시계열 데이터를 위한 스파이크 인코딩 최신 연구 동향

A Spiking Neural Network (SNN) processes information through discrete electrical events, emulating the behavior of neurons observed in the brain, known as spikes. Spikes are generated when the membrane potential exceeds a specific threshold, and these generated spikes are used to communicate information between nodes within SNNs. This event-driven method of information transmission is energy-efficient because of the sparsity of spike data. SNN models can provide the advantage of accomplishing tasks with reduced computational resources, while enabling comparable performance to Deep Neural Networks (DNNs) in processing temporal data by leveraging spike timing. This research investigates methods for efficiently handling time-related information, with a focus on applying recent trends in spike encoding techniques to analog time-series data. We categorize four encoding methods - HSA, BSA, Burst, and TTFS - into deconvolution-based encoding and temporal coding methods. We measure their accuracy by utilizing a simple classification model and conduct analysis to identify the most suitable encoding method for time-series data classification task.