Large Vision-Language Models (LVLMs) often suffer from object hallucination, making erroneous judgments about the presence of objects in images. We propose this primarily stems from spurious correlations arising when models strongly associate highly co-occurring objects during training, leading to hallucinated objects influenced by visual context. Current benchmarks mainly focus on hallucination detection but lack a formal characterization and quantitative evaluation of spurious correlations in LVLMs. To address this, we introduce causal analysis into the object recognition scenario of LVLMs, establishing a Structural Causal Model (SCM). Utilizing the language of causality, we formally define spurious correlations arising from co-occurrence bias. To quantify the influence induced by these spurious correlations, we develop Causal-HalBench, a benchmark specifically constructed with counterfactual samples and integrated with comprehensive causal metrics designed to assess model robustness against spurious correlations. Concurrently, we propose an extensible pipeline for the construction of these counterfactual samples, leveraging the capabilities of proprietary LVLMs and Text-to-Image (T2I) models for their generation. Our evaluations on mainstream LVLMs using Causal-HalBench demonstrate these models exhibit susceptibility to spurious correlations, albeit to varying extents.

This study addresses the challenge of noise in training datasets for Direct Preference Optimization (DPO), a method for aligning Large Language Models (LLMs) with human preferences. We categorize noise into pointwise noise, which includes low-quality data points, and pairwise noise, which encompasses erroneous data pair associations that affect preference rankings. Utilizing Distributionally Robust Optimization (DRO), we enhance DPO's resilience to these types of noise. Our theoretical insights reveal that DPO inherently embeds DRO principles, conferring robustness to pointwise noise, with the regularization coefficient $\beta$ playing a critical role in its noise resistance. Extending this framework, we introduce Distributionally Robustifying DPO (Dr. DPO), which integrates pairwise robustness by optimizing against worst-case pairwise scenarios. The novel hyperparameter $\beta'$ in Dr. DPO allows for fine-tuned control over data pair reliability, providing a strategic balance between exploration and exploitation in noisy training environments. Empirical evaluations demonstrate that Dr. DPO substantially improves the quality of generated text and response accuracy in preference datasets, showcasing enhanced performance in both noisy and noise-free settings.

Existing efforts to align multimodal large language models (MLLMs) with human preferences have only achieved progress in narrow areas, such as hallucination reduction, but remain limited in practical applicability and generalizability. To this end, we introduce **MM-RLHF**, a dataset containing **120k** fine-grained, human-annotated preference comparison pairs. This dataset represents a substantial advancement over existing resources, offering superior size, diversity, annotation granularity, and quality. Leveraging this dataset, we propose several key innovations to improve both the quality of reward models and the efficiency of alignment algorithms. Notably, we introduce the **Critique-Based Reward Model**, which generates critiques of model outputs before assigning scores, offering enhanced interpretability and more informative feedback compared to traditional scalar reward mechanisms. Additionally, we propose **Dynamic Reward Scaling**, a method that adjusts the loss weight of each sample according to the reward signal, thereby optimizing the use of high-quality comparison pairs. Our approach is rigorously evaluated across **10** distinct dimensions, encompassing **27** benchmarks, with results demonstrating significant and consistent improvements in model performance (Figure.1).

Preference learning is critical for aligning large language models (LLMs) with human values, with the quality of preference datasets playing a crucial role in this process. While existing metrics primarily assess data quality based on either explicit or implicit reward margins, their single-margin focus often leads to contradictory evaluations for the same data. To address this issue, we propose a new metric of alignment potential, $M_{AP}$, which integrates both margins to quantify the gap from the model’s current implicit reward margin to the target explicit reward margin, thereby estimating the model’s potential to align on the preference data. Empirical results demonstrate that training on the data selected by $M_{AP}$ consistently enhances alignment performance, surpassing existing metrics across different base models and optimization objectives. Furthermore, our method can be extended to self-play data generation frameworks, where we use this metric to identify high-quality data within the self-generated content by LLMs. Under this data generation scenario, our method surpasses current state-of-the-art methods across various training settings and demonstrates continuous improvements with increasing dataset size and training iterations.

