About this document
Assignment3 Report by Fehad Nazir is a document available to read on EtoBox.
The document discusses a model pruning experiment conducted on a fully connected Multi-Layer Perceptron (MLP) for the MNIST dataset, achieving high training and test accuracies. It highlights the effectiveness of loss curvature-based pruning over random pruning, showing that significant reductions in model size can be made without substantial accuracy loss. The findings emphasize the over-parameterization of neural networks and the importance of identifying critical connections for efficient model compressi
- Author
- Fehad Nazir
- Language
- EN