Research and practice
Methods from my research, alongside projects and technical explorations.
OCSD
Distilling signals from future observations to improve reinforcement learning for agents.
FutureBridge-OPD
Validating teacher guidance through the student's future trajectory before distillation.
MultiLCR
Retrieving legal cases through individual defendants' roles and judgment outcomes.
SelSkill
Teaching agents when to invoke a reusable skill—and when to skip it.
LawShift
Testing whether judgment prediction models adapt when statutes change.
LAWRE
Looking beyond recall to diagnose practical legal reasoning in case retrieval.
QR-LJP
Using monetary reasoning to better connect case facts with sentencing.