Reproduce “The Lack of Convexity of the Relevance-Compression Function” 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
Albert E Parker, Tomas Gedeon, Alexander Dimitrov
The authors analyze how stochastic quantizers compress responses into classes while preserving information about a stimulus. They compare a target-constrained information-distortion problem with its penalty-weighted Lagrangian and show that differentiable relevance-compression curves can change curvature. In a synthetic four-Gaussian example, a subcritical branch contains constrained solutions that are not Lagrangian optima; this finding is specific to that example.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.