상세정보
학술대회
Comparison between STDP and Gradient-descent Training Processes for Spiking Neural Networks Using MNIST Digits
- 발행일
- 202210
- 출처
- International Conference on Information and Communication Technology Convergence (ICTC) 2022, pp.1732-1734
- 협약과제
- 22HS4900, 경량 RISC-V 기반 초저전력 인텔리전트 엣지 지능형반도체 기술 개발, 구본태
- 초록
- Spiking neural networks (SNN) operated by the event-driven spiking process can perform rapid inferences with low power consumption, compared to other neural networks. This study presents a comparison between spike-timing-dependent plasticity (STDP)-based unsupervised training and backpropagation-based gradient-descent (BGD) training approaches for the SNNs by the evaluation of a selected MNIST subset.
- KSP 제안 키워드
- Event-driven, Spike timing dependent plasticity, gradient descent, low power consumption, spiking neural networks, unsupervised training