Revisiting Direct Encoding: Learnable Temporal Dynamics for Static Image Spiking Neural Networks

TL;DR

This paper investigates the limitations of static image encoding in spiking neural networks and proposes a learnable temporal encoding method that introduces meaningful temporal variation, improving the network's ability to model static images.

Contribution

It reveals that the performance gap is due to learnability and surrogate gradient issues, not the encoding scheme, and introduces a minimal learnable temporal encoding with adaptive phase shifts.

Findings

Performance gap attributed to learnability and surrogate gradients

Proposed minimal learnable temporal encoding with adaptive phase shifts

Enhanced temporal modeling of static images in SNNs

Abstract

Handling static images that lack inherent temporal dynamics remains a fundamental challenge for spiking neural networks (SNNs). In directly trained SNNs, static inputs are typically repeated across time steps, causing the temporal dimension to collapse into a rate like representation and preventing meaningful temporal modeling. This work revisits the reported performance gap between direct and rate based encodings and shows that it primarily stems from convolutional learnability and surrogate gradient formulations rather than the encoding schemes themselves. To illustrate this mechanism level clarification, we introduce a minimal learnable temporal encoding that adds adaptive phase shifts to induce meaningful temporal variation from static inputs.