Skip to content

research

Hooks in PyTorch 🪝

To quote myself in a most recently yet-to-be-published paper:

💪 The ability of deep neural networks (DNNs) come from extracting and interpreting features from the data provided.

What we call, deep features, are the abstract, latent representations that are naturally derived from the training data fed into the DNN. They reflect a consistent activation or response of a layer/node within the model hierarchy to an input.

The features from within a pretrained VGG-11 (top) and ResNet-18 (bottom) on layers of different depths visualized.