Reproduce “The Hidden Perception Constraint in Task-Aware Compression” 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
Sahan Liyanaarachchi, Semih Akkoc, Sennur Ulukus, Aylin Yener
The authors propose a compression method that preserves source-digit class while making reconstructions resemble class-conditional distributions where a receiver classifier works well. The codec is trained without that classifier. In SVHN-to-MNIST experiments, matching toward MNIST reduces classification error from 0.8930 on raw SVHN, while matching to SVHN leaves error around 0.90. This is a digit-task demonstration.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.