neural-research

This paper investigates how restricting evidence visibility in multi-module neural systems composed of language models can improve compositional generalization, showing that restricted-visibility societies outperform globally visible ones by over 20 percentage points in a relay communication setup.