Gradient representations in ReLU networks as similarity functions
This work addresses improving decision-making in neural networks for machine learning practitioners, but it appears incremental as it builds on existing interpretations of feed-forward networks.
The paper tackled the problem of refining decisions in ReLU networks by exploiting the tangent space, showing that a Riemannian metric parametrized on network parameters forms a similarity function at least as good as the original network and suggesting a sparse metric to increase the similarity gap.
Feed-forward networks can be interpreted as mappings with linear decision surfaces at the level of the last layer. We investigate how the tangent space of the network can be exploited to refine the decision in case of ReLU (Rectified Linear Unit) activations. We show that a simple Riemannian metric parametrized on the parameters of the network forms a similarity function at least as good as the original network and we suggest a sparse metric to increase the similarity gap.