Search papers, labs, and topics across Lattice.
Northeastern University, Microsoft Research
1
0
2
Off-policy degree in RLVR updates can drastically change which tokens drive learning, leading to a new adaptive method that outperforms traditional baselines.
