🤖 AI Summary
This work addresses the challenges of parallel training and reliance on surrogate gradients in spiking neural networks by proposing a novel resonator-and-fire neural architecture grounded in the state space model framework. The approach unifies the processing of both real-valued and spiking inputs and, for the first time, integrates state space models with resonator-and-fire neurons, enabling efficient parallel training without surrogate gradients while preserving biological plausibility. By incorporating short-time Fourier transforms, recurrent memory, and attention mechanisms, the architecture supports both parallel and sequential execution and establishes a clear connection to hyperdimensional computing, thereby offering a new paradigm for efficient learning in spiking neural networks.
📝 Abstract
State-space models (SSMs) provide a powerful theoretical framework to enable parallel training of recurrent networks. We expand on previous work adapting SSMs to spiking models to provide a novel interpretation of resonate-and-fire (R\&F) neural networks which is compatible both with real and spiking inputs, parallel and recurrent execution, has clear connections to hyperdimensional (HD) computing, and maintains biologically-realistic features. We demonstrate an implementation of this approach which integrates an STFT, recurrent memory, and attentional features within a single spike-compatible network.