Modality-Fair Preference Optimization for Trustworthy MLLM Alignment

• URL: http://arxiv.org/abs/2410.15334v1
• Date: Sun, 20 Oct 2024 08:56:52 GMT
• Title: Modality-Fair Preference Optimization for Trustworthy MLLM Alignment
• Authors: Songtao Jiang, Yan Zhang, Ruizhe Chen, Yeying Jin, Zuozhu Liu,
• Abstract summary: Direct Preference Optimization (DPO) is effective for aligning large language models (LLMs) It often favors text over image information, leading to unreliable outputs and visual hallucinations. We propose Modality-Fair Preference Optimization (MFPO) to balance text and image preferences.
• Score: 11.796170286878056
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Direct Preference Optimization (DPO) is effective for aligning large language models (LLMs), but when applied to multimodal models (MLLMs), it often favors text over image information, leading to unreliable outputs and visual hallucinations. To address this, we propose Modality-Fair Preference Optimization (MFPO) to balance text and image preferences. First, we found that the lack of image-related rewards in preference data biases optimization toward text, so we created automated, fine-grained image preference data to correct this. Then, we designed a learning objective to ensure the model captures both text and image preferences while maintaining high-quality outputs. Finally, we use a multi-stage alignment approach to stabilize training and improve learning across both modalities. Extensive experiments demonstrate that MFPO significantly enhances MLLM trustworthiness. On models like LLaVA-v1.5 (7B, 13B), our approach reduces hallucinations substantially. On the 7B model, MFPO outperforms GPT-4V and achieves a nearly 40\% improvement over previous methods on Object HalBench, as well as achieving state-of-the-art performance on both Object HalBench and AMBER when combined with the latest LLaVA-v1.6. Code will be released.

