HSCR: Hierarchical Self-Contrastive Rewarding for Aligning Medical Vision Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for medical vision-language models overlook modality misalignment . HSCR generates high-quality preference data with higher sampling probability . |
| Approach: | They propose a hierarchical self-contrastive reward approach that addresses two challenges in alignment . they leverage the inherent capability of Med-VLMs to generate dispreferred responses . |
| Outcome: | The proposed approach improves accuracy and trustworthiness of medical vision-label models with 2,000 training entries. |
Similar Papers
ProMedical: Hierarchical Fine-Grained Criteria Modeling for Medical LLM Alignment via Explicit Injection (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are difficult to align with high-stakes medical standards due to dissonance between coarse-grained preference signals and complex protocols. |
| Approach: | They propose a framework that aligns Large Language Models with medical standards . they use a dataset generated via a human-in-the-loop pipeline to augment medical instructions . |
| Outcome: | The proposed framework disentangles safety constraints from general proficiency, enabling precise guidance during reinforcement learning. |
Beyond Surface Features: Advancing Medical Vision-Language Alignment via Dynamic Evidence-Guided Preference Optimization (2026.acl-long)
Copied to clipboard
Zixuan Huang, Zhihong Zhu, Xiaolong Liu, Yanchao Hao, Manman Zhang, Zheng Wei, Bowen Xing, Xian Wu, Ye Li, Fen Miao, Yefeng Zheng
| Challenge: | Existing preference-based methods for medical large vision-Language Models face limitations in medical settings . existing methods are limited by overfitting to superficial cues and pseudo convergence of the preference signal. |
| Approach: | They propose a framework that enables evidence-aware and adaptive preference learning for Med-LVLMs. |
| Outcome: | The proposed framework improves evidence-aware and adaptive preference learning for Med-LVLMs. |
Re-Align: Aligning Vision Language Models via Retrieval-Augmented Direct Preference Optimization (2025.emnlp-main)
Copied to clipboard
Shuo Xing, Peiran Li, Yuping Wang, Ruizheng Bai, Yueqi Wang, Chan-Wei Hu, Chengxuan Qian, Huaxiu Yao, Zhengzhong Tu
| Challenge: | emergence of large Vision Language Models (VLMs) has broadened the capabilities of single-modal Large Language Model (LLM) but VLMs are prone to significant hallucinations, especially in the form of cross-modal inconsistencies. |
| Approach: | They propose a new alignment framework that leverages image retrieval to integrate both textual and visual preference signals. |
| Outcome: | The proposed framework mitigates hallucinations more effectively than previous methods . it maintains robustness and scalability across a wide range of VLM sizes and architectures . |
Token-level Inference-Time Alignment for Vision-Language Models (2026.findings-acl)
Copied to clipboard
Kejia Chen, Junjun Zheng, Jiawen Zhang, Manxi Lin, Xiao Pan, Jiacong Hu, Jian Lou, Zunlei Feng, Mingli Song
| Challenge: | Vision-Language Models (VLMs) often prioritize linguistic fluency over visual fidelity . despite widespread adoption, VLMs often exhibit a critical failure mode: hallucination . |
| Approach: | They propose a framework for Token-level Inference-Time Alignment that steers the decoding process without updating the base model parameters. |
| Outcome: | The proposed framework improves performance on 13 benchmarks across architectures . it boosts LLaVA-1.5-7B by 8.6% on MMVet and achieves a 74.0 MMStar score . |
Improving Alignment in LVLMs with Debiased Self-Judgment (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for aligning LVLMs rely on external datasets, human annotations or complex post-processing. |
| Approach: | They propose a method that generates a debiased self-judgment score for LVLMs . this self-evaluation metric is created internally by the model without external resources . |
| Outcome: | The proposed approach outperforms existing methods in reducing hallucinations and safety concerns. |
AlignDistil: Token-Level Language Model Alignment as Adaptive Policy Distillation (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for LLM alignment optimize tokens using a sparse, response-level reward or preference annotation. |
| Approach: | They propose an RLHF-equivalent distillation method for token-level reward optimization that incorporates the reward learned by DPO into the RLHG objective and builds a token-based teacher distribution. |
| Outcome: | The proposed method bridges the accuracy gap between the reward from the DPO model and the pure reward model by building a contrastive DPO reward with a normal and a reverse DPO. |
Learning What Matters: Dynamic Dimension Selection and Aggregation for Interpretable Vision-Language Reward Modeling (2026.acl-long)
Copied to clipboard
Qiyuan Chen, Hongsen Huang, Jiahe Chen, Qian Shao, Jintai Chen, Hongxia Xu, Renjie Hua, Ren Chuan, Jian Wu
| Challenge: | Existing multimodal reward models are interpretable but slow, while discriminative ones are opaque "black boxes." |
| Approach: | They propose a framework that dynamically decomposes evaluation into granular, interpretable dimensions. |
| Outcome: | The proposed framework outperforms open-source reward models on benchmarks like VL-RewardBench. |
Symmetrical Visual Contrastive Optimization: Aligning Vision-Language Models with Minimal Contrastive Images (2025.acl-long)
Copied to clipboard
| Challenge: | Recent studies have shown that Large Vision-Language Models (VLMs) tend to neglect image content and over-rely on language-model priors, resulting in errors in visually grounded tasks and hallucinations. |
| Approach: | They propose a novel finetuning objective that steers the model toward capturing important visual details and aligning them with corresponding text tokens. |
| Outcome: | The proposed method achieves up to 22% reduction in hallucinations and significant gains in vision-centric and general tasks while maintaining or improving the model's general abilities. |
Aligning with Your Own Voice: Self-Corrected Preference Learning for Hallucination Mitigation in LVLMs (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing preference learning-based approaches rely on proprietary models to construct preference datasets, causing a distributional mismatch between the proprietary and target models. |
| Approach: | They propose a framework that aligns LVLMs using in-distribution data derived from the model's intrinsic knowledge. |
| Outcome: | The proposed framework surpasses baselines in hallucination mitigation while requiring only 5.2k samples. |
Focus on What Matters: Enhancing Medical Vision-Language Models with Automatic Attention Alignment Tuning (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods rely on inference-time interventions, which are limited in attention adaptation or require additional supervision. |
| Approach: | They propose a framework for automatic attention alignment tuning that leverages weak labels from SAM and selectively modifies visually-critical attention heads to improve alignment while minimizing interference. |
| Outcome: | The proposed framework outperforms state-of-the-art models on medical VQA and report generation benchmarks. |