LOss-Based SensiTivity rEgulaRization: towards deep sparse neural networks
This work addresses the need for efficient neural network compression, which is incremental as it builds on sensitivity-based pruning methods.
The paper tackles the problem of training sparse neural networks by introducing LOBSTER, a method that prunes parameters based on their sensitivity to loss variation, achieving competitive compression ratios across various architectures and datasets with minimal computational overhead.
LOBSTER (LOss-Based SensiTivity rEgulaRization) is a method for training neural networks having a sparse topology. Let the sensitivity of a network parameter be the variation of the loss function with respect to the variation of the parameter. Parameters with low sensitivity, i.e. having little impact on the loss when perturbed, are shrunk and then pruned to sparsify the network. Our method allows to train a network from scratch, i.e. without preliminary learning or rewinding. Experiments on multiple architectures and datasets show competitive compression ratios with minimal computational overhead.