Direct Preference Optimization (DPO) has shown effectiveness in aligning multi-modal large language models (MLLM) with human preferences. However, existing methods exhibit an imbalanced responsiveness to the data of varying hardness, tending to overfit on the easy-to-distinguish data while underfitting on the hard-to-distinguish data. In this paper, we propose Data- and Model-aware DPO (DAMA) to dynamically adjust the optimization process from two key aspects: (1) a data-aware strategy that incorporates data hardness, and (2) a model-aware strategy that integrates real-time model responses. By combining the two strategies, DAMA enables the model to effectively adapt to data with varying levels of hardness. Extensive experiments on five benchmarks demonstrate that DAMA not only significantly enhances the trustworthiness, but also improves the effectiveness over general tasks. For instance, on the Object HalBench, our DAMA-7B reduces response-level and mentioned-level hallucination by 90.0% and 95.3%, respectively, surpassing the performance of GPT-4V.

Aligning large language models (LLMs) with human preferences requires balancing policy optimization with computational stability. While recent offline methods like DPO and SimPO bypass reinforcement learning’s complexity, they face critical limitations: DPO relies on static reference models that degrade with policy updates, and SimPO assumes a uniform target reward margin that ignores instance-wise preference strength. We propose AlphaDPO, an adaptive preference optimization framework that dynamically reparameterizes the reference distribution to address these issues. Our key innovation lies in an implicit reference model \(\hat{\pi}_{\text{ref}} \propto U(y|x)(\pi_\theta/\pi_{\text{ref}})^\alpha\), which interpolates between policy-driven specialization and uniform exploration while enabling instance-adaptive reward margins. Theoretically, we prove AlphaDPO implicitly controls sequential KL divergence between iterative policy updates, ensuring stability even with poorly calibrated reference models. Empirically, AlphaDPO achieves state-of-the-art performance on AlpacaEval 2 (58.7\% LC win rate) and Arena-Hard (35.7\% win rate) across Mistral2-7B, Llama3-8B, and Gemma2-9B, demonstrating robust alignment without multi-stage training. Our work establishes adaptive reference reparameterization as a principled mechanism for preference optimization.

Preference learning has become a common approach in various recent methods for aligning large language models with human values. These methods optimize the preference margin between chosen and rejected responses, subject to certain constraints for avoiding over-optimization. In this paper, we report surprising empirical findings that simple ReLU activation can learn meaningful alignments even using \emph{none} of the following: (i) sigmoid-based gradient constraints, (ii) explicit regularization terms. Our experiments show that over-optimization does exist, but a threshold parameter $\gamma$ plays an essential role in preventing it by dynamically filtering training examples. We further provide theoretical analysis demonstrating that ReLU-based Preference Optimization (RePO) corresponds to the convex envelope of the 0-1 loss, establishing its fundamental soundness. Our ``RePO'' method achieves competitive or superior results compared to established preference optimization approaches. We hope this simple baseline will motivate researchers to rethink the fundamental mechanisms behind preference optimization for language model alignment.

Adapting Large Language Models (LLMs) for agent tasks is critical in developing language agents.Direct Preference Optimization (DPO) is a promising technique for this adaptation with the alleviation of compounding errors, offering a means to directly optimize Reinforcement Learning (RL) objectives.However, applying DPO to multi-turn tasks presents challenges due to the inability to cancel the partition function.Overcoming this obstacle involves making the partition function independent of the current state and addressing length disparities between preferred and dis-preferred trajectories.In this light, we replace the policy constraint with the state-action occupancy measure constraint in the RL objective and add length normalization to the Bradley-Terry model, yielding a novel loss function named DMPO for multi-turn agent tasks with theoretical explanations.Extensive experiments on three multi-turn agent task datasets confirm the effectiveness and superiority of the DMPO loss.

