Perception-Enhanced Alignment DPO (PEA-DPO), a framework for multimodal LLMs alignment, which explicitly leverages visual preference signals to overcome visual insensitivity is proposed, which demonstrates that PEA-DPO enhances sensitivity to visual context while preserving the language modeling capacity of the base model.
Abstract
Direct Preference Optimization (DPO) has emerged as an effective approach for aligning large language models (LLMs) with human preferences. However, its adaptation to multimodal settings remains unexplored. Through representational analysis, we identify a key limitation in multimodal preference optimization, which we term visual insensitivity: models often fail to distinguish between images and those with critical visual context removed. Our theoretical analysis further uncovers two manifestations of this problem, namely Across-Image Insensitivity and Within-Image Insensitivity. To address these challenges, we propose Perception-Enhanced Alignment DPO (PEA-DPO), a framework for multimodal LLMs alignment, which explicitly leverages visual preference signals to overcome visual insensitivity. We further provide a theoretical analysis demonstrating that PEA-DPO provably mitigates both failure modes. Empirical results demonstrate that PEA-DPO enhances sensitivity to visual context while preserving the language modeling capacity of the base model. Evaluations across three hallucination benchmarks using MLLMs of varying scales show that PEA-DPO effectively mitigates visual insensitivity, achieves stronger multimodal alignment, and substantially reduces hallucinations.
Despite the rapid progress of Multimodal Large Language Models (MLLMs), they still suffer from untruthfulness issues, such as visual hallucinations, content fabrication, and unfaithful reasoning, which substantially undermine their faithfulness and practical utility. Alignment methods based on human preference, such as Direct Preference Optimization (DPO), have been widely adopted to address these issues. However, multimodal reasoning errors often propagate across stages, and final-answer errors can often be traced to mistakes in early grounding stages, yet standard DPO typically applies preference optimization at the final-answer level. This credit-assignment challenge means that supervision for early grounding stages is indirect rather than stage-specific, making it difficult to suppress error propagation arising from grounding drift and context inconsistency. To address this, we propose Grounded Context Preference Optimization (Groc-PO), a grounded preference optimization framework for MLLMs. We further construct the Grounded Context Preference Dataset (GCPD), organizing multi-stage preference samples around three stages of Object Grounding, Contextual Grounding, and Grounded Reasoning, to capture the formation, integration, and utilization of grounded context. By introducing more explicit preference supervision over multiple grounded stages, Groc-PO strengthens context-dependent reasoning and mitigates cross-stage error propagation. Extensive experiments show that, compared with standard DPO and other strong baselines, Groc-PO achieves improved performance in hallucination mitigation, faithful reasoning, and overall reliability, supporting the value of more explicit grounded supervision for trustworthy multimodal reasoning.
Zhi Zheng, Zheren Fu, Zhiyuan Yao et al.· 1 citation
Context-Calibrated DPO (C$^2$-DPO), which directly maximizes CPG while preserving the original preference ordering, is proposed, which substantially reduces hallucination without compromising general reasoning.
Byungoh Ko, Jinyoung Park, Jongha Kim et al.· 0 citations
Vision-language alignment, the stage that bridges pretrained vision encoders and large language models, is widely treated as a form of pretraining requiring full-parameter updates. We challenge this view and investigate what happens when low-rank adaptation is applied to the LLM during this stage instead. We find that low-rank alignment not only reduces computational costs but also outperforms full-parameter alignment on most benchmarks. To understand this phenomenon, we systematically characterize the implicit biases introduced by low-rank adaptation during alignment. Empirically, we find that low-rank alignment shifts model behavior from hallucinatory to conservative and preserves per-token linear separability of visual features that full-parameter alignment disrupts, a phenomenon we term LS-curse. Geometrically, low rank aligned models exhibit more homogeneous and structurally stable visual representations, maintaining modality-specific knowledge rather than prematurely fusing entity-level semantics. Theoretically, we establish two theorems showing that low-rank alignment induces preferences for parameter subspaces with flat gradients and feature subspaces robust to perturbations, providing a principled explanation for the observed structure-preserving behavior. Extensive experiments cover ablation over 100 alignment configurations, three families of low-rank operators, and various rank, encoder, and other settings.
Mingjia Shi, Shuo Wang, Xiaobo Wang et al.· 0 citations
This empirical research establishes the essential groundwork for predictably scaling multimodal foundation models by modeling the influence of data composition on compute laws and allocation exponents and derive an efficiency frontier specifying precise configurations of model size, token count, and data mixture.
Despite the progress of multimodal large language models (MLLMs), they continue to exhibit deficiencies in visual perception. Following visual instruction tuning, internal MLLM representations rapidly deviate from their original semantic states during inference, causing severe information degradation. While existing methods attempt to leverage external vision foundation models (VFMs) to align internal representations, we find that direct alignment with VFMs enhances visual semantics but fails to mitigate representation deviation. To address this, we propose Spatial-Spectral Visual Anchor Learning (SSVAL). The core of SSVAL is Visual Anchor Prompt Injection (VAPI), which introduces prompts that absorb rich knowledge from external VFMs during training, enabling them to serve as stable visual anchors that mitigate representation deviation during inference. Additionally, we incorporate auxiliary spatial and frequency-domain representation alignment losses to provide complementary vision-specific supervision at intermediate LLM layers. Extensive experiments demonstrate that SSVAL significantly outperforms existing methods. Code are available on our \href{https://msls38.github.io/SSVAL/}{project page}.
Qianlong Yang, Bowen Ye, Xianda Guo et al.· 0 citations
Direct preference optimization (DPO) is now a standard method for aligning large language models (LLMs) using human preference data. Each DPO example contains a prompt and a pair of candidate model responses. While prompts and responses are often public or model-generated, the relative preference between responses reflects subjective judgments and can reveal sensitive attributes of annotators or end users. Off-the-shelf privacy-preserving approaches are not well matched to this structure, leading to unnecessary noise injection and biased updates in training. In this paper, we formalize preference privacy, a label-DP-style privacy notion for DPO that protects only the relative preference between candidate responses, assuming an adversary who already knows the prompt and responses. We then design PrivDPO, a DPO variant that enforces preference privacy while remaining compatible with large-scale LLM training. Our main observation is that, for neighboring examples differing only in their preference signal, the gradient difference lies on a one-dimensional preference axis determined solely by the text; all preference information flows through this axis. PrivDPO adds calibrated randomness only along this axis via an unbiased randomized rescaling of the DPO objective, avoiding per-example gradient operations. Our experiments on three alignment benchmarks and three LLM families show that PrivDPO consistently achieves strong privacy-utility trade-offs compared with privacy-preserving baselines.
Yangfan Jiang, Fei Wei, Ergute Bao et al.· 0 citations