A range of methods aim to enhance the performance of vision-language models (VLMs) at test time. Among them, transduction has emerged as a promising paradigm due to its strong compatibility and efficiency. However, realistic evaluations often involve highly imbalanced class distributions, which cause performance degradation or even collapse. In this work, we systematically revisit transduction from the perspective of penalized likelihood estimation (PLE), showing that PLE with a KL-divergence anchor term naturally yields an adaptive shrinkage behavior between prior anchors and empirical estimates. From this viewpoint, the brittleness of transductive methods can be attributed to the absence of anchoring mechanism and static modeling of the shrinkage strength. Therefore, we propose Mixture of Von Mises-Fisher Models with Dynamic Shrinkage (MOON). MOON is built upon a mixture of von Mises-Fisher distributions to model feature representations on the unit hypersphere. To handle imbalance, MOON dynamically adjusts the shrinkage strength using zero-shot priors at both instance and class levels. Thus, it suppresses unreliable assignments and prevents harmful updates from outlier classes, thereby mitigating negative transfer. MOON is model-agnostic, training-free, and requires no task-specific hyperparameter tuning. Extensive experiments further validate the advantage of MOON in both performance and efficiency. Our code is available at https://github.com/walawalagoose/MOON
Jiazhe Huang, Zhiming Liu, Changhu Wang et al.· 2 citations
Lead optimization in drug discovery requires iteratively refining molecular candidates while preserving structural similarity to the original compound. Since each evaluation is costly, sample efficiency, the ability to achieve strong performance with limited oracle calls, becomes critical. Existing methods, from genetic algorithms to LLM-based approaches, rarely leverage intermediate outcomes and oracle feedback within an optimization trajectory, which reduces sample efficiency and forces more oracle calls to reach high success rates. To address this, we present POLO (Preference-guided Optimization for Lead Optimization), a multi-turn Reinforcement Learning framework for sample-efficient lead optimization. At its core, POLO introduces Preference-Guided Policy Optimization (PGPO), which extracts learning signals from each trajectory at two complementary levels: trajectory-level policy optimization reinforces successful multi-turn strategies, while turn-level preference learning ranks intermediate molecules by oracle-derived rewards and constructs within-trajectory pairwise comparisons to provide denser learning signals for effective modifications. Crucially, these preference signals require no additional oracle calls, as they are derived from evaluations already collected during rollouts. Experiments demonstrate that POLO achieves 85% success rate on single-property tasks (40% improvement over best baselines) and 50% on multi-property tasks using only 500 oracle calls. Our code is available at https://github.com/REAL-Lab-NU/POLO.
Ziqing Wang, Yibo Wen, William Pattie et al.· Proceedings of the 32nd ACM...· 0 citations
Retrieval-augmented generation (RAG) enables LLMs to access external knowledge for answering knowledge-intensive questions. For complex multi-hop questions, multi-turn retrieval-augmented reasoning extends RAG into an iterative process that repeatedly searches for and integrates evidence across documents. However, existing reinforcement-learning (RL) approaches for agentic RAG are typically optimized with final-answer rewards, which provide sparse supervision and overlook whether the model actually retrieves the required evidence chain. We present \textsc{GTA-RAG}, a graph-trajectory-augmented RL framework for multi-turn retrieval-augmented reasoning. From an entity--document graph, we sample connected document paths, synthesize multi-hop QA trajectories, and validate them with the deployed retriever to obtain executable trajectory-level supervision. We then optimize the retrieval policy with Group Relative Policy Optimization (GRPO) and a trajectory-guided reward that encourages both accurate answers and acquisition of target evidence documents, followed by answer-reward training on natural QA instances. Experiments on three multi-hop and two simple QA benchmarks show that \method{} consistently outperforms RL-based RAG baselines with both Qwen2.5-3B and Qwen2.5-7B backbones, while substantially improving evidence-chain coverage. Our code is available at https://github.com/cjcj46262/GTA-RAG.
Jun Chen, Yongchao Liu, Pengyu Qiu et al.· 0 citations
An evidence-guided retrieval-augmented framework EviRAG is proposed that decomposes retrieval into structured and unstructured alignment levels and substantially reduces clinical inconsistencies compared to strong medical vision-language baselines.
Yiyang Gu, Jiayue Fan, Kaili-May Liu et al.· Annual International ACM SIG...· 1 citation
D-SCAN (Document-level Signal Collapse Analysis), a lightweight detection framework that monitors attention dynamics to identify attacked generations and proposes D-SCAN, a lightweight detection framework that can detect attacks even when they fail to alter the final answer.
Yingtao Ren, Ziyi Zhao, Yiwei Fu et al.· Annual International ACM SIG...· 0 citations