Study on Neural Network Processor for Image Recognition
| Project/Area Number |
26730027
|
|---|
| Research Category |
Grant-in-Aid for Young Scientists (B)
|
| Allocation Type | Multi-year Fund |
|---|
| Research Field |
Computer system
|
|---|
| Research Institution | Kyoto University |
|---|
Principal Investigator |
|
|---|
| Co-Investigator(Renkei-kenkyūsha) |
SATO Takashi 京都大学, 大学院情報学研究科, 教授 (20431992)
|
|---|
| Project Period (FY) |
2014-04-01 – 2017-03-31
|
| Project Status |
Completed (Fiscal Year 2016)
|
| Budget Amount *help |
¥3,900,000 (Direct Cost: ¥3,000,000、Indirect Cost: ¥900,000)
Fiscal Year 2016: ¥2,340,000 (Direct Cost: ¥1,800,000、Indirect Cost: ¥540,000)
Fiscal Year 2015: ¥650,000 (Direct Cost: ¥500,000、Indirect Cost: ¥150,000)
Fiscal Year 2014: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000)
|
|---|
| Keywords | 画像認識 / ニューラルネットワーク / ディープラーニング / 低消費電力設計 / 近似計算 / メモリスタ / 低消費電力技術 |
|---|
| Outline of Final Research Achievements |
This work proposes efficient computing methods for image recognition by neural network processors, in which inaccurate computation is positively utilized based on the observation of the calculation redundancy in the recognition process. For the inaccurate computation, approximate computing and low-voltage operation of the circuits can be used to improve their energy efficiency. However, they may affect the recognition performance because of the calculation errors. In this work, novel algorithms to mitigate the calculation errors are proposed, and they achieved the energy efficiency improvement of the neural network processor while keeping the recognition accuracy.
|
Report
(4 results)
Research Products
(9 results)
