Reproduce “Taxonomy and Evaluation of Structured Compression of Convolutional Neural Networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Andrey Kuzmin, Markus Nagel, Saurabh Pitre, Sandeep Pendyam, Tijmen Blankevoort, Max Welling
The authors compare structured CNN compression methods across data and training budgets. On ImageNet, CP decomposition usually leads without fine-tuning, while spatial SVD leads among tested SVD and tensor methods after fine-tuning. For ResNet18, combining spatial SVD with probabilistic channel pruning gives the strongest combination result. Against scratch training, compression favors channel pruning, but spatial SVD only at lower compression.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.