Impact of Local Sequential Vision on Length Generalization

Investigating whether local, foveated visual processing improves state tracking and length generalization compared to global, single-shot computation. The study examines if sequential vision models can aggregate information more effectively than standard global vision architectures on simple visual tasks.