networks Visual Inspection.

  • The main features of Preferred Networks Visual Inspection:
  1. An inspection line can be set up with a small amount of training data (as few as 100 images of normal products and 20 images of defective products)
  2. Plastic, metal, cloth, food, and other materials with various shapes can be handled
  3. Results are well-explained through visualized anomalies such as scratches, foreign objects, and stains
  4. Training is made easy even for non-engineers with intuitive user interfaces

 

Preferred Networks Visual Inspection consists of a training support tool and CPU-based defect detection software. Depending on requirements, our licensed partners will install a combination of system components which include training workstations, inspection PCs, photographing equipment, UIs for visualization and operation. GPU-based, fast detection software is also available as an option.

The new product will enable users to build an easy-to-use and highly reliable auto-inspection system at a low cost in a short period of time. This product can be introduced with ease to the manufacturing lines which have been difficult to automate by existing products due to their high costs and inflexibility. In addition, defects are visualized so that its results can be easily explained. This is useful for passing down inspection skills and sharing knowledge with others in the company.

Comparison of Preferred Networks Visual Inspection and the existing solutions

  • New product announcement

PFN will announce Preferred Networks Visual Inspection at a New Technologies and Products Seminar (N3-5) entitled “Visual inspection system and picking robot solution based on deep learning” at CEATEC Japan in Makuhari Messe.

 

PFN will continue to promote practical applications of machine learning and deep learning technologies in the real world.

Preferred Networks wins second place in the Google AI Open Images – Object Detection Track, competed with 454 teams

Sept. 7, 2018, Tokyo Japan – Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) participated in the Google AI Open Images – Object Detection Track, an object detection challenge hosted by Kaggle*1, and won second place in the competition among 454 teams from around the world.

 

Object detection, which is one of the major research subjects in computer vision, is a basic technology that is critical for autonomous driving and robotics. Challenges in using large-scale datasets, such as ImageNet and MS COCO, to achieve better accuracy in object detection have been the unifying force of the research community, contributing to the rapid improvement of detection techniques and algorithms.

 

The Google AI Open Images – Object Detection Track held between July 3, 2018 and August 30, 2018 was a competition of an unprecedented scale that used Open Images V4*2, a large and complex dataset released by Google this year. As a result, the event attracted the attention of many researchers. A total of 454 teams from around the world participated in the competition.

PFN entered the competition as team “PFDet”, comprising interested members, mainly developers of ChainerMN and ChainerCV, PFN’s distributed deep learning library and computer vision library based on deep learning, respectively, as well as specialists in the fields of autonomous driving and robotics. During the competition, PFN’s large-scale cluster MN-1b that has 512 NVIDIA (R) Tesla(R) V100 32GB GPUs was in full operation for the first time since its launch in July this year. In addition, the team utilized a parallel deep learning technique to speed up training with a large-scale dataset and made full use of research results PFN had accumulated over the years in the fields of autonomous driving and robotics.  These efforts resulted in the team finishing in a close second place by a narrow margin of 0.023% behind the team who won first place.

 

We have published a paper, entitled “PFDet: 2nd Place Solution to Open Images Challenge 2018 Object Detection Track,” regarding our solution method in this competition, at https://arxiv.org/abs/1809.00778

We also plan to present the content of the paper at a workshop at the European Conference on Computer Vision (ECCV)2018.

 

A part of the techniques developed for this competition will be released as additional functionality to ChainerMN and ChainerCV.

 

PFN will continue to work on research and development of image analysis and object detection technologies, and promote their practical applications in our three primary business domains, namely, transportation, manufacturing, and bio/healthcare.

 

*1:A platform for machine learning competitions

*2:A very large training dataset comprised of 1.7 million images (including 12 million objects of 500 classes)

Preferred Networks will exhibit at CEATEC JAPAN 2018 with CEO Toru Nishikawa scheduled to make a keynote speech titled “Robots for Everyone”

Aug. 3, 2018, Tokyo Japan – Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) will set up an exhibit booth and unveil its new initiative at CEATEC Japan 2018 which will be held at the Makuhari Messe convention center in Chiba from Oct.16-19, 2018. PFN President and CEO Toru Nishikawa will also make a keynote speech titled “Robots for Everyone” on the first day of the event.   

 

  • CEATEC Keynote Future

・Date/time:12:30-13:15 on Tuesday, Oct. 16, 2018

・Venue:Makuhari Messe international convention complex

・Speaker: President and CEO Toru Nishikawa, Preferred Networks, Inc.

・Title and outline:

Robots for Everyone
The advancement of machine learning technology is rapidly expanding the possibilities of what robots can do. Fusing the machine learning technology with robotics is essential for making robots that can flexibly respond to unexpected situations and execute various tasks much like a human. Shortly, we will begin to see an increasing number of robots helping perform tasks in many places, working alongside humans. PFN President Nishikawa will talk about how we can leverage current technologies in the new era of robots, provide some insight into what lies ahead, and introduce PFN’s new initiative.

 

 

  • PFN booth

・Duration: Tuesday, Oct. 16 to Friday, Oct. 19, 2018

・Exhibit zone:Total Solutions (Booth No. : A060)

・What will be exhibited?:PFN’s new initiative (to be shown to the public for the first time)

Preferred Networks raises a total of about 900 million yen in capital from Chugai Pharmaceutical and Tokyo Electron

July 26, 2018, Tokyo Japan – Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) has agreed to receive investments of about 700 million yen from Chugai Pharmaceutical Co., Ltd. (Chugai, Headquarters: Chuo-ku, Tokyo, President and CEO: Tatsuro Kosaka) and about 200 million yen from Tokyo Electron Ltd. (TEL, Headquarters: Minato-ku, Tokyo, President & CEO: Toshiki Kawai) through its subsidiary in August 2018.

PFN will use the raised capital to strengthen its financial base, improve computing resources, and continue recruiting talented people.

PFN and Chugai have established a comprehensive partnership to develop innovative drugs and services that create new value. PFN and Chugai will collaborate on joint projects to solve open problems in drug research and development as well as those for more exploratory initiatives by utilizing deep learning technology. Also, PFN and TEL have begun joint research on applications of deep learning to such areas as optimization and automation in semiconductor manufacturing.

PFN will strive to drive innovation not only in the fields of transportation system, manufacturing, and bio/healthcare but in a broad range of business areas to increase its corporate value.

 

We have received the following comments from Chugai and TEL:

 

“The fusion of existing and new technologies such as IoT and AI will become essential in all value-chain activities including research and development in the domain of healthcare and life science. By applying PFN’s cutting-edge data analysis techniques, such as machine learning and deep learning, to Chugai’s overall business operations centering on “technology-driven drug discovery,” we are aiming to deliver innovative drugs and services that address high unmet medical needs and contribute to the medical community and human health around the world.”

Osamu Okuda
Executive Vice President of Chugai in charge of Project & Lifecycle Management (Marketing) and Corporate Planning

 

“I have high expectations that PFN and TEL will be able to fuse the world’s most advanced technologies of deep learning and chipmaking to produce an epoch-making result that leads to innovation in semiconductor manufacturing.”

Toshihiko Nishigaki
Deputy General Manager, Corporate Innovation Division, Corporate Marketing, Information Technology

 

Related link:

Chugai Enters into Comprehensive Partnership Agreement with Preferred Network
https://www.chugai-pharm.co.jp/english/news/detail/20180726153001.html

*Company names and product names written in this release are the trademarks or the registered trademarks of each company.

Office expansion

Preferred Networks, Inc. (hereinafter PFN) has expanded its office space in Otemachi head office to accommodate the increasing number of employees as we continue to expand our business.

New office location (3rd floor, Otemachi Building)

In addition to a comfortable workspace that enhances individual productivity, our new office has a group work area to facilitate communication among team members, a Japanese-style sitting area with tatami mats, a counter area to enjoy coffee and catered food, among others. With a variety of workspaces added, we aim to build a more comfortable working environment where our employees can collaborate and develop innovative ideas more easily.

With this expansion, we have also set up a manned reception desk for visitors on the third floor of the Otemachi Building and increased the number of meeting rooms as well.  

Workspace with electric, height-adjustable desks

Group work area that can be rearranged to accommodate small to large groups

Sitting area with tatami traditional Japanese floor covering

Counter area

Catered lunch

Manned reception desk (Section 352 on the 3rd floor of Otemachi Building)

 

Preferred Networks received the Best Paper Award on Human-Robot Interaction in ICRA 2018.

ICRA(International Conference on Robotics and Automation), one of the top conferences in Robotics organized by Institute of Electrical and Electronics Engineers (IEEE), was held in Brisbane, Australia from May 21-25, 2018. In this conference, our paper “Interactively Picking Real-World Objects with Unconstrained Spoken Language Instructions” was awarded the Best Paper Award on Human-Robot Interaction (HRI).

At the award ceremony (from left, Sosuke Kobayashi, Jun Hatori, skipping one person, Kuniyuki Takahashi, and Wilson Ko)

After the ceremony (from left, Sosuke Kobayashi, Jun Hatori, Kuniyuki Takahashi, and Wilson Ko)

 

At PFN, we are applying the latest Image processing and natural language processing technologies as a means of communication between humans and robots. Our latest work has succeeded in building an interactive system in which you can use unconstrained spoken language instructions to operate a common object picking task.

PFN will continue to research and develop the cutting-edge technology and promote its application to the industry.

 

The details of the PFN’s paper “Interactively Picking Real-World Objects with Unconstrained Spoken Language Instructions” and video are available on the following website.

https://pfnet.github.io/interactive-robot/

Chainer awarded the Open Source Data Science Project Award Winner at the ODSC East 2018

The Open Source Data Science Project award is given in recognition for the significant contribution to the field of data science. Winners in previous years were the Pandas Project and scikit-learn.

Chainer, an open source deep learning framework, won the award this year, in the recognition of its dynamic and flexible neural network definition by “define-by-run”.

 

 

Chainer is evaluated for the award as follows:
Chainer strives to “bridge the gap between algorithms and deep learning implementations” in its flexible and intuitive Python-based framework for neural networks. Chainer was the first framework to provide the “define-by-run” neural network definition which allows for dynamic changes in the network. Since flexibility is a significant part of the foundations of Chainer, the framework allows for customization that similar platforms do not so easily provide and supports computations on either CPUs or GPUs.

https://opendatascience.com/odsc-east-2018-open-source-data-science-project-award-winner-the-chainer-framework/

 

About the Open Data Science Conference (ODSC)

ODSC is a conference for people to connect with the data science community and contribute to the open source applications they use every day. Its goal is to bring together the global data science community to help foster the exchange of innovative ideas and encourage the growth of open source software.