Masked modeling has recently achieved remarkable success in specific fields of vision and language, sparking a surge of interest in graph-related research. However, Masked Graph Modeling (MGM), which captures fine-grained local information by masking low-level elements such as nodes, edges, and features, limits itself to a sub-optimal position, particularly on tasks requiring high-quality graph-level representations. Such a local perspective disregards the graph's global information and structure. To address these limitations, we propose a novel graph pre-training framework called Graph Contrastive Masked Autoencoder (GCMAE). GCMAE leverages the strengths of both MGM and Graph Contrastive Learning (GCL) to provide a more comprehensive perspective of both local and global. Our frame-work uses instance discrimination to learn global representations of graphs and reconstructs the graph using masked low-level elements. We augment the framework with a novel multi-view augmentation module to further enhance the pre-trained model's robustness and generalization ability. We evaluate GCMAE on real-world biochemistry and social network datasets, conducting extensive experiments on both node and graph classification tasks and transfer learning on downstream graph classification tasks. Our experimental results demonstrate that GCMAE's comprehensive perspective of both local and global benefits model pre-training. Moreover, GCMAE outperforms existing MGM and GCL baselines, proving its effectiveness on downstream tasks. Our code is available at https://github.com/lyc0930/GCMAE.

Loss functions steer the optimization direction of recommendation models and are critical to model performance, but have received relatively little attention in recent recommendation research. Among various losses, we find Softmax loss (SL) stands out for not only achieving remarkable accuracy but also better robustness and fairness. Nevertheless, the current literature lacks a comprehensive explanation for the efficacy of SL. Toward addressing this research gap, we conduct theoretical analyses on SL and uncover three insights: 1) Optimizing SL is equivalent to performing Distributionally Robust Optimization (DRO) on the negative data, thereby learning against perturbations on the negative distribution and yielding robustness to noisy negatives. 2) Comparing with other loss functions, SL implicitly penalizes the prediction variance, resulting in a smaller gap between predicted values and and thus producing fairer results. Building on these insights, we further propose a novel loss function Bilateral SoftMax Loss (BSL) that extends the advantage of SL to both positive and negative sides. BSL augments SL by applying the same Log-Expectation-Exp structure to positive examples as is used for negatives, making the model robust to the noisy positives as well. Remarkably, BSL is simple and easy-to-implement - requiring just one additional line of code compared to SL. Experiments on four real-world datasets and three representative backbones demonstrate the effectiveness of our proposal. The code is available at https://github.com/junkangwu/BSL.

Optimization metrics are crucial for building recommendation systems at scale. However, an effective and efficient metric for practical use remains elusive. While Top-K ranking metrics are the gold standard for optimization, they suffer from significant computational overhead. Alternatively, the more efficient accuracy and AUC metrics often fall short of capturing the true targets of recommendation tasks, leading to suboptimal performance. To overcome this dilemma, we propose a new optimization metric, Lower-Left Partial AUC (LLPAUC), which is computationally efficient like AUC but strongly correlates with Top-K ranking metrics. Compared to AUC, LLPAUC considers only the partial area under the ROC curve in the Lower-Left corner to push the optimization focus on Top-K. We provide theoretical validation of the correlation between LLPAUC and Top-K ranking metrics and demonstrate its robustness to noisy user feedback. We further design an efficient point-wise recommendation loss to maximize LLPAUC and evaluate it on three datasets, validating its effectiveness and robustness.

Selection bias hinders recommendation models from learning unbiased user preference. Recent works empirically reveal that pursuing invariant user and item representation across biased and unbiased data is crucial for counteracting selection bias. However, our theoretical analysis reveals that simply optimizing representation invariance is insufficient for addressing the selection bias — recommendation performance is bounded by both representation invariance and discriminability. Worse still, current invariant representation learning methods in recommendation neglect even hurt the representation discriminability due to data sparsity and label shift. In this light, we propose a new Discriminative-Invariant Representation Learning framework for unbiased recommendation, which incorporates label-conditional clustering and prior-guided contrasting into conventional invariant representation learning to mitigate the impact of data sparsity and label shift, respectively. We conduct extensive experiments on three real-world datasets, validating the rationality and effectiveness of the proposed framework. Code and supplementary materials are available at: https://github.com/HungPaan/DIRL.

