Posts on Apr 2018

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.”