Kang, Yechan Seo, Mingyeong Jo, Jeonghee Jang, Hyun Jae Park, Jongkil Kim, Jae wook Lee, Su youn Kim, Jinkyu Park, Seongsik 2025-02-21T01:00:23Z 2025-02-21T01:00:23Z 2025-02-11 2024-12-14 https://pubs.kist.re.kr/handle/201004/151778 Deep spiking neural networks (SNNs), gaining attention as the next generation of artificial neural networks, have been successfully applied to many applications thanks to the development of various algorithms, such as spike encoding. Spike encoding represents input information as discrete spikes, significantly influencing the performance and efficiency of deep SNNs. Most state-of-the-art deep SNN models have greatly improved their performance by using direct encoding. However, performance and efficiency have been limited by the lack of consideration for the brain’s efficiency mechanisms, such as homeostasis. To overcome this limitation, we propose H-Direct, a spike encoding technique designed to balance both effectiveness and efficiency, based on a comprehensive analysis of conventional direct encoding. Furthermore, experimental results confirm that our proposed encoding surpasses traditional direct encoding in both performance and efficiency across multiple image classification benchmarks. English NIPS Homeostasis-aware Direct Spike Encoding for Deep Spiking Neural Networks Conference 1 Conference on Neural Information Processing Systems (NeurIPS 2024), v.00, no.00 Conference on Neural Information Processing Systems (NeurIPS 2024) 00 00 CN Vancouver Convention Center, Canada 2024-12 Workshop on NeuroAI; Fusing Neuroscience and AI for Intelligent Solutions