 

 

About the Chainer Open Source Deep Learning Framework

Chainer is a Python-based deep learning framework developed mainly by PFN, which has unique features and powerful performance that allow for designing complex neural networks easily and intuitively, thanks to its “Define-by-Run” approach. Since it was open-sourced in June 2015, as one of the most popular frameworks, Chainer has attracted not only the academic community but also many industrial users who need a flexible framework to harness the power of deep learning in their research and real-world applications.
Chainer incorporates the results of the latest deep learning research. With additional packages such as ChainerMN (distributed learning), ChainerRL (reinforcement learning), ChainerCV (computer vision) and through the support of Chainer development partner companies, PFN aims to promote the most advanced research and development activities of researchers and practitioners in each field. (http://chainer.org/)

 

Preferred Networks Executive Appointments

Preferred Networks (hereinafter referred to as PFN) introduces corporate officers and a PFN Fellow system as part of its new initiative to research and develop various technical elements in a wide range of areas and drive its expanded business forward.

The introduction of the new system is aimed at enhancing PFN corporate culture by providing more growth opportunities for younger generations, supported by experienced staff, as well as enabling it to make swift decisions and act quickly while maintaining the flat hierarchy in its rapidly growing organization as much as possible. In addition, an employee who is highly respected by people both inside and outside the company for his/her significant contribution to research for many years can be appointed as a PFN Fellow.  

Through the proper delegation and transfer of responsibilities, PFN will continue to move forward with its efforts to further grow as a team and become a sustainable organization where each individual plays a responsible role in business and research and builds relationships of trust with each other.

 

  • Directors

Toru Nishikawa, Representative Director & President

Daisuke Okanohara, Representative Director & Executive Vice President

Ryosuke Okuta, Director

 

  • Corporate Officers

    Takuya Akiba

    Daisuke Okanohara

    Ryosuke Okuta

    Masakazu Takahashi

    Toru Nishikawa

    Junichi Hasegawa

    Kiyoshi Yamamoto

 

  • PFN Fellow

    Hiroshi Maruyama

Preferred Networks released open source deep learning framework Chainer v4 and general-purpose array calculation library CuPy v4.

Tokyo, Japan, April 17, 2018 — Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) has released v4 of Chainer™ and CuPy™, major updates of the open source deep learning framework and the general-purpose array calculation library, respectively.

This major upgrade to Chainer and CuPy incorporates the results of the latest deep learning research over the last six months. The newly released v4 is largely compatible with previous versions of Chainer.

 

Main features of Chainer and CuPy v4 include:

  • Additional functions for fast, memory-efficient training on NVIDIA(R) GPUs *1

Chainer now supports NVIDIA TensorCore to speed up convolutional operations. Loss scaling has also been implemented to alleviate the vanishing gradient problem when using half-precision floats.

  • Quick installation of CuPy

We have begun providing a binary package of CuPy to reduce the installation time from 10 minutes down to about 10 seconds.

  • Optimized for Intel(R) Architecture

An Intel Deep Learning Package (iDeep) *2 backend has been added to make training and inference on Intel CPUs faster. This delivers an 8.9-fold improvement of GoogLeNet (a neural network used for image recognition) inference speed on CPUs, according to our benchmark results*3.

  • More functions supporting second order differentiation

Enhanced support for second order differentiation, which was first introduced in v3, allows easier implementation of the latest networks and algorithms.

  • A new function to export results of training with Chainer in the Caffe format

A function to export Chainer’s computational procedure and learned weights in the Caffe format has been added as experimental. This makes it easier to use the results of training with Chainer even in an environment where Python cannot be executed. (Exporting into the ONNX format is also available via the onnx-chainer package.)

 

◆Chainer ReleaseNote: https://github.com/chainer/chainer/releases/tag/v4.0.0

◆Update Guide:https://docs.chainer.org/en/latest/upgrade.html

 

Chainer and CuPy have taken in a number of development results from external contributors. PFN will continue working with supporting companies and the OSS community to promote the development and popularization of Chainer and CuPy.

 

* 1:http://docs.nvidia.com/deeplearning/sdk/mixed-precision-training/index.html

* 2:NumPy-compatible library for performing general arithmetic operations in deep learning at a high speed on Intel CPUs https://github.com/intel/ideep

* 3:The results of comparison in time to process an image between when iDeep was enabled and disabled. Intel Math Kernel Library was enabled in both cases. Intel Xeon(R) CPU E5-2623 v3 was used.

 

About the Chainer Open Source Deep Learning Framework

Chainer is a Python-based deep learning framework developed mainly by PFN, which has unique features and powerful performance that allow for designing complex neural networks easily and intuitively, thanks to its “Define-by-Run” approach. Since it was open-sourced in June 2015, as one of the most popular frameworks, Chainer has attracted not only the academic community but also many industrial users who need a flexible framework to harness the power of deep learning in their research and real-world applications.

Chainer incorporates the results of the latest deep learning research. With additional packages such as ChainerMN (distributed learning), ChainerRL (reinforcement learning), ChainerCV (computer vision) and through the support of Chainer development partner companies, PFN aims to promote the most advanced research and development activities of researchers and practitioners in each field. (http://chainer.org/

FANUC’s new AI functions utilizing machine learning and deep learning

Tokyo, Japan, April 16, 2018 — FANUC CORPORATION (hereinafter, FANUC) in collaboration with Preferred Networks, Inc. (hereinafter, PFN) has developed new AI functions that apply machine learning or deep learning to its FA, ROBOT, and ROBO-MACHINE products.

 

FA:AI Servo Tuning (Machine Learning)

FANUC has developed AI Feed Forward as the first to come out of its development efforts in a group of AI Servo Tuning functions that realize high-speed, high-precision, high-quality machining. It utilizes machine learning to easily tune parameters for controlling servo motors in a sophisticated manner. AI Feed Forward is a feed-forward controller based on a high-dimensional model that represents mechanical characteristics more accurately. This model has too many parameters to tune manually as has been done up to now. Therefore, machine learning is used in the process to determine parameters for this advanced feed-forward control. AI Feed Forward offers high-quality machining as it reduces mechanical vibration caused when servo motors accelerate or decelerate.

Shipment estimated to start in April 2018

 

ROBOT:AI Bin Picking (Deep Learning/FIELD System Application)

 FANUC released AI Bin Picking FIELD application with 3D object scoring function to identify suitable picking order with higher success rate. This Deep Learning based application enables FANUC Robot Bin Picking system to learn the picking order automatically, and reduces robot user’s burden of the lengthy manual setup process. Also, this function makes FANUC Robot to pick up the object with higher success rate, which had only been possible with detail parameter tuning by experienced operators. Picking success rate can be even improved by creating Deep Learning trained model for each workpiece type. 

Left: FIELD BASE Pro (with NVIDIA GPU)

Right: Picking robot system with sensor (Demo unit)

Shipment started in April 2018

 

ROBOMACHINE:AI Thermal Displacement Compensation (Machine Learning)

FANUC has developed and begun selling an AI thermal displacement compensation function for FANUC’s ROBODRILL series, following the release of the same AI function for its wire-cut electric discharge machine ROBOCUT in November last year. The second AI function is for ROBOMACHINE and utilizes machine-learning technology to predict and compensate for the thermal displacement caused by temperature fluctuations, which are detected by the thermal sensors measuring ambient temperatures as well as ROBODRILL’s temperature rise while in motion. Machining accuracy has improved by about 40%, compared with an existing function. Furthermore, the optimal placement of the thermal sensors and the effective use of thermal data enable it to continue to perform optimal compensation without interrupting machining work even if there is sensor malfunction.

ROBOCUT with the first AI thermal displacement compensation function (released on November 2017)

ROBODRILL with the second AI thermal displacement compensation function

Shipment started in March 2018 (already released)

 

Comment from Toru Nishikawa,
President & CEO of Preferred Networks

“We have been working with FANUC on the AI Bin Picking project since the commencement of our R&D alliance in 2015. I feel it is of great significance that we announce its release today as the first product to apply deep learning to robots. We will continue to bring a new value to manufacturing floors by stepping up our efforts to introduce to the market smart robots and machine tools that utilize deep learning in a broader field.”

Call for applications for PFN summer internship 2018

Preferred Networks (PFN) is looking for enthusiastic interns who can work with us in our Tokyo office this summer. Students who participated in the previous programs are also eligible to apply. We welcome students who want to help us develop new technologies, software, and services in a wide range of computer science areas including machine learning.  

 

Important notice:

Note that this program is only for students who already have visa eligibility to work as an intern this summer in Japan. We are not accepting applications from students who need support for obtaining the designated activity visa to work as an intern because the due date for processing such applications has already passed.

 

Guidelines for applicants

 

● Characteristics of PFN Internship

  • Over the two-month period, PFN engineers will be assigned to work with each of you as a mentor. You will have opportunities to discuss and study your theme with specialists in various fields including deep learning, computer vision, natural language processing, robotics, bio-healthcare, reinforcement learning, and distributed processing.
  • After the internship, you can make public your research result by writing a paper or making it OSS, to the extent possible.  

 

● Period

Start date:Between late July and early August depending on your schedule

End date:Friday, Sept.21, 2018

※You can choose to continue to work in the week of Sept. 24-28 under the same terms and conditions.

Note that this year’s internship will end on Sept. 21 in consideration of the fact that many schools start their fall semester in late September. If you need more time to finalize your research or want to spend more time with our staff, you can continue to work until the end of September under the same terms and conditions. We understand you may have school or family commitments during the internship which might range from lab activities to attending academic conferences, to returning home. We are very flexible about your need to take days of absence due to these reasons.

 

● Key Qualifications

PFN is seeking highly motivated and skillful individuals who can develop applications, tools, etc. on your own. Having knowledge or development experience in the themes listed below is a plus but not a must. Minimum requirements are:

  • Currently enrolled in high school, technical college, university, or graduate school. Negotiable for those attending other higher educational institution
  • Fluency in Japanese or English
  • Strong communication skills
  • Prior experience in programming (any language)
  • Willingness to come to work in our Tokyo office on weekdays

Do not hesitate to apply even if you don’t have prior experience in full-scale development.

 

【Important notes before you apply】

  • We are not accepting applications from students who need support for obtaining the designated activity visa to work as an intern because the due date for processing such applications has already passed.
  • You need to let us know in advance for any administrative work required for receiving academic credit from your school. Please note that depending on the complexity of the work, PFN may not be able to accommodate your request.

 

● Place of work

PFN Tokyo Office

Otemachi Bldg. 2F, 1-6-1, Otemachi, Chiyoda-ku, Tokyo, Japan 100-0004

 

● Basic working conditions and benefits

  • Salaries:2,500JPY an hour for a technical college, university, graduate school students. 2,000JPY an hour for high school students
  • Work hours:Eight work hours in principle. Five days a week excluding Saturdays, Sundays, public holidays.  
  • Commuting fee support:PFN will pay for your daily commute to and from office in an approved route.
  • Travel cost:For students traveling a long distance by plain or Shinkansen bullet train to participate in the internship, PFN will support a round trip to relocate to the Tokyo area.
  • Accommodation support:For students coming from distant parts who would take roughly 60 minutes or longer to commute, PFN will provide a housing allowance of 5,000JPY a day covering the entire period of your internship. You need to arrange a place to stay by yourself. Reasonable weekly rental apartments are available near PFN office ranging from 100,000 to 150,000 JPY a month. Please note that the accommodation allowance is taxable.

