ITS 2021
We made a paper and a presentation at ITS World Congress 2021 about a research on object recognition performance shift between before and after Spiking Neural Network (SNN) conversion at Intelligent Transportation System - World Congress (ITS-WC) 2023 in Hamburg.
This work is entitled "Investigation of Performance Shift by Conversion from Convolutional Neural Networks to Spiking Neural Networks". The issue we struggled with in this work is performance decrease by conversion from Convolutional Neural Networks (CNN) to Spiking Neural Networks (SNN). One of the difficulties in SNN handling is its training. As we cannot use Back Propagation (BP) technique in general in SNN training due to its asynchrony, SNN conversion method (ANN2SNN or CNN2SNN) which converts trained CNN to SNN is useful. On the other hand, performance after conversion decreases because signal propagation is different between in training and in inference. Therefore, we investigated the tolerance until how many convolutional layers can keep the performance.