Spiking Neural Networks Exhibit Higher Resistance to Model Stealing

Benchmarking four model stealing attacks reveals that Spiking Neural Networks (SNNs) are more difficult to replicate than conventional artificial neural networks. The temporal, spike-based computation mode makes it harder for adversaries to reconstruct decision behavior through black-box queries.