A Formal Tool for Verification of Probabilistic Spiking Neural Networks Based on Quotient Abstractions

The paper introduces CogSpike, a formal verification tool for probabilistic Spiking Neural Networks (SNNs) that utilizes a weight-discretized quotient model abstraction to mitigate state space explosion in probabilistic model checking. This approach allows the mapping of continuous synaptic weights to a compact integer range while preserving synaptic contributions, achieving approximately 17-fold state space reduction per neuron with a discretization parameter of W=3. This enables the verification of larger SNNs, which are typically intractable, making it a significant advancement for practitioners working on the formal verification of probabilistic models in neural network architectures.