本文へジャンプ

成果報告書詳細
管理番号20190000000694
タイトル*平成30年度中間年報 高効率・高速処理を可能とするAIチップ・次世代コンピューティングの技術開発/革新的AIエッジコンピューティング技術の開発/マルチモーダルな超低消費電力エッジシステムに向けたAIコンピューティング技術の研究開発
公開日2019/6/25
報告書年度2018 - 2018
委託先名国立大学法人神戸大学
プロジェクト番号P16007
部署名IoT推進部
和文要約
英文要約Title: Project for Innovative AI Chips and Next-Generation Computing Technology Development / Development of Innovative AI Edge Computing Technologies / Development of AI Computing Technologies for Multimodal Ultra-low Power Edge Systems (FY2018-FY2019) FY2018 Annual Report

Algorithm research(Salty Star)
Research on multimodal algorithms such as deep learning / machine learning / statistical analysis that can operate stably under limited computing resources (CPU, memory, storage, etc.) in edge systems were conducted.
1) Technical investigation of AI learning inference engine
As a result of investigating the ET / IoT comprehensive technology exhibition and the research of open source software, there are still many cases of AI analysis of image recognition processing system, and multimodal AI analysis case of sensor system were not found in the market introduction.
2) Preparation of data set for analysis
Multi-modal data (various sensing data) and quality inspection data (visual inspection data) in the manufacturing process of a specific machine component were obtained from a manufacturer of the industry.

Architecture research(Kobe University)
We verified the feasibility for SoC architecture to realize "micro inference engine".
1) Technical research on AI inference engine
The most actively researched inference engines at this time still use images as input, and there are few inference engines for one-dimensional time-series data.
2) Architecture feasibility verification of "micro inference engine"
We proposed an architecture configuration of "micro inference engine" that operates on edge sensor nodes, specialized to one-dimensional time-series data. This is used to set up a task to estimate the heart rate interval of 1 channel ECG data from 2 channels of PPG data. We compared the performance of the inference engine in the case of learning / inference by three methods: Random Forest (RF), Support Vector Machine (Support Vector Machine: SVM) and Convolutional Neural Network (CNN).
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