Reproduction and extension of contrastive decoding (CD) reveal its effects are mainly due to distribution shifts, not true hallucination mitigation, across multiple models and datasets.

Arnav Bendre, Guneesh Gupta, Kavish Grover et al.

PARED employs inverse RL in feature space to extract implicit rewards from demonstrations, improving language model alignment without task-specific annotations.

Michał Wiliński, Liu Leqi, Chirag Nagpal

FlowCTS replaces KL-OPD with trajectory supervision, raising GenEval to 0.93, OCR to 0.92, and PickScore to 23.06.

Kaiyang Ye, Yuan Ge, Junxiang Zhang et al.

Proposed AAMSF framework for extreme volatility warning via multi-source anomaly fusion; achieved test AUC of 0.680.

Jin Qian, Zhangzhi Xiong, Mingrui Li et al.

Proposed MCOF framework addresses sampling and constraint challenges in VAE-based data-driven problems.

Ye Shi

Proposes NOPD, a self-distillation method that improves vision-language models without external supervision by leveraging prediction discrepancies between clean and corrupted inputs.

Shuai Wang, Daoan Zhang, Zhe Tang et al.

MetaEvolve uses reinforcement learning to cultivate self-evolution meta-skills, boosting code task performance by over 10%.

Shujin Wu, Cheng Qian, Xiusi Chen et al.

DomainPilot employs token-level domain loss monitoring and a two-stage optimization to improve language model fine-tuning by +1.8% to +3.8%.

He Zhang

Proposes Zero-Flow Two-Sample Test (ZF2ST) with strong performance on synthetic and image datasets.

Yakun Wang, Leyang Wang, Song Liu et al.

Exact excluded-pool audits certify missed mass; in the zero-miss regime, the optimal label complexity is Ω(N₀/m).

Martin Anthony, Kaveh Salehzadeh Nobari

Proposes context-weighted discrete flow matching, reducing perplexity by 63% via local context-aware sampling and loss.

Daniil Cherniavskii, Daniel Severo, Karen Ullrich

M2S method predicts posterior mean and maps it to scores, improving generative PPL to 143.3.

Jingyuan Li, Xiaoyi Jiang, Yixuan Jiang et al.

BoE framework optimizes candidate selection under partial verification, improving medical VQA accuracy by 0.4% with limited evidence budget.

Cenwei Zhang, Teng Fang, Yuxia Wang et al.

Transforms active inference into a convex Markov decision process, enabling efficient EFE minimization with theoretical guarantees.

Nikola Milosevic, Nicolás Hinrichs, Nico Scherf

Using XGB-C2ST to reveal dependency deficits in TabbyFlow and TabDiff models.

Jie Zhang

Proposes a new method for evaluating personal LLM agents under temporal interventions; no existing protocol meets all four conditions.

Pin Qian, Su Wang, Yihang Chen et al.

MotifRole-Diff allocates masking by structural risk, raising QM9 validity from 0.905 to 0.944 and lowering FCD to 1.609.

Tasfia Nuzhat Ornee, Elias Hossain, Ivan Garibay et al.

RRPO uses stratified conditional rollouts and contrastive learning to optimize policies without explicit verification signals.

Yuxin Xiong, Xunyi Jiang, Rohan Surana et al.

This paper establishes label complexity bounds for class-conditional coverage under distribution shift, showing pseudo labels cannot fully recover class thresholds without target labels.

Weijia Han, Lisha Qu

Harness-based LLM-assisted GPU kernel system achieves 1.62-29.68× speedups, integrating human skills and profile-backed search.

Yue Shui, Chenyu Ma, Hangfei Xu et al.