Leveraging spike-timing information in digital systems

Determine how spike-timing information—the precise timing of discrete spike events used by biological neurons—can be effectively leveraged in digital computing systems to encode and process information efficiently.

Background

The paper motivates event-based encoding by noting that biological neurons communicate via spikes whose timing likely carries rich information. While vector-quantized autoencoders dominate neural audio compression, they do not naturally support event-based representations.

To explore leveraging spike timing in a practical digital context, the authors propose a binary autoencoder for audio (Spiking Music compression) and demonstrate competitive performance on the MAESTRO dataset, including emergent alignment of latent events with piano key strikes. The abstract frames the broader question as unresolved in general digital systems, beyond the specific audio compression approach studied.