[17a-A33-1~11] FS.1 Focused Session "AI Electronics"

9:30 AM - 9:45 AM JST(12:30 AM - 12:45 AM UTC)Presentation by Applicant for JSAP Young Scientists Presentation Award

[17a-A33-3] Realtime prediction based on all-photonic reservoir circuit integrated with a photonic readout layer

〇(M2)Keigo Takabayashi¹, Takeo Maruyama¹, Tomoaki Niiyama¹, Satoshi Sunada¹ (1.Kanazawa Univ.)

Keywords:

photonic computing,reservoir computing,online learning

The purpose of this study is to realize an all-photonic reservoir circuit that can process all information from input to output light on a silicon photonic circuit, thereby enabling low-power and high-speed inference. In this reservoir circuit, readout weights are adjusted using phase shifters, etc. However, applying existing learning algorithms requires many measurements and calibrations because each phase shift amount needs to be measured in advance. Therefore, in this study, we apply an evolutionary strategy and simultaneous perturbation stochastic approximation (SPSA) as a learning algorithm that can adjust readout weights without requiring prior knowledge of photonic circuits or many measurements, and show that it can predict chaotic time series in real time.