Cog-DRIFT reformulates difficult reasoning problems into easier variants and applies an adaptive curriculum in RLVR to enable learning from previously unsolvable tasks, delivering 3-10% gains over baselines on reasoning benchmarks.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Cog-DRIFT: Exploration on Adaptively Reformulated Instances Enables Learning from Hard Reasoning Problems
Cog-DRIFT reformulates difficult reasoning problems into easier variants and applies an adaptive curriculum in RLVR to enable learning from previously unsolvable tasks, delivering 3-10% gains over baselines on reasoning benchmarks.