 

● How to Apply

Go to: Application form

※ Click the above form to apply. To access the application form, you will need to log in with a Google account.

Deadline:By 23:59 Monday, April 30, 2018, Japan time

For inquiries:Send us an email at intern2018@preferred.jp

 

※About your portfolio

Summarize your skills and qualifications freely in a A4 paper to pitch yourself and highlight and showcase some of your best work such as software you have developed, a list of published papers, awards or prizes you have received, programming contests you have participated in, your blog, twitter account, and other social media sites.

 

● Themes

Let us know which of the following areas of study you would like to work on during the internship. We will decide your theme after speaking with you. You must choose your 1st and 2nd preferences in the application form. If you have more than two areas of interest, select the 3rd preference, which is optional.

  1. Theoretical study of Machine learning/deep learning
  2. Computer vision
  3. Deep reinforcement learning
  4. Robotics
  5. Bio-healthcare
  6. HPC and distributed data management for distributed deep learning/deep learning
  7. Natural language processing
  8. Speech processing
  9. VR/AR
  10. Human computer interaction, human machine interaction
  11. Applications of deep learning to animation, creator support
  12. Development of Chainer
  13. Development of area-specific libraries on Chainer
  14. R&D of machine learning algorithms such as anomaly detection
  15. Information visualization tool and front-end development for machine learning
  16. Machine learning research support, cluster management, experiment management system development
  17. Development of dedicated accelerator/processor for deep learning
  18. Development of compiler/optimizer for deep learning
  19. Development of IoT/Edge Heavy Computing platform
  20. Other
  21. (New) R&D of automatic tuning methods for deep learning
  22. (New) Video analytics (sports, etc…)

 

● Selection process

▼First screening

After sending the application by April 30, you will receive two tests: (1) Online self-interview (recording) (2) Coding test. Deadline for completing these tests is  May 14 (subject to change).

▼Interview

An interview will be scheduled sometime during the two weeks starting from June 5. For students living in distant areas, PFN will arrange a video chat such as Skype.

▼Letter of acceptance (by late June)

Preferred Networks to Launch “MN-1b” Private Sector Supercomputer Adopting NVIDIA Tesla V100 32GB GPUs Will expand NTT Com Group’s multi-node GPU platform

TOKYO, JAPAN — Preferred Networks, Inc. (PFN), a provider of IoT-centric deep learning systems, NTT Communications Corporation (NTT Com), the ICT solutions and international communications business within the NTT Group, and NTT Com subsidiary NTT PC Communications Incorporated (NTT PC) announced today that PFN will launch an expanded version of its MN-1 private sector supercomputer equipped with NTT Com and NTTPC’s next-generation GPU platform by July. The new MN-1b supercomputer will adopt the NVIDIA(R) Tesla(R) V100 32GB, that was announced at GTC 2018 on March 27, 2018 (U.S. time).

PFN plans to enhance MN-1 by adding 512 NVIDIA Tesla V100 32GB GPUs and have them up and running by July, with the added GPUs having a theoretical peak performance of about 56 PetaFLOPS1, a massive 56,000 trillion floating-point operations per second, based on a mixed precision floating-point operation2 used in deep learning. This means the expansion alone will contribute to a roughly threefold increase from the current peak.

PFN expects the new supercomputer’s extra high speed and massive processing environment leveraging the latest GPUs will accelerate the real-world applications of its research and development in deep learning and related technologies and thereby strengthen PFN’s global competitiveness. NTT Com and NTT PC will build and operate the multi-node platform leveraging the latest GPUs that meets PFN’s requirements, using their knowledge of intra-GPU communication and waste heat processing.

“We are truly honored that Preferred Networks has chosen NVIDIA Tesla V100 32GB, most advanced data center GPU with 2X the memory, for its next-generation private supercomputer’s computation environment, “MN-1b”. With NTT Com Group’s experience of establishing and managing highly reliable data center services, combined with NVIDIA’s latest high-speed GPUs for deep learning, we sincerely look forward to R&D results in the fields of transportation systems, manufacturing and biotech/healthcare.”
said Masataka Osaki, Vice President of Corporate Sales and NVIDIA Japan Country Manager.

Emmy Chang, Board Director, Supermicro KK and VP of Strategic Sales, Supermicro said
“Preferred Networks is the first in the world to deploy our SuperServer(R) 4029GP-TRT2 equipped with the latest version of Intel(R) Xeon(R) Scalable processors and supporting eight NVIDIA Tesla V100 32GB GPU accelerators,” “Preferred Networks has developed the world-class private supercomputer through cooperative work with NTT Com Group, and Supermicro continues to support them with our latest innovative hardware and solutions.  We are confident that Preferred Networks will achieve new heights with its new private supercomputer.”

 

PFN will use the new MN-1b to raise the speed of its ChainerTM open source deep-learning framework and further accelerate its research and development in fields that require a huge amount of computing resources, namely transportation systems, manufacturing, bio-healthcare, and creativity.

Going forward, NTT Com expects to increasingly support the delivery of AI technologies and related platforms for advanced research and commercialized deep learning, including the AI business initiatives of PFN.

 

Related links:

Chainer:

Enterprise Cloud:

Nexcenter:

 

Notes:

1 A unit measuring computer performance. Peta is 1,000 trillion (10 to the power of 15) and FLOPS is used to count floating-point operations per second. So, 1 PetaFLOPS means that a computer is capable of performing 1,000 trillion floating-point calculations per second.

2 Mixed precision floating-point operation is a method of floating point arithmetic operations with a combination of multiple precisions.

ChainerTM is a trademark or a registered trademark of Preferred Networks, Inc. in Japan and other countries. Other company names and product names written in this release are the trademarks or the registered trademarks of each company.

 

About Preferred Networks, Inc.

Founded in March 2014 with the aim of promoting business utilization of deep learning technology focused on IoT, PFN advocates Edge Heavy Computing as a way to handle the enormous amounts of data generated by devices in a distributed and collaborative manner at the edge of the network, driving innovation in three priority business areas: transportation, manufacturing and bio/healthcare. PFN develops and provides Chainer, an open source deep learning framework. PFN promotes advanced initiatives by collaborating with world leading organizations, such as Toyota Motor Corporation, Fanuc Corporation and the National Cancer Center.

https://www.preferred-networks.jp/

 

About NTT Communications Corporation

NTT Communications provides consultancy, architecture, security and cloud services to optimize the information and communications technology (ICT) environments of enterprises. These offerings are backed by the company’s worldwide infrastructure, including the leading global tier-1 IP network, the Arcstar Universal One™ VPN network reaching over 190 countries/regions, and over 140 secure data centers worldwide. NTT Communications’ solutions leverage the global resources of NTT Group companies including Dimension Data, NTT DOCOMO and NTT DATA.
www.ntt.com | Twitter@NTT Com | Facebook@NTT Com | LinkedIn@NTT Com

 

NTT PC Communications Incorporated

NTTPC Communications Incorporated (NTTPC), established in 1985 is a subsidiary of NTT Communications, is a network service and communication solution provider in Japanese telco market, The company has been the most strategic technology company of the group throughout of years. NTTPC launched the 1st ISP service of the NTT group, so called “InfoSphere” at 1995, and also launched the 1st Internet Data Center and server hosting services of Japan so called “WebARENA” at 1997. NTTPC have always started something new in ICT market.

Preferred Networks’ private supercomputer ranked first in the Japanese industrial supercomputers TOP 500 list

Preferred Networks Launches one of Japan’s Most Powerful Private Sector Supercomputers

Preferred Networks’ automatic coloring service PaintsChainer receives the Excellence Award at the 21st Japan Media Arts Festival

TOKYO, JAPAN — PaintsChainer™, an automatic coloring service developed and provided by Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa), has received the Excellence Award of the Entertainment Division at the 21st Japan Media Arts Festival organized by Agency for Cultural Affairs, Government of Japan.

The Japan Media Arts Festival publicly honors highly creative, artistic works in four fields of media arts: Art, Entertainment, Animation, and Manga. The annual festival was first organized in 1997 and has since recognized and awarded outstanding works. This year, it has received 4,192 entries from 98 countries and regions around the world, the most in its history, and has announced a grand prize winner, four excellence award winners, and three new face award winners in each of the four categories.

PaintsChainer has drawn a huge reaction on Twitter and other social media sites the moment it was released in January 2017 as a free online service that colors black and white sketches automatically. It uses a deep learning technology to recognize faces, clothing, and other background objects on an uploaded sketch image or picture, and color them automatically or based on a specified color. Currently, it provides three different coloring models Tanpopo, Satsuki, and Canna.

https://paintschainer.preferred.tech/index_en.html

The jury cites the significance of providing an automatic coloring platform as a free web service as the main reason for giving the Excellence Award to PaintsChainer.

A commemorative photo of award winners taken at the press conference at the National Art Center, Tokyo. The fourth person from the left in the middle row is Taizan Yonetsuji.

 

PaintsChainer was developed by one of PFN engineers Taizan Yonetsuji, who has commented as follows:

It all started as my personal project, so I could also study deep learning. I feel very honored to receive such an award and am truly grateful to my senior colleagues who have taught me about deep learning, everyone at Preferred Networks who has supported the launch and operation of the service, and all the users who have said PaintsChainer was fun and enjoyed using it. We will make PaintsChainer even greater while continuing to take on other new challenges.

 

● Summary

 

● 21st Japan Media Arts Festival Exhibition of Award-winning Works

All the award-winning works that represent the media arts of the modern times will be shown to the public during the exhibition, which also includes symposia, artists’ talks, workshops, and other related events.

 

※PaintsChainer™ is the registered trademark of Preferred Networks, Inc. in Japan and other countries. Other company names and product names written in this release are the trademarks or the registered trademarks of each company.

Call for applications for PFN AI residency program 2018-2019 in Tokyo

Preferred Networks (PFN) will be organizing an AI residency program in Tokyo for students from outside of Japan in 2018-2019.

We are a growing startup with about 120 members based in Tokyo, Japan, focusing on applying deep learning to industrial problems such as autonomous driving, manufacturing, and bio-healthcare. We are actively developing the deep learning framework Chainer.

We are looking for brilliant students who have expertise in various topics, such as deep learning, reinforcement learning, computer vision, bioinformatics, natural language processing, distributed computing, simulation, etc.

In previous years, by selecting highly capable interns and encouraging them to tackle challenging and important problems, some of our interns were able to have their work published at top conferences such as ICML, ICCV,  ICRA, and ICLR.

This year, we would like to expand our reach to attract talented students around the world and collaborate with them to tackle challenging problems in AI for a longer period, by introducing this AI residency program.

During the residency program, you will have a unique opportunity to collaborate with our excellent research team members at PFN and work on real-world applications of deep learning, while living in Tokyo; one of the most attractive cities in the world.

