[17p-Z23-1~17] 12.2 Characterization and Materials Physics

▼ [17p-Z23-14] Physical Reservoir Device for Supervised Learning by Random Network of Single-Walled Carbon Nanotube/Porphyrin-Polyoxometalate

〇(D)Deep Banerjee¹, Takumi Kotooka¹, Takuji Ogawa², Hakaru Tamukoh¹, Yuki Usami¹, Hirofumi Tanaka¹ (1.KYUTECH, 2.Osaka Univ)

Keywords:Reservoir computing, Single-walled carbon nanotube, Polyoxometalate

Reservoir computing (RC) has emerged an efficient architecture for supervised learning by training only the output weights. We demonstrate such a physical RC device using single-walled carbon nanotube/porphyrin-polyoxometalate random network in this work. By using the time-series tactile input grasping data of Toyota Human Support Robot, we succefully implemented the supervised prediction of one-hot-vector object classification of different hardness. We infer that intrinsic properties of non-linearity, high dimensionality and excho-state property of the material are indeed important for such reservoir learning and hence can be utilised for higher complex cognitive tasks in the near future.