Amortized Inference of Neuron Parameters on Analog Neuromorphic Hardware

Researchers developed an amortized simulation-based inference algorithm to efficiently tune seven key parameters of neuron models on the BrainScaleS-2 analog neuromorphic chip. Their method uses a neural density estimator to approximate the posterior distribution, constraining a vast parameter space to find combinations that produce realistic spiking behavior. A trained summary network yielded more accurate membrane potential dynamics than handcrafted statistics, validating the approach for parameterizing complex, brain-inspired analog circuits.