 

We are looking forward to receiving your applications! Please see below for the instructions.

●Target of this program

  • Students outside of Japan

 

●Work time & Location:

  • Business hours:
    8 hours/day, 5 days/week (excluding national holidays)
  • Location: Center of Tokyo
    Preferred Networks Tokyo office: Otemachi Bldg. 2F, 1-6-1, Otemachi, Chiyoda-ku, Tokyo, Japan 100-0004
    https://www.preferred-networks.jp/en/about

 

●Period & Compensation:

  • The period of the residency program can be flexibly arranged between September 2018 and August 2019, the minimum term is 6 months
  • AI residency program students are paid a competitive salary
  • We will cover residence and travel cost

 

●Requirements:

  • We can only accept Ph.D. students or new graduates who have been accepted for Ph.D. program starting 2018 Fall
    • Applicants are responsible for negotiating with their university for one year leave or deferring the admission to next year
  • Experience in at least one of the technology areas (listed below) other than only attending lectures
    e.g., published a paper, won a competition, part-time work, open source contribution
  • Strong programming skill (any programming language)
  • Fluent in either English or Japanese
  • Able to work fulltime on weekdays at our Tokyo office during the period

 

●Preferred experience & skills:

  • Machine learning and deep learning
  • Experience with NumPy / SciPy / deep learning frameworks
  • Experience with software & service development
  • Experience working with shared codebases (e.g. GitHub / bitbucket / etc)
  • Contribution to open source projects

 

●Candidate themes (subject to change)

   1. Technology areas: Sub-field of machine learning, such as

a. Deep learning theory

b. Reinforcement learning

c. Computer vision

d. Natural language processing

e. Parallel / distributed computing

 

   2. Application areas: Advanced applications, such as

a. Object detection / tracking / segmentation from image / video

b. Robotics / factory automation / predictive maintenance

c. Life science / healthcare / medicine

d. Human machine interaction

e. Design / content creation / visualization

f. Deep learning software (Chainer, CuPy, ChainerMN/CV/RL, etc)

g. Optimization for deep learning hardware

 

●Application information:

  • Resume / CV (PDF format only. Please DO NOT include any personal or private information [e.g., age, race, nationality, religion, personal address, phone number] except name, email address, affiliation)
  • Github account (optional)

 

●How to apply:

  • Please fill the google forms and submit
  • Due: March 20th, 11:59 pm Tuesday (PDT)

  • No late submission will be accepted
  • The review process takes about 6-8 weeks after submission
  • Usually, getting a visa for working in Japan takes up to 3 months

 

●Interview process:

  1. Document review
  2. One-way video interview (webcam, recording)
  3. Skype interview in English or Japanese (multiple times if necessary)

 

If you have questions, please contact us at hr-pfn@preferred.jp (Sorry but no late application is accepted for fairness)

Preferred Networks support the 30th International Olympiad in Informatics held in Japan

Preferred Networks, Inc. (hereinafter referred to as PFN) supports the 30th International Olympiad in Informatics (IOI 2018 Japan) and participating students in the event. IOI2018 will be held in the city of Tsukuba, Ibaraki Prefecture from Sept. 1-8, 2018.

Kazuo Furukawa, Chairman of IOI 2018 JAPAN Organizing Committee

The support from PFN is a great help in our endeavor to ensure that IOI 2018 JAPAN in which we welcome students from various countries and regions around the world will be held smoothly. Receiving support from companies with employees who have participated in the previous contests such as PFN is quite encouraging for future contestants as well. I expect all the participants who represent the next generation to take this opportunity to expand their network and hope that a new wave of technological innovation will be created through the contest.

 

Toru Nishikawa, President and CEO of Preferred Networks

PFN has six employees who participated in the previous IOI contests. Acquiring such high-level skills as problem analysis, design of algorithms, and programing will be a great advantage after entering the world of business. I hope IOI2018 will be a wonderful opportunity for students to feel the great joy of programming and improve their skills through friendly competition with both Japanese and international friends.

 

About International Olympiad in Informatics

International Olympiad in Informatics is one of the international science olympiads that focuses on the field of informatics. Selected groups of students in secondary education from more than 80 countries and regions participate in IOI held every year.
Contestants design algorithms to solve assigned tasks and compete to get the best score based on the performance of their algorithms such as efficiency and quality as well as the programming skill needed to implement the algorithms properly. One of the primary objectives of IOI is to nurture talent who have a network of personal connections around the world and play major roles in the future of advanced IT society by bringing together young students in the same generation gathered from all over the world.

Hakusensha and Hakuhodo DY Digital Announces the Launch of Colorized Manga Products Using PaintsChainer, a Deep Learning Coloring Technology created by Preferred Networks

Tokyo, Japan, February 6, 2018 -Hakusensha Inc. (Headquarters: Chiyoda-Ku, Tokyo; President: Kazuhiko Torishima; “Hakusensha”)and Hakuhodo DY digital Inc. (Headquarters: Minato-Ku, Tokyo; President: Akira Tsuji; “Hakuhodo DY digital”), with the cooperation of Preferred Networks, Inc. (Headquarters: Chiyoda-Ku, Tokyo; President & CEO: Toru Nishikawa, “PFN”), have started distributing and marketing color version manga products with automatic coloring using deep learning technology.

We have customized the automatic coloring service PaintsChainer *1 provided by PFN, to develop a new manga coloring model. The PaintsChainer enables users to enjoy a unique look and feel with color gradations that are only possible with automatic coloring based on deep learning.

 

  • Example of coloring using PaintsChainer

Kekkon X Renai (Akira Hagio)

  • Initial release titles
Kekkon X Renai (Akira Hagio)
Watashitachi XX Shimashita (Asuka Sora)

The initial releases have been distributed through Hakusensha e-net and other major e-book stores since as early as January 24, 2018.

 

Hakusensha have published a large number of Webcomics, such as Love Silky *2, the pioneer of monthly Web magazines, and have been involved in a large number of projects collaborating with CONPYRA *3, which is produced by Hakuhodo DY digital. This project is another product of this collaboration.

In this collaboration, PFN have developed the PaintsChainer manga coloring model, and Hakuhodo DY digital have been working on requirement definitions and specifications and directed the progress of the production. Starting with the first distribution this time, they will release other colored product versions as well.

Hakusensha, Hakuhodo DY digital, and PFN aim to further innovate manga expression and aggressively work on development of new manga production technologies using deep learning.

 

*1 PaintsChainer(R)
PaintsChainer is an online automatic coloring service developed and provided by PFN, which became a big topic on Twitter and other social media sites on its release in January 2017.
After uploading black and white pictures, it colors them automatically or as specified based on deep learning technologies. PaintsChainer provides three different coloring models Tanpopo, Satsuki, and Kanna, free of charge.
Official site: https://paintschainer.preferred.tech

*2 Love Silky
Love Silky is the monthly Web magazine distributed by Hakusensha. It features a large number of manga works for women and it distributes a new issue every third Wednesday of the month. It has been five years since the first release of Love Silky in January 2013, and currently its sixty-first issue is being released in January 2018. It has two sister magazines Love Jossie (published since July 2015), which features manga works for women, and Jossie Bunko (started publication November 2017), which specializes in novels.
Official site: http://www.hakusensha.co.jp/silky_web

*3 CONPYRA
CONPYRA is the author agent business produced by Hakuhodo DY digital.
Currently a lot of new content is generated and issued by users on the digital platform every day. CONPYRA uncovers particularly excellent works using their proprietary data analytics capabilities and produces these works collaborating with the authors/creators based on agent agreements.
In addition to agent work, CONPYRA is promoting the development of a large number of businesses focusing on novels and manga, in collaboration with a variety of publishing companies.
Official site: https://conpyra.com

 

FANUC, Hitachi, and Preferred Networks to establish a joint venture company for the development of Intelligent Edge Systems

Jan. 31, 2018,  Tokyo Japan – FANUC CORPORATION (TSE: 6594, “FANUC”), Hitachi, Ltd. (TSE:6501, “Hitachi”), and Preferred Networks, Inc. (“PFN”) today announced that they have reached an agreement to establish a joint venture company (the “new company”) on April 2, 2018, to develop Intelligent Edge Systems(1) that utilize artificial intelligence (AI) technologies in edge devices in the industrial and social infrastructure field. Yutaka Saito, who is currently Executive Vice President and Executive Officer at Hitachi, and who will be appointed Senior Executive Vice President at FANUC on April 1, will hold a concurrent position as CEO of the new company(2).

 

(1) Intelligent Edge Systems: Systems that use AI as an intermediary between the Cloud and edge devices such as machine tools, industrial machinery, and robots to achieve cyclic, real-time control

(2) Yutaka Saito is scheduled to resign as Executive Vice President at Hitachi on March 31.

 

In recent years, innovations using AI technologies have been expanding rapidly in a variety of fields. In the field of industrial and social infrastructures in particular, AI technologies are expected to play an important role in the part close to edge devices, for example in the case of vehicles and robots.

 

In this backdrop, FANUC, Hitachi, and PFN have agreed to undertake joint development of the world’s most advanced Intelligent Edge Systems in the industrial and social infrastructure field, and to establish a joint venture company for that purpose. The new company will develop these Intelligent Edge Systems by combining FANUC’s technologies and expertise in machine tools and robots, Hitachi’s knowledge of control technologies and other aspects of OT and IT in the front lines of manufacturing, and PFN’s deep learning and distributed computing technologies. After establishing a joint venture company, the three companies will test business potential and create business plans, and then undertake actual system development and expand application fields.

 

Through the new company’s activities, FANUC, Hitachi, and PFN will develop Intelligent Edge Systems and other next-generation control systems as part of efforts to promote collaborative creation aimed at the realization of Society 5.0(3).

 

(3) A human-centered society that balances economic advancement with the resolution of social problems by a system that highly integrates cyberspace and physical space.

 

 

Comment from Yoshiharu Inaba,
Representative Member of the Board, Chairman and CEO of FANUC

“FANUC is planning to continue investing efforts into the development of FIELD systems as part of its activities targeting IoT. We believe that the activities of the joint venture company, as it strives to achieve faster, cyclic, real-time control, will have a positive effect on these development activities. As a company that has specialized in the field of factory automation, we will respond quickly and flexibly to the rapid introduction of IoT in the manufacturing industry, and continue to contribute to the growth and development of this industry in the future. We have great expectations for this joint venture company as a new phase of these activities.”

 

Comment from Toshiaki Higashihara,
President and CEO of Hitachi

“It is a great honor to work with FANUC and PFN in establishing this joint venture company. Through the Social Innovation Business, where we combine knowledge and expertise accumulated through more than 100 years of experience in OT and more than 50 years in IT, Hitachi strives to resolve issues faced by society, and achieve safer, more secure, and more comfortable lifestyles for everyone. Now, by merging the strengths of these three companies, we will develop and offer the world’s most advanced Intelligent Edge Systems, to contribute to advancements in the field of industrial and social infrastructures.”