Related papers

• AdaViP: Aligning Multi-modal LLMs via Adaptive Vision-enhanced Preference Optimization [26.03204301595711]
  We propose an Adaptive Vision-enhanced Preference optimization (AdaViP) that addresses limitations through two key innovations. vision-based preference pair construction integrates multiple visual foundation models to strategically remove key visual elements from the image. AdaViP-7B achieves 93.7% and 96.4% reductions in response-level and mentioned-level hallucination respectively on the Object HalBench.
  arXiv  Detail & Related papers  (2025-04-22T06:19:38Z)
• From Captions to Rewards (CAREVL): Leveraging Large Language Model Experts for Enhanced Reward Modeling in Large Vision-Language Models [58.16075709485292]
  CAREVL is a novel method for preference reward modeling by reliably using both high- and low-confidence data. CAREVL achieves performance improvements over traditional distillation-based methods on VL-RewardBench and MLLM-as-a-Judge benchmark.
  arXiv  Detail & Related papers  (2025-03-08T16:13:18Z)
• Re-Align: Aligning Vision Language Models via Retrieval-Augmented Direct Preference Optimization [19.37373012848517]
  Large Vision Language Models (VLMs) are prone to significant hallucinations, particularly in the form of cross-modal inconsistencies. We introduce Re-Align, a novel alignment framework that leverages image retrieval to construct a dual-preference dataset. We also introduce rDPO, an extension of the standard direct preference optimization that incorporates an additional visual preference objective during fine-tuning.
  arXiv  Detail & Related papers  (2025-02-18T18:59:57Z)
• CHiP: Cross-modal Hierarchical Direct Preference Optimization for Multimodal LLMs [107.21334626890713]
  Multimodal Large Language Models (MLLMs) still struggle with hallucinations despite their impressive capabilities. We propose a Cross-modal Hierarchical Direct Preference Optimization (CHiP) to address these limitations. We evaluate CHiP through both quantitative and qualitative analyses, with results across multiple benchmarks demonstrating its effectiveness in reducing hallucinations.
  arXiv  Detail & Related papers  (2025-01-28T02:05:38Z)
• Multimodal Preference Data Synthetic Alignment with Reward Model [23.978820500281213]
  We propose a new framework in generating synthetic data using a reward model as a proxy of human preference for effective multimodal alignment with DPO training. Experiment results indicate that integrating selected synthetic data, such as from generative and rewards models can effectively reduce reliance on human-annotated data.
  arXiv  Detail & Related papers  (2024-12-23T09:29:40Z)
• Scalable Ranked Preference Optimization for Text-to-Image Generation [76.16285931871948]
  We investigate a scalable approach for collecting large-scale and fully synthetic datasets for DPO training. The preferences for paired images are generated using a pre-trained reward function, eliminating the need for involving humans in the annotation process. We introduce RankDPO to enhance DPO-based methods using the ranking feedback.
  arXiv  Detail & Related papers  (2024-10-23T16:42:56Z)
• MIA-DPO: Multi-Image Augmented Direct Preference Optimization For Large Vision-Language Models [85.30735602813093]
  Multi-Image Augmented Direct Preference Optimization (MIA-DPO) is a visual preference alignment approach that effectively handles multi-image inputs. MIA-DPO mitigates the scarcity of diverse multi-image training data by extending single-image data with unrelated images arranged in grid collages or pic-in-pic formats.
  arXiv  Detail & Related papers  (2024-10-23T07:56:48Z)
• CLIP-DPO: Vision-Language Models as a Source of Preference for Fixing Hallucinations in LVLMs [37.98496239547762]
  Large Vision Language Models are prone to hallucinating details like objects and their properties or relations, limiting their real-world deployment. We present CLIP-DPO, a preference optimization method that leverages contrastively pre-trained Vision-Language (VL) embedding models, such as CLIP, for DPO-based optimization of LVLMs.
  arXiv  Detail & Related papers  (2024-08-19T21:56:20Z)
• mDPO: Conditional Preference Optimization for Multimodal Large Language Models [52.607764280030196]
  Direct preference optimization (DPO) has shown to be an effective method for large language model (LLM) alignment. Recent works have attempted to apply DPO to multimodal scenarios but have found it challenging to achieve consistent improvement. We propose mDPO, a multimodal DPO objective that prevents the over-prioritization of language-only preferences by also optimizing image preference.
  arXiv  Detail & Related papers  (2024-06-17T17:59:58Z)
• Aligning Large Language Models via Fine-grained Supervision [20.35000061196631]
  Pre-trained large-scale language models (LLMs) excel at producing coherent articles, yet their outputs may be untruthful, toxic, or fail to align with user expectations. Current approaches focus on using reinforcement learning with human feedback to improve model alignment. We propose a method to enhance LLM alignment through fine-grained token-level supervision.
  arXiv  Detail & Related papers  (2024-06-04T20:21:45Z)
• Multi-Reference Preference Optimization for Large Language Models [56.84730239046117]
  We introduce a novel closed-form formulation for direct preference optimization using multiple reference models. The resulting algorithm, Multi-Reference Preference Optimization (MRPO), leverages broader prior knowledge from diverse reference models. Our experiments demonstrate that LLMs finetuned with MRPO generalize better in various preference data, regardless of data scarcity or abundance.
  arXiv  Detail & Related papers  (2024-05-26T00:29:04Z)
• Aligning Modalities in Vision Large Language Models via Preference Fine-tuning [67.62925151837675]
  In this work, we frame the hallucination problem as an alignment issue, tackle it with preference tuning. Specifically, we propose POVID to generate feedback data with AI models. We use ground-truth instructions as the preferred response and a two-stage approach to generate dispreferred data. In experiments across broad benchmarks, we show that we can not only reduce hallucinations, but improve model performance across standard benchmarks, outperforming prior approaches.
  arXiv  Detail & Related papers  (2024-02-18T00:56:16Z)
• Aligning Large Multimodal Models with Factually Augmented RLHF [176.54751941088819]
  Large Multimodal Models (LMM) are built across modalities and misalignment between two modalities can result in "hallucination" We adapt the Reinforcement Learning from Human Feedback (RLHF) from the text domain to the task of vision-language alignment. We propose a new alignment algorithm called Factually Augmented RLHF that augments the reward model with additional factual information. Our approach achieves remarkable improvement on the LLaVA-Bench dataset with the 94% performance level of the text-only GPT-4.
  arXiv  Detail & Related papers  (2023-09-25T20:59:33Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.