Questions about ann-to-snn conversion

[MayWeHappy]

Dear Authors,

I trained an MLP on fashionMNIST with 2 hidden layers and converted it to SNN using your ann_to_snn function, with database normalization. However, testing shows the SNN generates the same output for all inputs.

I believe I trained the MLP properly but parameter selection may be wrong for the SNN. Adjusting SNN parameters like thresholds/weight decay could resolve this, allowing effective feature extraction and classification.

How can I modify parameters to fix this? Any guidance would be appreciated.

GuYe

[Hananel-Hazan]

Its recommended to run a heuristic search on the parameters to find the best match.