 

Comment from Toru Nishikawa,
President & CEO of Preferred Networks

“I am very pleased that we are able to begin these new activities so soon after the announcement of our capital tie-up with Hitachi in December of last year. PFN entered a capital tie-up with FANUC in 2015, and since then, we have been working together to create innovative manufacturing bases where machine learning and deep learning technologies are used to establish intelligent links between machine tools, robots, and other devices. Now, by establishing this joint venture company, our three companies will accelerate the development and provision of innovative technologies that leverage each of our respective specialty fields, to further advance this trend on a global scale.”

 

Outline of the joint venture company (Plan)

Corporate Name Intelligent Edge System, LLC
Capital 30 million yen
Investments FANUC: 10 million yen; Hitachi: 10 million yen;

PFN: 10 million yen

CEO Yutaka Saito
Head Office 3580, Shibokusa Aza-Komanba, Oshino-mura, Minamitsuru-gun, Yamanashi Prefecture, Japan
Established April 2, 2018
Outline of Business Conceptual testing and development of Intelligent Edge Systems

 

About FANUC CORPORATION

FANUC CORPORATION, headquartered at the foot of Mt. Fuji, Japan, is the global leader and the most innovative manufacturer of FA, ROBOT and ROBOMACHINE in the world. More than 260 offices in 45 countries, FANUC provides world-class customer service and support.  Since its inception in 1972, FANUC has contributed to the automation of machine tools as a pioneer in the development of computer numerical control equipment. FANUC technology has been a leading force in a worldwide manufacturing revolution, which evolved from the automation of a single machine to the automation of entire production lines.  For more information visit: http://www.fanuc.co.jp/eindex.htm.

 

About Hitachi, Ltd.

Hitachi, Ltd. (TSE: 6501), headquartered in Tokyo, Japan, delivers innovations that answer society’s challenges. The company’s consolidated revenues for fiscal 2016 (ended March 31, 2017) totaled 9,162.2 billion yen ($81.8 billion). The Hitachi Group is a global leader in the Social Innovation Business, and it has approximately 304,000 employees worldwide. Through collaborative creation, Hitachi is providing solutions to customers in a broad range of sectors, including Power / Energy, Industry / Distribution / Water, Urban Development, and Finance / Government & Public / Healthcare. For more information on Hitachi, please visit the company’s website at http://www.hitachi.com.

 

About Preferred Networks, Inc.

Preferred Networks, Inc. (PFN, Headquarters: Tokyo, Japan, President and CEO:  Toru Nishikawa) was founded in March 2014 with the aim of promoting business utilization of deep learning technology focused on IoT. PFN advocates Edge Heavy Computing as a way to handle the enormous amounts of data generated by devices in a distributed and collaborative manner at the edge of the network, driving innovation in various fields, with a focus on three business areas: transportation, manufacturing, and bio/healthcare. PFN develops and provides Chainer(R), an open source deep learning framework. PFN promotes advanced initiatives by collaborating with world leading organizations, such as Toyota Motor Corporation, FANUC CORPORATION, and the National Cancer Center. https://www.preferred-networks.jp/en/

Second call for application for PFN summer internship 2018 in Tokyo

Preferred Networks (PFN) will be organizing internship programs next summer in Tokyo. In order to make the process smooth for students from outside of Japan, we open an early bird application for them.

We are a growing startup with about 110 members based in Tokyo, Japan, focusing on applying deep learning to industrial problems such as autonomous driving, manufacturing, and bio-healthcare. We are actively developing the deep learning framework Chainer.

We look for brilliant students who have expertise on various topics, such as deep learning, reinforcement learning, computer vision, bioinformatics, natural language processing, distributed computing, simulation, etc.

In previous years, by selecting highly capable interns and encouraging them to tackle challenging and important problems, some of the internship results have been published at top conferences such as ICML or workshops at ICRA and ICCV.

During the internship, you will have unique opportunity to collaborate with highly motivated experts for working on real-world applications of deep learning, while staying in Tokyo, one of the most attractive cities in the world.

We are looking forward to receiving your applications, following the instructions below.

 

● Target of this program:

  • Intern Students outside of Japan

 

Work time & Location:

  • Business hours:
    8 hours/day, 5 days/week (excluding national holidays)
  • Location: Center of Tokyo
    Preferred Networks Tokyo office: Otemachi Bldg. 2F, 1-6-1, Otemachi, Chiyoda-ku, Tokyo, Japan 100-0004
    https://www.preferred-networks.jp/en/about

 

Period & Compensation:

  • The period of the internship can be flexibly arranged
    • You can choose the beginning date as any Wednesday from May 8th or later, and the final date from July 27th, August 16th, or September 21st
  • We require a minimum of eight weeks (40 business days), in order to be able to tackle a challenging task
  • Interns are paid a competitive salary
  • We will cover residence and travel cost

 

Requirements:

  • Experience in at least one of the technology areas (listed below) other than lectures
    e.g., published a paper, won a competition, part-time work, open source contribution
  • Strong programming skill (any programming language)
  • Formally enrolled in university or research institute outside of Japan during 2018-2019 school year
  • Fluent in either English or Japanese
  • Able to work fulltime on weekdays at our Tokyo office during the period

 

Preferred experience & skills:

  • Machine learning and deep learning
  • Experience with numpy / scipy / deep learning frameworks
  • Experience with software & service development
  • Experience working with shared codebases (e.g. github / bitbucket / etc)
  • Contribution to open source projects

 

Candidate themes (subject to change)

1. Technology areas: Sub-field of machine learning, such as

a. Deep learning theory

b. Reinforcement learning

c. Computer vision

d. Natural language processing

e. Parallel / distributed computing

 

2. Application areas: Advanced applications, such as

a. Object detection / tracking / segmentation from image / video

b. Robotics / factory automation / predictive maintenance

c. Life science / healthcare / medicine

d. Human machine interaction

e. Design / content creation / visualization

f. Deep learning software (Chainer, CuPy, ChainerMN/CV/RL, etc)

g. Optimization for deep learning hardware

 

Application information:

  • Resume / CV (PDF format only. Please DO NOT include any personal or private information [e.g., age, race, nationality, religion, personal address, phone number] except name, email address, affiliation)
  • Github account (optional)

 

How to apply:

  • Please fill the google forms and submit
  • Due: January 14th 11:59 pm Sunday (PST)

  • No late submission will be accepted
  • The review process takes about 6-8 weeks after submission
  • Usually, getting a visa for working in Japan takes up to 3 months

 

Interview process:

  1. Document review
  2. One-way video interview (webcam, recording)
  3. Skype interview in English or Japanese (multiple times if necessary)

 

If you have questions, please contact us at hr-pfn@preferred.jp (Sorry but no late application is accepted for fairness)

Preferred Networks raises a total of over 2 billion yen from FANUC, Hakuhodo DYHD, Hitachi, Mizuho Bank, and Mitsui & Co.

December 11, 2017, Tokyo Japan – Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) has agreed to allocate new shares in December 2017 to Hakuhodo DY Holdings Inc. (Headquarters: Minato-ku, Tokyo, President and CEO: Hirokazu Toda), Hitachi, Ltd. (Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toshiaki Higashihara), Mizuho Bank, Ltd. (Headquarters: Chiyoda-ku, Tokyo, President and CEO: Koji Fujiwara) and Mitsui & Co., LTD. (Headquarters: Chiyoda-ku, Tokyo, President and CEO: Tatsuo Yasunaga). PFN will receive an investment of about 500 million yen from each of them.

Similarly, FANUC CORPORATION (FANUC, Headquarters: Oshino-mura, Yamanashi Prefecture, Chairman and CEO: Yoshiharu Inaba) has also agreed to acquire additional shares of PFN for about 500 million yen. FANUC and PFN formed an R&D alliance in June 2015 and a capital alliance in August of the same year.

With these deals, PFN will complete a series of fund-raising activities since the additional investment of about 10.5 billion yen from Toyota Motor Corporation, which was announced in August 2017.

PFN will use the raised capital to strengthen its financial base, improve computing resources, and hire talented people while collaborating with each of the investing companies in respective fields for growth over the mid- to long-term.

PFN will strive to drive innovation not only in the fields of transportation system, manufacturing, and bio/healthcare but in a broad range of business areas to increase its corporate value.

 

We received comments from the companies investing in PFN.

 “FANUC will apply Preferred Networks’ cutting-edge machine-learning and deep-learning techniques to our FIELD system* and all the other products to provide the most effective production system that processes information quickly and in real time at the edges of machining and assembly lines in production sites, enabling machines to collaborate with each other in a flexible and smart way.”

Yoshiharu Inaba
Chairman and CEO of FANUC

 

*FANUC Intelligent Edge Link and Drive system is an open platform to optimize manufacturing by interconnecting various devices in production sites and is available for a variety of companies to join.

 

“I have high expectations that we can create innovative solutions and services by combining Preferred Networks’ deep learning and other machine-learning techniques and implementation capabilities, and Hakuhodo DY Group’s proprietary data and creativity in the fields of advertising, marketing, media, and content. We will, in collaboration with Preferred Networks, promote the fusion of creativity and technology for implementation in the real world and realize innovations that make life richer and more enjoyable.”

Hirokazu Toda
President and CEO of Hakuhodo DY Holdings

 

“Hitachi is utilizing leading-edge technologies, including IoT and AI, and enhancing collaborative creation with various partners in the Social Innovation Business. Taking the investment as an opportunity to combine the strengths of both companies such as Preferred Networks’ knowledge of deep learning and Hitachi’s technology and know-how in the integration of OT, IT, and products, we will begin collaborative creation towards the realization of Society 5.0.”

Yutaka Saito
Representative Executive Officer, Executive Vice President and Executive Officer of Hitachi

 

“Preferred Networks is a leading company in the industry that has a high level of AI technology. Through the new partnership with Preferred Networks, Mitsui will not only promote higher operation efficiency and higher value added of our global business assets but also accelerate Digital Transformation activities that lead to the creation of new businesses with the aim of strengthening competitiveness.”

Nobuaki Kitamori
Representative Director, Executive Managing Officer, and Chief Digital Officer of Mitsui & Co.

 

Related link:

Hitachi Invests in Preferred Networks
http://www.hitachi.com/New/cnews/month/2017/12/171211.html

Mitsui to Invest in AI Company Preferred Networks, Inc.
https://www.mitsui.com/jp/en/release/2017/1225186_10832.html

 

 

 

◆ About Preferred Networks, Inc.

Founded in March 2014 with the aim of promoting business utilization of deep learning technology focused on IoT, PFN advocates Edge Heavy Computing as a way to handle the enormous amounts of data generated by devices in a distributed and collaborative manner at the edge of the network, driving innovation in various fields, with a focus on three business areas: transportation, manufacturing, and bio/healthcare. PFN develops and provides Chainer®, an open source deep learning framework. PFN promotes advanced initiatives by collaborating with world leading organizations, such as Toyota Motor Corporation, FANUC CORPORATION, and the National Cancer Center. (https://www.preferred-networks.jp/en/)

 

*Company names and product names written in this release are the trademarks or the registered trademarks of each company.

