Reproduce “Finding the Muses: Identifying Coresets through Loss Trajectories” 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
Manish Nagaraj, Deepak Ravikumar, Efstathia Soufleri, Kaushik Roy
LTC selects a class-balanced coreset by correlating training examples’ epoch-to-epoch loss changes with those of query examples, averaging correlations across queries, and ranking samples. On CIFAR-100, ResNet-18 achieved 35.47% mean test accuracy with 10% of training data, 1.06 percentage points above GraphCut across five seeds. However, the test set was used to select samples, so this accuracy may be optimistic.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.