Dynamically Adaptive Approximate Computing Framework

In general, reducing the computational accuracy of a program reduces the computational load on the processor, resulting in higher speed and power savings. In many past studies, computational accuracy has been determined by the programmer at the time of writing the program. However, it is difficult for programmers to properly and safely determine the necessary and sufficient computational accuracy at the time of program writing, since it varies greatly with the moment-to-moment input values and the user state at the time of execution. Therefore, we develop a computer infrastructure that automatically changes the computational accuracy based on feedback from the system and the user at runtime. In our proposed “Dynamic Approximation Processor,” we introduce a unique instruction class to realize execution of the same program by different approximation at runtime. In parallel, we study methods to estimate user satisfaction and provide feedback through the HCI approach.

  • 冨田 和孝, 中村 朋生, 小泉 透, 入江 英嗣, 坂井 修一: 「近似レベルを動的制御可能なアーキテクチャのためのコンパイラフレームワークの検討」, 第19回情報科学技術フォーラム講演論文集, CC-002, pp. 33–38, Sep., 2020. (FIT船井ベストペーパー賞)
  • 河野 翔太, 中村 朋生, 門本 淳一郎, 入江 英嗣, 坂井 修一: 「ユーザ視線検出を用いた効率的な動的映像デコード手法」, マルチメディア,分散,協調とモバイルシンポジウム, pp. 293–303, Jul., 2021.
  • Yuya Degawa, Toru Koizumi, Tomoki Nakamura, Ryota Shioya, Junichiro Kadomoto, Hidetsugu Irie, Shuichi Sakai: “Accurate and Fast Performance Modeling of Processors with Decoupled Front-end”, Int. Conf. on Computer Design, pp. 88–92, Oct., 2021.
  • Tomoki Nakamura, Kazutaka Tomida, Shota Kohno, Hidetsugu Irie, Shuichi Sakai: “Stochastic Iterative Approximation: Software/hardware techniques for adjusting aggressiveness of approximation”, Int. Conf. on Computer Design, pp. 74–82, Oct., 2021.
  • 冨田 和孝, 中村 朋生, 小泉 透, 出川 祐也, 入江 英嗣, 坂井 修一: 「近似の積極性を動的制御可能なアーキテクチャのための コンパイラフレームワーク」, 情報処理学会論文誌, Vol. 63, No. 4, pp. 1019–1028, Apr., 2022.

Researches

Search this site

Recent News

SAKAI・IRIE LAB