Preferred Networks’ private supercomputer ranked first in the Japanese industrial supercomputers TOP 500 list

MN-1, a private supercomputer used exclusively by Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) has recorded a LINPACK  performance of about 1.39 PetaFLOPS 2.  As a result, MN-1 is ranked 12th in the world and 1st in Japan among industrial supercomputers in the TOP500 List (http://www.top500.org), which shows the most powerful supercomputers as of November 2017. When including supercomputers for research purposes, it is listed as 91st in the world and 13th in Japan.   

 

 

About PFN’s private supercomputer MN-1   

MN-1 is equipped with a Graphics Processing Unit (GPU) platform from NTT Communications Corporation (NTT Com, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Tetsuya Shoji) and NTT PC Communications Incorporated (NTTPC, Headquarters: Minato-ku, Tokyo, President and CEO: Motoo Tanaka), and contains 1,024 of NVIDIA (R) ’s Tesla (R) multi-node P100 GPUs. It utilizes Mellanox’s Infiniband interconnect to make high-speed distributed deep learning possible using ChainerMN 4, a distributed deep learning package developed by PFN.    

Using MN-1, PFN will further accelerate its research and development activities in various fields that require a huge amount of computing resources such as transportation systems, manufacturing and bio/healthcare.

 

 

 

1.  A benchmark to compare practical operation speed of computers

2.  A unit measuring computer performance. Peta is 1,000 trillion (10 to the power of 15) and FLOPS is used to count floating-point operations per second. So, 1 PetaFLOPS means that a computer is capable of performing 1,000 trillion floating-point calculations per second.

3.  Preferred Networks launches one of Japan’s most powerful private sector supercomputers
https://www.preferred-networks.jp/en/news/pr20170920

4.  Preferred Networks achieved the world’s fastest training time in deep learning
https://www.preferred-networks.jp/en/news/pr20171110

 

About Preferred Networks, Inc.

Founded in March 2014 with the aim of promoting business utilization of deep learning technology focused on IoT, PFN advocates Edge Heavy Computing as a way to handle the enormous amounts of data generated by devices in a distributed and collaborative manner at the edge of the network, driving innovation in three priority business areas: transportation, manufacturing and bio/healthcare. PFN promotes advanced initiatives by collaborating with world leading organizations, such as Toyota Motor Corporation, FANUC CORPORATION, and the National Cancer Center. (https://www.preferred-networks.jp/en/)

 

*Chainer (R)  is the trademark or the registered trademark of Preferred Networks, Inc. in Japan and other countries. Other company names and product names written in this release are the trademarks or the registered trademarks of each company.

Preferred Networks achieved the world’s fastest training time in deep learning, completed training on ImageNet in 15 minutes,using the distributed learning package ChainerMN and a large-scale parallel computer

November 10, 2017, Tokyo – Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) has achieved the world’s fastest training time in deep learning by using its large-scale parallel computer MN-1 1.

With the size of training data and the number of parameters expanding for the sake of better accuracy of deep learning models, computation time is also increasing. It is not unusual to take several weeks to train a model. Getting multiple GPUs to link with one another for faster training is very important to reduce the time spent on trial and error and verification of new ideas, and produce results quickly.

On the other hand, it is generally known in parallel/distributed learning that the accuracy and learning rate of a model decrease gradually with increased GPUs, due to larger batch sizes and GPU communication overhead.

This time, we have improved learning algorithms and parallel performance to address these issues, and used one of Japan’s most powerful parallel computers with 1,024 of NVIDIA(R)’s Tesla(R) multi-node P100 GPUs and leverages Chainer’s distributed learning package ChainerMN 2 for training.

As a result, we completed training ResNet-50 3 for image classification on the ImageNet 4 dataset in 15 minutes, which is a significant improvement from the previously best known result 5.

The research paper on this achievement is available in the following URL under the title “Extremely Large Minibatch SGD: Training ResNet-50 on ImageNet in 15 Minutes”. (https://www.preferred-networks.jp/docs/imagenet_in_15min.pdf)

Based on this research result, PFN will further accelerate its research and development activities in the fields of transportation systems, manufacturing, and bio/healthcare, which require large-scale deep learning.

 

1 One of the most powerful private supercomputer in Japan, contains 1,024 of NVIDIA(R)’s Tesla(R) multi-node P100 GPUs.https://www.preferred-networks.jp/en/news/pr20170920

2 A package adding distributed learning functionality with multiple GPUs to the open source deep learning framework Chainer

3 A network frequently used in the field of image recognition

4 A dataset widely used for image classification

5 Training completed in 31 minutes using Intel(R) Xeon(R) Platinum 8160 x 1,600(Y. You et al. ImageNet Training in Minutes. CoRR,abs/1709.05011, 2017)

 

■ About the Chainer Open Source Deep Learning Framework

Chainer is a Python-based deep learning framework being developed mainly by PFN, which has unique features and powerful performance that allow for designing complex neural networks easily and intuitively, thanks to its “Define-by-Run” approach. Since it was open-sourced in June 2015, as one of the most popular frameworks, Chainer has attracted not only the academic community but also many industrial users who need a flexible framework to harness the power of deep learning in their research and real-world applications.

Chainer incorporates the results of the latest deep learning research. With additional packages such as ChainerMN (distributed learning), ChainerRL (reinforcement learning), ChainerCV (computer vision) and through the support of Chainer development partner companies, PFN aims to promote the most advanced research and development activities of researchers and practitioners in each field. (http://chainer.org/

■ About Preferred Networks, Inc.

Founded in March 2014 with the aim of promoting business utilization of deep learning technology focused on IoT, PFN advocates Edge Heavy Computing as a way to handle the enormous amounts of data generated by devices in a distributed and collaborative manner at the edge of the network, driving innovation in three priority business areas: transportation, manufacturing and bio/healthcare. PFN promotes advanced initiatives by collaborating with world leading organizations, such as Toyota Motor Corporation, Fanuc Corporation, and the National Cancer Center. (https://www.preferred-networks.jp/en/)

*Chainer(R) is the trademark or the registered trademark of Preferred Networks, Inc. in Japan and other countries.

 

 

Preventive Maintenance Feature of Injection Molding Machine Using AI (Deep Learning)

FANUC CORPORATION (hereinafter, FANUC) and Preferred Networks, Inc. (hereinafter, PFN) have jointly developed AI Backflow Monitor that performs preventive maintenance on FANUC’s electric injection molding machine ROBOSHOT α-SiA series. This is the latest example of our joint initiative to apply deep learning to machine tools.

AI Backflow Monitor uses deep learning to evaluate and predict the wear state of ROBOSHOTs consumable parts (non-return valve) to let operators know before a part starts to malfunction. A conventional method requires that operators visually check waveform data for shape changes that indicate backflow of resin to assess the wear and estimate the replacement timing of the valve. The new feature utilizes the deep learning techniques to effectively analyze the waveform and digitize the wear amount, enabling it to notify operators of the best timing to replace the valve.

Additionally, AI Backflow Monitor takes advantage of its Edge Heavy feature to process data mainly on ROBOSHOT-LINKi, not in the cloud.

AI Backflow Monitor will be provided as an optional feature that can improve the operating rate of ROBOSHOT through preventive maintenance. (FANUC plans to begin taking orders in January next year.)

ROBOSHOT with this new feature will be exhibited at International Plastic Fair 2017, which will be held in Makuhari Messe from Oct. 24-28.

FANUC and PFN will continue to work together and make steady progress, step by step, towards realizing innovative and advanced manufacturing fields using AI.

ROBOSHOT α-SiA series: http://www.fanuc.co.jp/en/product/roboshot/index.html

High-level system structure

AI (Machine Learning) Improves Wire-cut EDM Accuracy

FANUC CORPORATION (hereinafter, FANUC) in collaboration with Preferred Networks, Inc. (hereinafter, PFN) has developed an AI thermal displacement compensation function which will improve the machining accuracy of ROBOCUT α-CiB series, FANUC’s wire-cut electric discharge machine (see Note 1 below). ROBOCUT with this function will be the first product using AI since FANUC and PFN began collaborating.

 

FANUC and PFN formed an R&D alliance*1 in June 2015, followed by a capital alliance*2 in August of the same year to promote a joint development of AI functions for the manufacturing industry that can efficiently improve the performance and operation rates of FANUC products. The newly developed function utilizes machine-learning (ML) technology to predict and control the variable machining accuracy caused by ROBOCUT’s temperature fluctuations, with 30% more accurate compensation than existing method. The new function is applicable from small to large workpieces.

The AI thermal displacement compensation function will be provided as an optional function to ROBOCUT, and FANUC plans to start accepting orders in November of this year. FANUC will also display the ROBOCUT with this new function at Mechatronics Technology Japan, which will be held in Port Messe Nagoya on Oct. 18-21, 2017.

ROBOCUT α-CiB series

FANUC is also developing a similar function for the ROBODRILL series that utilizes ML and expect to release it in the near future.
FANUC and PFN will continue making gradual but steady progress towards realizing innovative manufacturing fields through AI.

“It is my pleasure to announce our first product based on the machine-learning technology since the tie-up with FANUC. Through this product, we can demonstrate using ML is effective in optimizing control parameters, which is one of the most important issues facing the manufacturing industry. PFN will continue to contribute to the intelligence of machine tools and robots by applying machine learning and deep learning techniques.”
Toru Nishikawa, Chief Executive Officer of PFN

 

*1 Announcement for R&D alliance with FANUC Corporation
https://www.preferred-networks.jp/en/news/8731
*2 Announcement for capital tie-up between FANUC and PFN
http://www.fanuc.co.jp/en/profile/pr/newsrelease/notice20150821.html

 

Note 1. Wire-cut EDM is a precision and fine shape machining tool that uses discharge phenomenon between the ultrathin wire electrode and the metal workpiece (electric conductor).

 

Preferred Networks released open source deep learning framework Chainer v3 and NVIDIA GPU array calculation library CuPy v2

Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) has released Chainer v3, a major update of the open source deep learning framework Chainer(R), as well as NVIDIA(R) GPU array calculation library CuPy™ v2.

We release a major upgrade of Chainer every three months that quickly incorporates the results of the latest deep learning research. The newly released Chainer v3 will run without the need to change most of your code.

 

Main features of Chainer v3 and CuPy v2 include:

1.  Automatic differentiation of second and higher order derivatives

Chainer now supports automatic differentiation of second order and higher derivatives in many functions. This will enable users to easily implement deep learning methods that require second order differentiation as per equations written in papers.

 

2. Improved CuPy memory allocation

In many neural nets, memory efficiency when using GPUs will improve significantly, and reallocation of memory will be reduced in some cases, increasing speed.

 

3. Sparse matrix support has been added to CuPy

Large-scale graph analysis and natural language processing, which have previously been highly costly to implement on GPUs, can now be implemented more easily thanks to sparse matrix calculation being available on the GPU.

