exploration-strategy

N-GRPO introduces semantic neighbor mixing in the GRPO framework to enhance mathematical reasoning diversity while preserving semantic consistency, achieving improvements on math benchmarks and out-of-distribution tasks.

ScreenSearch introduces a system for ambiguity-aware desktop exploration, combining structural screen retrieval and deduplication with a PUCT graph-bandit to handle partial observability in GUI agents. It collects over 1M screenshots across 11 applications and demonstrates a novelty–ambiguity trade-off in exploration policies.