Reproduce “Learning Instance-Specific Augmentations by Capturing Local Invariances” 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

Ning Miao, Tom Rainforth, Emile Mathieu, Yann Dubois, Yee Whye Teh, Adam Foster, Hyunjik Kim

InstaAug learns an input-dependent distribution of image transformations, rather than applying one shared augmentation policy. An invariance module predicts transformations for each image, balancing task loss with transformation diversity. On Tiny-ImageNet cropping, it reaches 66.02±0.18% top-1 accuracy versus 63.20±0.12% for a no-input ablation. This supports input dependence in that experiment, not universal label preservation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.