◆ Chainer ReleaseNote: https://github.com/chainer/chainer/releases/tag/v3.0.0

Chainer v3 has taken in a number of development results from external contributors as its previous versions did. PFN will continue working with supporting companies and the OSS community to promote the development and popularization of Chainer.

 

◆ About the Chainer Open Source Deep Learning Framework

Chainer is a Python-based deep learning framework developed by PFN, which has unique features and powerful performance that enables users to easily and intuitively design complex neural networks, thanks to its “Define-by-Run” approach. Since it was open-sourced in June 2015, as one of the most popular frameworks, Chainer has attracted not only the academic community but also many industrial users who need a flexible framework to harness the power of deep learning in their research and real-world applications.

Chainer incorporates the results of the latest deep learning research. With additional packages such as ChainerMN (distributed learning), ChainerRL (reinforcement learning), ChainerCV (computer vision) and through the support of Chainer development partner companies, PFN aims to promote the most advanced research and development activities of researchers and practitioners in each field. (http://chainer.org/)

*Chainer(R) and CuPyTM are the trademark or the registered trademark of Preferred Networks, Inc. in Japan and other countries.

Preferred Networks Launches one of Japan’s Most Powerful Private Sector Supercomputers

Features NTT Com Group’s Cloud-based GPU platform

TOKYO, JAPAN — This September, Preferred Networks, Inc. (PFN), a provider of IoT-centric deep learning systems, NTT Communications Corporation (NTT Com), the ICT solutions and international communications business within the NTT Group, and NTT Com subsidiary NTT PC Communications Incorporated (NTTPC) announced today the launch of a private supercomputer designed to facilitate research and development of deep learning, including autonomous driving and cancer diagnosis.

The new supercomputer is one of the most powerful to be developed by the private sector in Japan and is equipped with NTT Com and NTTPC’s Graphics Processing Unit (GPU) platform, and contains 1,024 of NVIDIA(R)’s Tesla(R) multi-node P100 GPUs. Theoretically, the processing speed of the new supercomputer can reach 4.7 PetaFLOPS—a massive 4,700 trillion floating point operations per second—the fastest levels of any computing environment in Japan.

Overview of the private supercomputer

 

PFN’s deep learning research demands an ultra high-speed, high capacity, state-of-the-art computing environment. Existing GPU platforms require massive electricity supplies, generate excessive heat and offer inadequate network speed. To address these issues, PFN adopted the NTT Com Group’s proven GPU platform, which boasts significantly advanced technology. They additionally leveraged the latest data center design, building a large-scale multi-node platform using ChainerMN, PFN’s technology that significantly accelerates the speed of deep learning by parallelizing calculations over multiple nodes.

NTT Com group has developed and released a multi-node GPU platform on Enterprise Cloud, Nexcenter(TM), a world-leading data center service, which incorporates the group’s extensive know-how in GPU performance maximization.

Following the supercomputer launch, PFN plans to increase the processing speed of its open source deep learning framework Chainer. They will further accelerate their research and development in the field of transportation systems, manufacturing, bio and healthcare industry which require a huge amount of computing resources. PFN will additionally consider the deployment of NVIDIA(R) Tesla(R) V100 GPUs, which are based on next-generation Volta GPU technology. NTT Com group will continue to support PFN’s research and the commercialization of their developed solutions with AI-related technologies and platforms.

“NVIDIA is excited to see the launch of Preferred Networks’ private supercomputer, built in partnership with NTT Com Group. Computing power is the source of competitive advantage for deep learning, the core technology of modern AI. We have high expectations that the new system will accelerate Preferred Networks’ business and contribute to Japan’s economic growth.”

Masataka Osaki
NVIDIA Japan Country Manager, Vice President of Corporate Sales

 

Related links:

Chainer
Enterprise Cloud
Nexcenter

 

◆ About Preferred Networks, Inc.
Founded in March 2014 with the aim of promoting business utilization of deep learning technology focused on IoT, PFN advocates Edge Heavy Computing as a way to handle the enormous amounts of data generated by devices in a distributed and collaborative manner at the edge of the network, driving innovation in three priority business areas: transportation, manufacturing and bio/healthcare. PFN develops and provides Chainer, an open source deep learning framework. PFN promotes advanced initiatives by collaborating with world leading organizations, such as Toyota Motor Corporation, Fanuc Corporation, and the National Cancer Center.
https://www.preferred-networks.jp/en/

◆ About NTT Communications Corporation
NTT Communications provides consultancy, architecture, security and cloud services to optimize the information and communications technology (ICT) environments of enterprises. These offerings are backed by the company’s worldwide infrastructure, including the leading global tier-1 IP network, the Arcstar Universal One™ VPN network reaching over 190 countries/regions, and 140 secure data centers worldwide. NTT Communications’ solutions leverage the global resources of NTT Group companies including Dimension Data, NTT DOCOMO and NTT DATA.
www.ntt.com | Twitter@NTT Com | Facebook@NTT Com | LinkedIn@NTT Com

◆ NTT PC Communications Incorporated
NTTPC Communications Incorporated (NTTPC), established in 1985 is a subsidiary of NTT Communications, is a network service and communication solution provider in Japanese telco market, the company has been the most strategic technology company of the group throughout of years. NTTPC launched the 1st ISP service of the NTT group, so called “InfoSphere” at 1995, and also launched the 1st Internet Data Center and server hosting services of Japan so called “WebARENA” at 1997. NTTPC have always started something new in ICT market.
http://www.nttpc.co.jp/english/

 

 

Notes
1. Chainer(R) is the trademark or the registered trademark of Preferred Networks, Inc. in Japan and other countries.

2. Other company names and product names written in this release are the trademarks or the registered trademarks of each company.

Call for application for PFN summer internship 2018 in Tokyo

Preferred Networks (PFN) will be organizing internship programs next summer in Tokyo. In order to make the process smooth for students from outside of Japan, we open an early bird application for them.

We are a growing startup with about 100 members based in Tokyo, Japan, focusing on applying deep learning to industrial problems such as autonomous driving, manufacturing, and bio-healthcare. We are actively developing the deep learning framework Chainer.

We look for brilliant students who have expertise on various topics, such as deep learning, reinforcement learning, computer vision, bioinformatics, natural language processing, distributed computing, simulation, etc.

In previous years, by selecting highly capable interns and encouraging them to tackle challenging and important problems, some of the internship results have been published at top conferences such as ICML or workshops at ICRA and ICCV.

During the internship, you will have unique opportunity to collaborate with highly motivated experts for working on real-world applications of deep learning, while staying in Tokyo, one of the most attractive cities in the world.

We are looking forward to receiving your applications, following the instructions below.

 

● Target of this program

  • Students outside of Japan

 

Work time & Location:

  • Business hours:
    8 hours/day, 5 days/week (excluding national holidays)
  • Location: Center of Tokyo
    Preferred Networks Tokyo office: Otemachi Bldg. 2F, 1-6-1, Otemachi, Chiyoda-ku, Tokyo, Japan 100-0004
    https://www.preferred-networks.jp/en/about

 

Period & Compensation:

  • The period of the internship can be flexibly arranged though it usually starts in June and finish by the end of August
  • We require minimum of two months (40 business days), in order to be able to tackle a challenging task
  • Interns are paid a competitive salary
  • We will cover residence and travel cost

 

Requirements:

  • Experience in at least one of the technology areas (listed below) other than lectures
    e.g., published a paper, won a competition, part-time work, open source contribution
  • Strong programming skill (any programming language)
  • Formally enrolled in university or research institute outside of Japan during 2018-2019 school year
  • Fluent in either English or Japanese
  • Able to work fulltime on weekdays at our Tokyo office during the period

 

Preferred experience & skills:

  • Machine learning and deep learning
  • Experience with numpy / scipy / deep learning frameworks
  • Experience with software & service development
  • Experience working with shared codebases (e.g. github / bitbucket / etc)
  • Contribution to open source projects

 

Candidate themes (subject to change)

1. Technology areas: Sub-field of machine learning, such as

     a. Deep learning theory

     b. Reinforcement learning

     c. Computer vision

     d. Natural language processing

     e. Parallel / distributed computing

 

2. Application areas: Advanced applications, such as

     a. Object detection / tracking / segmentation from image / video

     b. Robotics / factory automation / predictive maintenance

     c. Life science / healthcare / medicine

     d. Human machine interaction

     e. Design / content creation / visualization

     f. Deep learning software (Chainer, CuPy, ChainerMN/CV/RL, etc)

     g. Optimization for deep learning hardware

 

Application information:

  • Resume / CV (PDF format only. Please DO NOT include any personal or private information [e.g., age, race, nationality, religion, personal address, phone number] except name, email address, affiliation)
  • Github account (optional)

 

How to apply:

  • Please fill the google forms and submit
  • Due: September 29th, 11:59 pm Friday (PST)

  • No late submission will be accepted (we are planning to open the 2nd call for application by January, and another call for students in Japan by May)
  • The review process takes about 6-8 weeks after submission
  • Usually, getting a visa for working in Japan takes up to 3 months

 

Interview process:

  1. Document review
  2. One-way video interview (webcam, recording)
  3. Skype interview in English or Japanese (multiple times if necessary)

 

If you have questions, please contact us at hr-pfn@preferred.jp (Sorry but no late application is accepted for fairness)

Preferred Networks officially released ChainerMN version 1.0.0, a multi-node distributed learning package, making it even faster with stablized data-parallel core functions

Tokyo, Japan, September 1, 2017 – Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) has released the official version 1.0.0 of ChainerMN※1, which is a package adding distributed learning functionality with multiple GPUs to Chainer, the open source deep learning framework developed by PFN.

For practical application of machine learning and deep learning technologies, the ever-increasing complexity of neural network models, with a large number of parameters and much larger training datasets requires more and more computational power to train these models.

ChainerMN is a multi-node extension to Chainer that realizes large-scale distributed deep learning by high-speed communications both intra- and inter-node. PFN released the beta version of ChainerMN on May 9, 2017 and this is the first official release. The following features have been added to ChainerMN v1.0.0.

● Features of ChainerMN v1.0.0

1. Increased stability in core functions during data parallelization

With this improved stability, ChainerMN can be used more comfortably.

2. Compatibility with NVIDIA Collective Communications Library (NCCL) 2.0.

By supporting the latest version, it has become even faster.

3. More sample code (machine translation, DCGAN) is available.

These examples will help users learn more advanced ways of using ChainerMN.

4. Expansion of supported environments (non-CUDA-Aware MPI).

CUDA – Aware MPI implementation such as Open MPI and MVAPICH was necessary for the beta version, but ChainerMN is now compatible with non-CUDA-Aware MPI.

5. Initial implementation of model parallelism functions.

More complex distributed learning has become possible by getting multiple GPUs to work in the model parallelism method.
The conventional data parallelism approach is known to limit the possible batch size when increasing the nodes while maintaining accuracy. o overcome this, we have done the initial part of the more challenging implementation of model parallelism for greater speed than possible with data parallelism.

