2022 Fiscal Year Final Research Report
Performance enhancement of photonic reservoir computing based on parallel and deep architecturs
| Project/Area Number |
20K15185
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 30020:Optical engineering and photon science-related
|
|---|
| Research Institution | Saitama University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2020-04-01 – 2023-03-31
|
| Keywords | リザーバコンピューティング / 物理リザーバコンピューティング / 機械学習 / 時間遅延システム / レーザ / 記憶容量 / 非線形性 |
|---|
| Outline of Final Research Achievements |
Energy efficiency issues have been pointed out in machine learning in recent years. Photonic reservoir computing using a laser and a time-delayed loop has attracted attention as a fast and efficient machine learning scheme. This research aims to improve the information processing capability of photonic reservoir computing to make it adaptable to advanced and versatile information processing.
As a result of this study, we showed that a reservoir's memory capacity and nonlinearity can be controlled via the feedback delay time of the reservoir. We also show that the parallelization of reservoirs effectively improves information processing performance in photonic reservoir computing, and this knowledge is useful for reservoir computing based on photonic integrated circuits. Furthermore, we found that mutually coupled different reservoirs in memory capacity and nonlinearity can be applied to various tasks that require both memory capacity and nonlinearity.
|
| Free Research Field |
光工学，機械学習
|
| Academic Significance and Societal Importance of the Research Achievements |
光リザーバコンピューティングは，高速かつ高効率な情報処理を実現可能であると期待されている．特に本研究により光リザーバコンピューティングの情報処理精度を向上することで，より高度で多目的な情報処理に適応可能となり得る．特に光を用いた機械学習技術であることから，光信号のまま情報処理を行うことが可能である．例えば歪みを生じた信号を光リザーバコンピューティングに直接入力することで，信号復元の実時間処理が実現できる．またホログラフィックメモリと呼ばれる大容量記憶媒体技術において，複素光信号をリザーバコンピューティングに直接入力することで，直接観測できない複素振幅の直接再生も期待できる．
|
