All of Davis Brown's Comments + Replies

Thanks for this write-up! In case it’s of interest, we have also performed some exploratory interpretability work using the SVD of model weights.  

We examine convolutional layers in models on a couple common vision tasks (CIFAR-10, ImageNet). In short, we similarly take the SVD of the weights in a CNN layer, , and project the hidden layer activations  onto the th singular vector . These singular direction “neurons” can then be studied with interpretability methods: we use hypergraphs, feature visualizati... (read more)