成果報告書詳細
管理番号20160000000575
タイトル*平成27年度中間年報 次世代ロボット中核技術開発 (次世代人工知能技術分野)プログラミング言語 メニーコアを活用するデータフロー型プログラミング言語の開発
公開日2016/7/27
報告書年度2015 - 2015
委託先名株式会社トプスシステムズ
プロジェクト番号P15009
部署名ロボット・AI部
和文要約
英文要約Next-Generation Robotics Control requires a system powered by Software that realizes advanced Artificial Intelligence. Accordingly, we have proceeded research and study on Programming Language that utilizes parallelisms efficiently by bridging between “Parallelisms inherent in processing of Next-Generation Artificial Intelligence Applications” and “Parallelisms supported by Computing Platforms”, and that improves productivity of development of high performance Artificial Intelligence software and makes it easy for maintenance and extensions of the software.
In the fiscal year of 2015, we started the project with overlooking Applications of Next-Generation Artificial Intelligence, Software Development Environments, Programming Models, Programming Languages, and Computing Platforms for Next-Generation Artificial Intelligence Applications. Especially, we have done research and study on the latest development framework for Artificial Intelligence Applications as well as GP-GPU as a most popular computing platform for such applications, then based on them we chose a specific GP-GPU hardware and development framework, and build a GP-GPU Evaluation Environment, such as TITANX a latest high performance GPU and its software development environment; DIGITS, Caffe, Torch, CUDA, cuDNN, etc.
Also, we have done deep research on programming models and computation models for Dataflow Processing, then defined “DeepPN (Deep Process Network)” that is optimum for stream processing of Massive Data Flow that is specific for Next-Generation Artificial Intelligence Applications. Then, we have started definition of Programming Language Specification for Next-Generation Artificial Intelligence Applications based on the “DeepPN”.
Furthermore, we have done research and study on benchmarks for quantitative comparison of performance between GP-GPU and Manycore for Dataflow Processing, and productivity, maintenance, scalability of the new Programming Language. We chose (1)LeNet、(2)AlexNet、(3)GoogLeNet、(4)Restricted Boltzmann Machine as benchmark application candidates, and confirmed that these applications can be executed for benchmarking on the above GP-GPU Evaluation Environment.
Also, we have procured several types of programming language systems for Artificial Intelligence and Machine Learning, then established evaluation environment of these programming language systems.
In addition to these programming language systems, we have designed and implemented prototype of programming language system based of DeepPN for Dataflow Processing.
ダウンロード成果報告書データベース(ユーザ登録必須)から、ダウンロードしてください。

▲トップに戻る