Poster: Hybrid Encoding in SNN-Based Transformers for Accurate Time Series Forecasting

This study proposes a hybrid encoding method for accurate time-series forecasting in a SNN-based Transformer architecture. It combines a convolutional encoding and a delta modulation encoding, enabling to capture both the spatial and temporal characteristics of time series data effectively. This study additionally introduces RLeaky neurons at the final output layer of the SNN model's pipeline to strengthen the model's ability to analyze time-varying data. An experimental study, using the Metr-la dataset, shows that the proposed methods can improve accuracy performance compared to the state-of-the-arts encoding methods and enumerates the impact of RLeaky neurons.