2017 Fiscal Year Annual Research Report
Self-learnable Analog-Digital-Mixed VLSI Processors for Smart Human-Computer-Interaction
| Project/Area Number |
26870227
|
|---|
| Research Institution | Nara Institute of Science and Technology |
|---|
Principal Investigator |
ZHANG Renyuan 奈良先端科学技術大学院大学, 情報科学研究科, 助教 (00709131)
|
|---|
| Project Period (FY) |
2014-04-01 – 2018-03-31
|
| Keywords | Approximate computing / Analog Calculation Unit / Image processing / Support Vector Machine |
|---|
| Outline of Annual Research Achievements |
In the last year of this project, we extend our research from Multi-valued logic to real analog calculation, which is more efficient and powerful. One of advanced machine learning algorithms ``support vector regression (SVR)” is realized by VLSI circuits to retrieve arbitrary functions.
The analog calculation circuits have been developed for high speed computations with vectors in a very small size. The proposed processor for vector computation is named as ``ACU” (Analog Calculation Unit). The current prototype of ACU accepts nine operands and retrieves arbitrary complicated functions with those nine operands, theoretically. ACU is especially suggested for the use of image processing tasks inside the human-computer-interfaces. We use those analog calculators in several image filters with the size of 3X3; achieve real-time processing with an acceptable inaccuracy of 10%. Our paper from this work won the Outstanding Award on CANDAR’17. The newest progresses are submitted to the high ranking journals IEEE Micro, and IEEE TCAS.
So far, we offered several practical types of approximate computing strategies on the hardware side; and applied them in some computations of image processing. The hardware-effort of approximate computing is one of solutions to efficiently realize HCI with an acceptable loss of computing accuracy inside.
|