 

These features provide a more stable and faster than ever deep learning experiences with ChainerMN, as well as improved usability.

The following is the result of the performance measurement of ChainerMN using the image classification dataset of ImageNet. It is about 1.4 times faster than the first announcement in January this year, and 1.1 times faster than the beta version released in May. Please visit the following Chainer Blog to learn more about the experiment settings:

https://chainer.org/general/2017/02/08/Performance-of-Distributed-Deep-Learning-Using-ChainerMN.html

 

In addition, from October 2017, ChainerMN will become available on “XTREME DNA”, an unmanned  cloud-based super-computer deployment and operation service, provided by XTREME Design Inc.(Head office: Shinagawa-ku, Tokyo, CEO: Naoki Shibata)

ChainerMN will be added on the distributed parallel environment templates for GPU instances of the pay-per-load public cloud, Microsoft Azure. This not only eliminates the need to construct infrastructure required for large-scale distributed deep learning but also makes it easy to manage research-and-development costs.

ChainerMN aims to provide an environment in which deep learning researchers and developers can easily concentrate on the main parts of research and development including the design of neural networks. PFN will continue to improve ChainerMN by adding more features and expanding its usage environment.

 

◆ The Open Source Deep Learning Framework Chainer (http://chainer.org)
Chainer is a Python-based deep learning framework developed by PFN, which has unique features and powerful performance that enables users to easily and intuitively design complex neural networks, thanks to its “Define-by-Run” approach. Since it was open-sourced in June 2015, as one of the most popular frameworks, Chainer has attracted not only the academic community but also many industrial users who need a flexible framework to harness the power of deep learning in their research and real-world applications.

 

※1:MN in ChainerMN stands for Multi-Node. https://github.com/pfnet/chainermn

Regarding the additional investment made by Toyota Motor Corporation announced on August 4.

Thank you for your inquiries concerning additional investment made by Toyota Motor Corporation announced on August 4.
We received a lot of questions about the shareholding ratio, so allow me to clarify.

The shareholding of Toyota Motor, which will be changed as a result of additional investment in PFN, will not change our existing business and management policy.

Preferred Networks received about 10.5 billion yen in investments from Toyota Motor Corporation

Preferred Networks received about 10.5 billion yen in investments from Toyota Motor Corporation. Accelerate joint research and development of AI technology in the mobility field

Tokyo, Japan, August 4, 2017 – Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) agreed to receive an additional investment of approximately 10.5 billion yen from Toyota Motor Corporation (Toyota, Headquarters: Toyota  City, Aichi Prefecture, President: Akio Toyoda). PFN will allocate new shares to Toyota. After the allocation, Toyota will be the largest external shareholder of PFN.

PFN and Toyota started the joint research and development in October 2014, and Toyota invested 1.0 billion yen in PFN in December 2015 to strengthen the relationship.

Through the past joint research and development of object recognition technologies and vehicle information analysis, PFN’s world-leading intelligent technologies in machine learning, deep learning, and big data processing have been highly appreciated as essential for Toyota aiming at realizing the next generation of mobility society, including autonomous driving.

With this funding, PFN will enhance the computational environment and accelerate talent acquisition. We will further accelerate joint research and development by strengthening our relationship with Toyota in the field of mobility.

 

◆ About Preferred Networks, Inc. https://www.preferred-networks.jp/

Founded in March 2014 with the aim of business utilization of deep learning technology focused on IoT. PFN advocates Edge Heavy Computing as a way to handle the enormous amount of data generated by devices in a distributed and collaborative manner at the edge of the network and realizes innovation in the three priority business areas of the transportation system, manufacturing industry, and bio/healthcare.

PFN develops and provides Chainer, and open source deep learning framework, and solutions based on the Deep Intelligence in Motion (DIMo) platform that provides state-of-the-art deep learning technology. PFN promotes advanced initiatives by collaborating with world leading organizations, such as Toyota Motor Corporation, Fanuc Corporation, National Cancer  Center.

*Chainer® and DIMoTM are trademarks of Preferred Networks, Inc. in Japan and other countries.

 

Preferred Networks recieved the “Emerging Leader Award” at the 2017 Japan-U.S. Innovation Award

Preferred Networks recognized for pioneering application of Deep Learning Technology

    

 Receiving 2017 Emerging Leader Award (Japan) with Dr. Richard Dasher, Director, Stanford US-Asia Technology Management Center and Hiroshi Maruyama, Chief strategy officer of PFN.

Preferred Networks, Inc. was selected as the “2017 Emerging Leader Award”, the winner of the 2017 Japan-U.S. Innovation Award held at Stanford University on Friday, July 28th. PFN was one of the “Innovation ShowCase” start-up companies in 2016, and the rapid growth in the past year was highly appreciated.

US-Japan Innovation Awards is operated by the Japan Society of Northern California in collaboration with the Stanford University US–Asia Technology Management Center.

“Emerging Leader Awards” are presented to young, dynamic, and growing, entrepreneurial companies founded upon innovations that are demonstrating impressive success and that have the potential to make substantial global impacts. One U.S. and one Japanese award recipients are selected each year through a rigorous process guided by a distinguished 40-person Innovation Advisory Council composed of venture capitalists, academic experts, and prominent business executives based in Japan and in San Francisco/Silicon Valley.

Preferred Networks appoints Professor Pieter Abbeel of UC Berkeley as a Technical Advisor

Preferred Networks, Inc. (PFN, HQ: Chiyoda-ku, Tokyo, President & CEO: Toru Nishikawa) agreed with Pieter Abbeel (Professor at University of California, Berkeley, and Research Scientist at OpenAI) that he will be appointed as a technical advisor at PFN as of August 1st, 2017.

Professor Abbeel is a pioneer in machine learning based optimization and automation in robotics. He has published many research papers on robot control using deep reinforcement learning. Recently he joined OpenAI, a non-profit AI research company as research scientist.

In this appointment as a technical advisor, Professor Abbeel will give technical advice and guidance on cutting-edge deep learning to robotics research in PFN, aiming to accelerate the development and practical use of PFN’s technology. 

For more information, please refer to the research blog: https://preferredresearch.jp/2017/07/22/abbeel/

 

  • Pieter Abbeel

  • Biography

    He received a BS/MS in Electrical Engineering from KU Leuven (Belgium) and received his Ph.D. degree in Computer Science from Stanford University in 2008. He joined the faculty at UC Berkeley in Fall 2008, with an appointment in the Department of Electrical Engineering and Computer Sciences. His research focuses on robotics, machine learning and control. Professor Abbeel has won various awards, including IEEE Robotics and Automation Society Early Career Award, Sloan Research Fellowship. His work has been featured in many popular press outlets, including BBC, New York Times, and MIT Technical Review.

*DIMo™ are trademarks of Preferred Networks, Inc. in Japan and other countries.

Preferred Networks released Version 2 of Chainer, an Open Source framework for Deep Learning

New functions developed, including a significant increase of memory efficiency during learning

Tokyo, Japan, June 2, 2017 – Preferred Networks, Inc. (PFN, Headquarters: Chiyoda-ku, Tokyo, President and CEO: Toru Nishikawa) has released a major update of its open source deep learning framework Chainer, called Chainer v2.

This is the first major version update since the official release of Chainer in 2015, and it enables more powerful, flexible, and intuitive functions to implement and study deep learning methods.

With the rapid evolution of deep learning technology and an expanding field of target applications, user demands for the functionality of deep learning frameworks is rapidly changing and diversifying.

Chainer incorporates the results of the latest deep learning research. With additional packages such as ChainerMN (distributed learning), ChainerRL (reinforcement learning), ChainerCV (computer vision) and through the support of Chainer development partner companies, PFN aims to promote the most advanced research and development activities of researchers and practitioners in each field.

 

Chainer v2 has three major enhancements and improvements.

 

1. Improved memory efficiency during learning

Chainer v2 shows significantly reduced memory usage without sacrificing learning speed. It has been confirmed that the memory usage can be reduced by 33% or more when learning using the network ResNet50 used in the field of image recognition. This makes it easier to design larger networks and allows to learn using larger batch sizes in usual networks.

 

2. Chainer’s accompanying array library CuPy has been separated and made into an independent project, allowing a broader range of HPC applications to be easily accelerated using GPUs

The general-purpose array calculation library CuPy is highly compatible with library NumPy, which is very popular in the field of scientific computing, making it possible to run faster using the GPU without altering the code written for use with NumPy. By separating CuPy and developing it as a separate library, we aim to increase users for expanding our application not only in deep learning field but also in other research and development fields.

 

3. Organized the API and made it more intuitive

One of the major features of Chainer is its ability to intuitively describe a complex neural network as a program. We have taken into consideration the various use cases and needs of the community to remove unnecessary options and organize interfaces to provide a more sophisticated API. Due to a more intuitive description, unintentional bugs occur less frequently.

 

● Chainer Release Note: https://github.com/chainer/chainer/releases/tag/v2.0.0

● Chainer Upgrade Guide: https://docs.chainer.org/en/stable/upgrade.html

● Chainer Blog: https://chainer.org/announcement/2017/06/01/released-v2.html

 

The Chainer team plans to release major version updates every four months to support the most advanced research and development activities for researchers and practitioners in each field.

Development results of many external contributors are also included in the Chainer V2 release. PFN will continue to work with support companies and the OSS community to promote the development and dissemination of Chainer.

 

◆ Chainer Meetup # 05

Community event for developers and researchers who use Chainer.

  • Date: June 10, 2017, 14:00–18:30
  • Place: Microsoft Japan Co., Ltd. Shinagawa Office, Seminar Room A
    (Shinagawa Grand Central Tower 31f, 2-16-3 Konan, Minato-ku, Tokyo)
  • Application: https://chainer.connpass.com/event/57307/

 

◆ Kick-off for the Deep Learning Lab Community

The Deep Learning Lab is a community of professionals who are well versed in both technology and business to apply the latest deep learning technology to real business. Microsoft Azure and Chainer are used in key platforms/frameworks to disseminate information about use examples and the latest technology trends.

  • Date: Monday, June 19, 2017, 9:00-12:30
  • Place: Microsoft Japan Co., Ltd. Shinagawa Office
    (Shinagawa Grand Central Tower 31F, 2-16-3 Konan, Minato-ku, Tokyo)
  • Application: https://dllab.connpass.com/event/57981/

 

◆ About the Chainer Open Source Deep Learning Framework (http://chainer.org)

Chainer is a Python-based deep learning framework developed by PFN, which has unique features and powerful performance that enables users to easily and intuitively design complex neural networks, thanks to its “Define-by-Run” approach. Since it was open-sourced in June 2015, as one of the most popular frameworks, Chainer has attracted not only the academic community but also many industrial users who need a flexible framework to harness the power of deep learning in their research and real-world applications.

*Chainer(R) and DIMo(TM) are trademarks of Preferred Networks, Inc. in Japan and other countries.