Reproduce “Rethinking GA2Ms: performance and out-of-distribution generalisation” 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

George Dimitriadis, Damian Machlanski, Paul Clarke, Spyridon Samothrakis

CoGA2M combines smooth, multi-resolution spline effects for individual features with pairwise interactions in an interpretable tabular predictor. The authors evaluate it on standard benchmarks and constructed diversity shifts, where test inputs occupy regions absent from training. On 22 selected numerical-feature regression datasets, its average ID-to-OOD performance loss was smaller than LightGBM’s; this result does not establish universal OOD superiority or a causal mechanism. The bootstrap mean difference was 0.247 (p = 0.006).

Summary by alphaXiv.

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