[15p-B414-1~11] FS.1 Focused Session "AI Electronics"

[15p-B414-9] Future Prediction of Event Data and Frame Images using Spiking Neural Networks

〇Shunsuke Koshino¹, Naoko Misawa¹, Chihiro Matsui¹, Ken Takeuchi¹ (1.Univ. Tokyo)

Keywords:Spiking Neural Network, Event-based Vision Sensor, Computation-in-Memory

Focusing on future prediction as a task that takes advantage of event data and the high speed of CiM, we propose a new neural network using SNN. The proposed future prediction neural network does not use convolution in the spatial direction but uses only convolution in the channel direction. The motion features of each pixel can be extracted without spatial convolution at the encoder side. In addition, because there is no spatial convolution, it is structurally compatible with CiM and can be efficiently implemented in circuits.