Long-term memory is a critical capability for multimodal large language model (MLLM) agents, particularly in conversational settings where information accumulates and evolves over time. However, existing benchmarks either evaluate multi-session memory in text-only conversations or assess multimodal understanding within localized contexts, failing to evaluate how multimodal memory is preserved, organized, and evolved across long-term conversational trajectories. Thus, we introduce Mem-Gallery, a new benchmark for evaluating multimodal long-term conversational memory in MLLM agents. Mem-Gallery features high-quality multi-session conversations grounded in both visual and textual information, with long interaction horizons and rich multimodal dependencies. Building on this dataset, we propose a systematic evaluation framework that assesses key memory capabilities along three functional dimensions: memory extraction and test-time adaptation, memory reasoning, and memory knowledge management. Extensive benchmarking across twelve memory systems reveals several key findings, highlighting the necessity of explicit multimodal information retention and memory organization, the persistent limitations in memory reasoning and knowledge management, as well as the efficiency bottleneck of current models. Our benchmark and dataset are available at https://github.com/YuanchenBei/Mem-Gallery.

Despite substantial efforts toward improving the moral alignment of Vision-Language Models (VLMs), it remains unclear whether their ethical judgments are stable in realistic settings. This work studies moral robustness in VLMs, defined as the ability to preserve moral judgments under textual and visual perturbations that do not alter the underlying moral context. We systematically probe VLMs with a diverse set of model-agnostic multimodal perturbations and find that their moral stances are highly fragile, frequently flipping under simple manipulations. Our analysis reveals systematic vulnerabilities across perturbation types, moral domains, and model scales, including a sycophancy trade-off where stronger instruction-following models are more susceptible to persuasion. We further show that lightweight inference-time interventions can partially restore moral stability. These results demonstrate that moral alignment alone is insufficient and that moral robustness is a necessary criterion for the responsible deployment of VLMs.

Large language models (LLMs) exhibit complementary strengths arising from differences in pretraining data, model architectures, and decoding behaviors. Inference-time ensembling provides a practical way to combine these capabilities without retraining. However, existing ensemble approaches suffer from fundamental limitations. Most rely on fixed fusion granularity, which lacks the flexibility required for mid-generation adaptation and fails to adapt to different generation characteristics across tasks. To address these challenges, we propose AdaFuse, an adaptive ensemble decoding framework that dynamically selects semantically appropriate fusion units during generation. Rather than committing to a fixed granularity, AdaFuse adjusts fusion behavior on the fly based on the decoding context, with words serving as basic building blocks for alignment. To be specific, we introduce an uncertainty-based criterion to decide whether to apply ensembling at each decoding step. Under confident decoding states, the model continues generation directly. In less certain states, AdaFuse invokes a diversity-aware scaling strategy to explore alternative candidate continuations and inform ensemble decisions. This design establishes a synergistic interaction between adaptive ensembling and test-time scaling, where ensemble decisions guide targeted exploration, and the resulting diversity in turn strengthens ensemble quality. Experiments on open-domain QA, arithmetic reasoning, and machine translation demonstrate that AdaFuse consistently outperforms strong ensemble baselines, achieving an average relative improvement of 6.88%.

Providing Language Models (LMs) with relevant evidence in the context (either via retrieval or user-provided) can significantly improve their ability to provide better-grounded responses. However, recent studies have found that LMs often struggle to fully comprehend and utilize key evidence from the context, especially when it contains noise and irrelevant information—an issue common in real-world scenarios.To address this, we propose SelfElicit, an inference-time approach that helps LMs focus on key contextual evidence through self-guided explicit highlighting.By leveraging the inherent evidence-finding capabilities of LMs using the attention scores of deeper layers, our method automatically identifies and emphasizes key evidence within the input context, facilitating more accurate and grounded responses without additional training or iterative prompting.We demonstrate that SelfElicit brings consistent and significant improvement on multiple evidence-based QA tasks for various LM families while maintaining computational efficiency.Our code and documentation are available at https://github.com/ZhiningLiu1998/SelfElicit.

The advancement of Large Language Models (LLMs) has made ensuring their trustworthiness increasingly critical, especially in terms of fairness across diverse human groups. While modern LLMs are aligned with user preferences through Reinforcement Learning from Human Feedback (RLHF), the reward models used for alignment are trained on preference data that may both reflect societal biases and suffer from demographic skewness, as labeler populations are often uneven due to systemic accessibility or participation gaps. In this work, we reveal that reward models can exhibit significant discrepancies across different demographic groups, posing a fundamental challenge to fair and robust alignment. Using real-world datasets, we conduct the most comprehensive study to date, auditing various state-of-the-art reward models across nine sensitive attributes, including age, gender, ethnicity, etc. Our evaluation spans both (1) the agreement level between reward models and specific user groups, and (2) the reward model’s preference toward responses associated with different groups. Based on these findings, we propose the first method to mitigate group disparities in reward modeling. Code is available at https://github.com/Violet24K/FaRM.

With a knowledge graph and a set of if-then rules, can we reason about the conclusions given a set of observations? In this work, we formalize this question as the cognitive inference problem, and introduce the Cognitive Knowledge Graph (CogKG) that unifies two representations of heterogeneous symbolic knowledge: expert rules and relational facts. We propose a general framework in which the unified knowledge representations can perform both learning and reasoning. Specifically, we implement the above framework in two settings, depending on the availability of labeled data. When no labeled data are available for training, the framework can directly utilize symbolic knowledge as the decision basis and perform reasoning. When labeled data become available, the framework casts symbolic knowledge as a trainable neural architecture and optimizes the connection weights among neurons through gradient descent. Empirical study on two clinical diagnosis benchmarks demonstrates the superiority of the proposed method over time-tested knowledge-driven and data-driven methods, showing the great potential of the proposed method in unifying heterogeneous symbolic knowledge, i.e., expert rules and relational facts, as the substrate of machine learning and reasoning models.