Algorithmic Research Group / Code & Benchmarks

DeltaMLBench

A benchmark for measuring whether AI agents can improve on published ML research, focused on the gap between implementing known approaches and discovering better ones.