This study reveals the inherent tolerance of contrastive learning (CL) towards sampling bias, wherein negative samples may encompass similar semantics (\eg labels). However, existing theories fall short in providing explanations for this phenomenon. We bridge this research gap by analyzing CL through the lens of distributionally robust optimization (DRO), yielding several key insights: (1) CL essentially conducts DRO over the negative sampling distribution, thus enabling robust performance across a variety of potential distributions and demonstrating robustness to sampling bias; (2) The design of the temperature $\tau$ is not merely heuristic but acts as a Lagrange Coefficient, regulating the size of the potential distribution set; (3) A theoretical connection is established between DRO and mutual information, thus presenting fresh evidence for ``InfoNCE as an estimate of MI'' and a new estimation approach for $\phi$-divergence-based generalized mutual information. We also identify CL's potential shortcomings, including over-conservatism and sensitivity to outliers, and introduce a novel Adjusted InfoNCE loss (ADNCE) to mitigate these issues. It refines potential distribution, improving performance and accelerating convergence. Extensive experiments on various domains (image, sentence, and graph) validate the effectiveness of the proposal.

Recent years have witnessed the great successes of embedding-based methods in recommender systems. Despite their decent performance, we argue one potential limitation of these methods — the embedding magnitude has not been explicitly modulated, which may aggravate popularity bias and training instability, hindering the model from making a good recommendation. It motivates us to leverage the embedding normalization in recommendation. By normalizing user/item embeddings to a specific value, we empirically observe impressive performance gains (9% on average) on four real-world datasets. Although encouraging, we also reveal a serious limitation when applying normalization in recommendation — the performance is highly sensitive to the choice of the temperature τ which controls the scale of the normalized embeddings.

Negative sampling has been heavily used to train recommender models on large-scale data, wherein sampling hard examples usually not only accelerates the convergence but also improves the model accuracy. Nevertheless, the reasons for the effectiveness of Hard Negative Sampling (HNS) have not been revealed yet. In this work, we fill the research gap by conducting thorough theoretical analyses on HNS. Firstly, we prove that employing HNS on the Bayesian Personalized Ranking (BPR) learner is equivalent to optimizing One-way Partial AUC (OPAUC). Concretely, the BPR equipped with Dynamic Negative Sampling (DNS) is an exact estimator, while with softmax-based sampling is a soft estimator. Secondly, we prove that OPAUC has a stronger connection with Top-K evaluation metrics than AUC and verify it with simulation experiments. These analyses establish the theoretical foundation of HNS in optimizing Top-K recommendation performance for the first time. On these bases, we offer two insightful guidelines for effective usage of HNS: 1) the sampling hardness should be controllable, e.g., via pre-defined hyper-parameters, to adapt to different Top-K metrics and datasets; 2) the smaller the K we emphasize in Top-K evaluation metrics, the harder the negative samples we should draw. Extensive experiments on three real-world benchmarks verify the two guidelines.

Knowledge graph completion (KGC) has become a focus of attention across deep learning community owing to its excellent contribution to numerous downstream tasks. Although recently have witnessed a surge of work on KGC, they are still insufficient to accurately capture complex relations, since they adopt the single and static representations. In this work, we propose a novel Disentangled Knowledge Graph Attention Network (DisenKGAT) for KGC, which leverages both micro-disentanglement and macro-disentanglement to exploit representations behind Knowledge graphs (KGs). To achieve micro-disentanglement, we put forward a novel relation-aware aggregation to learn diverse component representation. For macro-disentanglement, we leverage mutual information as a regularization to enhance independence. With the assistance of disentanglement, our model is able to generate adaptive representations in terms of the given scenario. Besides, our work has strong robustness and flexibility to adapt to various score functions. Extensive experiments on public benchmark datasets have been conducted to validate the superiority of DisenKGAT over existing methods in terms of both accuracy and explainability.