I’ve been looking into Captum library to do some parameter/neuron attribution. I.e., given a model and an input, quantify neurons’ impact on model’s prediction. Captum seems to have some tutorials for feature attribution but not any for parameter attribution. Is that the case or am I overlooking it ?
Hi @TinfoilHat0! Thank you for the question! What do you exactly mean by parameter ? Do you mean the weights of the neurons ? Current attribution for neurons is meant for neuron / layer activation.
Hi @TinfoilHat0, you can take a look at this getting started tutorial on the Titanic dataset, it covers example usage of Captum for layer and neuron attribution.