Dynamically Adaptive Approximate Computing Framework
We are researching an adaptive/learning approximate computing system in which the computer automatically adjusts the calculation accuracy on run-time to achieve both task completion and calculation amount reduction. Approximate computing technology has a great effect if it is done properly, but it is often difficult to predict the appropriate degree of approximation statically, which increases the cost of software development. In this research, we are developing an on-demand software/hardware approximate computing framework that feeds back user satisfaction and finely adjusts the approximation level at run-time. In this system, even with the same program, processing with different degrees of approximation is performed according to the user state at that time.
- 道上 和馬, 中村 朋生, 小泉 透, 入江 英嗣, 坂井 修一: 「近似度合いを動的制御可能なアーキテクチャの提案」, 情報処理学会研究報告システム・アーキテクチャ, Vol. 2020-ARC-240, No. 31, pp. 1–9, Feb., 2020.