A new way to find the parameters for analogue neurons

Researchers from the EBRAINS community have demonstrated a new approach to estimate these parameters automatically on BrainScaleS-2. 

The method uses amortised simulation-based inference to learn how different parameter settings affect a neuron's activity. Once trained, it can estimate the parameters that are most likely to produce a given observation, and can be reused for different observations without having to repeat the entire training process. 

The researchers tested the approach on the adaptive exponential integrate-and-fire neuron model implemented on BrainScaleS-2. A neural network trained to identify the relevant features of the recorded membrane activity and estimate the model parameters produced narrow posterior distributions over the parameters. The inferred parameter values reproduced the observed neuron dynamics. 

The results demonstrate that amortised simulation-based inference can help parameterise analogue neuron circuits on BrainScaleS-2. The approach could support more efficient calibration and programming of analogue neuromorphic hardware. 

The work was presented at the Neuro Inspired Computational Elements (NICE) Conference 2026 in Atlanta, USA. The slides and a video of the talk, as well as materials from other sessions, are available on the NICE agenda webpage. 

BrainScaleS-2 is available to researchers through EBRAINS for experiments in computational neuroscience and neuromorphic machine learning. 

The next BrainScaleS hands-on tutorials will be offered on 7 October 2026 at the EBRAINS Users Days in Heidelberg, Germany, and at the virtual training days online on 13 October 2026. 

Read the publication: 

J. Kaiser, E. Müller and J. Schemmel, "Amortized Inference of Neuron Parameters on Analog Neuromorphic Hardware," 2026 Neuro Inspired Computational Elements (NICE), Atlanta, GA, USA, 2026, pp. 1-8, doi: 10.1109/NICE69539.2026.11567515.