Topic overview
Browse by topic, or explore all entries here.
- PapersAgentic Reinforcement Learning with Observation-Calibrated Self-Distillation
Preprint on arXiv (2026) — Paper
OCSD contrasts structurally matched replay views with and without future observations to isolate observation-specific supervision. It uses the resulting residual to modulate token-level GRPO updates at high-uncertainty steps while preserving the trajectory-level update direction.
- PapersLook Ahead Before You Distill: Future Trajectory Validation of Teacher Guidance for Agentic On-Policy Distillation
Preprint on arXiv (2026) — Paper
FutureBridge-OPD (FTB) validates teacher guidance through the future trajectories it induces in the student. At a state with high teacher-student disagreement, it executes a short teacher bridge and compares student continuations from the original and bridged states. The bridge is retained for distillation only when the resulting continuation is more strongly preferred by the teacher.
- PapersBeyond Case-Level Similarity: Multi-Defendant Legal Case Retrieval via Individual-Aware Reasoning
Published in EMNLP main conference (2026) — Paper
MultiLCR studies legal case retrieval beyond case-level textual similarity by modeling defendants' roles and judgment outcomes. It introduces the Multi-ELAM and Multi-LeCaRDv2 datasets with defendant-level annotations and a role- and judgment-aware retrieval framework.
- PapersSkill or Skip? Learning Selective Skill Invocation in Agentic Tasks via Dual-Granularity Preference Learning
Published in Findings of EMNLP (2026) — Paper
SelSkill learns when an agent should invoke a reusable skill and when it should skip it. It combines episode-level and step-level preference learning with controlled invoke-or-skip comparisons to improve selective skill use in agentic tasks.
- PapersLawShift: Benchmarking Legal Judgment Prediction Under Statute Shifts
Published in NeurIPS Datasets and Benchmarks Track (2025) — Paper
LawShift benchmarks legal judgment prediction under statutory revisions. Covering 31 fine-grained change types, it uses metamorphic testing to examine whether models adapt their judgments to changes in the law.
- PapersLAWRE: Evaluating Legal Case Retrieval for Practical Legal Reasoning
Published in IJCNN (2026) — Paper
LAWRE evaluates whether legal case retrieval systems possess the reasoning skills needed in practice, beyond high recall on existing benchmarks. It provides diagnostic tests for capabilities such as distinguishing easily confused cases and identifying legal elements.
- PapersImproving Legal Judgment Prediction via Quantitative Reasoning
Published in ACM Transactions on Knowledge Discovery from Data (TKDD) (2026) — Paper
QR-LJP incorporates legal quantitative reasoning into judgment prediction. It extracts monetary values from case facts, determines the total crime amount using legal reasoning, and uses this measure of crime severity to improve judgment predictions, particularly prison terms.