Multilingual NLP faces challenges of data heterogeneity, privacy, and limited computational resources, especially for low-resource languages. Centralised methods risk privacy breaches, while federated learning struggles with communication overhead and poor cross-lingual generalisation. We propose FLiP (Federated Lightweight Prompt-tuning), a privacy-preserving, resource-efficient, generalizable framework integrating prompt-based learning with federated optimisation. FLiP eliminates communication overhead, reduces trainable parameters to 16%, and cuts GPU memory use by 90%. Experiments show superior generalisation and efficiency under both IID and Non-IID settings, establishing FLiP as a scalable, privacy-aware solution for multilingual NLP, particularly in low-resource and indigenous language contexts.

Multi-agent large language model systems offer a promising paradigm for multidisciplinary medical decision support by integrating heterogeneous expert perspectives. However, existing frameworks often rely on loosely structured interactions, where disagreement is insufficiently diagnosed and discussion may collapse into superficial consensus or parallel opinion aggregation. We propose Structured Adaptive Multi-disciplinary Collaboration (SAMAC), a structured and strategy-aware multi-agent framework for adaptive MDT-style medical reasoning. SAMAC introduces Structured Deliberation to detect opinion disagreement and majority not confident, along with a Consensus Purification mechanism that transforms disagreement into constructive and collaborative refinement. When disagreement arises, SAMAC adaptively selects between Self-refinement and Counter-questioning to enable evidence-grounded negotiation. Experiments on three medical datasets—RADCURE, HANCOCK, and OSCCPI—show that SAMAC consistently improves overall performance and collaborative gains over existing multi-agent baselines. In particular, SAMAC improves the Average Score by 4.29 points on the RADCURE dataset and achieves higher Relative Self-Correction Scores (RSS), indicating more effective resolution of disagreement and stable collective decisions. These results highlight the importance of controlled and diverse interaction dynamics in multi-agent medical reasoning. Code will be released upon publication.

Real-time streaming video understanding in domains such as autonomous driving and intelligent surveillance poses challenges beyond conventional offline video processing, requiring continuous perception, proactive decision making, and responsive interaction based on dynamically evolving visual content. However, existing methods rely on alternating perception-reaction or asynchronous triggers, lacking task-driven planning and future anticipation, which limits their real-time responsiveness and proactive decision making in evolving video streams. To this end, we propose a StreamAgent that anticipates the temporal intervals and spatial regions expected to contain future task-relevant information to enable proactive and goal-driven responses. Specifically, we integrate question semantics and historical observations through prompting the anticipatory agent to anticipate the temporal progression of key events, align current observations with the expected future evidence, and subsequently adjust the perception action (e.g., attending to task-relevant regions or continuously tracking in subsequent frames). To enable efficient inference, we design a streaming KV-cache memory mechanism that constructs a hierarchical memory structure for selective recall of relevant tokens, enabling efficient semantic retrieval while reducing the overhead of storing all tokens in the traditional KV-cache. Extensive experiments on streaming and long video understanding tasks demonstrate that our method outperforms existing methods in response accuracy and real-time efficiency, highlighting its practical value for real-world streaming scenarios.

Retrieval-augmented generation (RAG) remains challenging for multi-hop question answering when supporting evidence is distributed across heterogeneous sources, including text, tables, and knowledge graphs (KGs). Existing systems often rely on modality-specific retrieval pipelines or uncontrolled iterative search, making it difficult to preserve source structure while maintaining efficient evidence acquisition. We present **RELOOP**, a structure-aware retrieval framework built on **Hierarchical Sequence (HSEQ)**, a reversible representation that maps heterogeneous sources into provenance-preserving typed segments with lightweight structural links. Operating on this unified evidence interface, a lightweight iterator selects and expands candidate evidence through structure-respecting actions, while an explicit sufficiency signal determines when the accumulated evidence is adequate for answering. A head module then synthesizes the final answer from a compact, canonicalized evidence package, with an optional refinement step when inconsistent evidence is detected. Experiments on HotpotQA, HybridQA, TAT-QA, and MetaQA show that RELOOP consistently improves answer quality over strong single-pass and iterative RAG baselines while requiring substantially fewer retrieval steps than graph-expansion approaches. These results suggest that a reversible, structure-aware evidence interface provides an effective foundation for accurate and budget-controlled retrieval across heterogeneous QA settings.

Reliable evaluation of vision-language models (VLMs) and medical vision-language models (Medical-VLMs) requires calibrated confidence, particularly under realistic clinical conditions. However, existing efforts mainly focused on improving accuracy, leaving calibration in the medical domain underexplored. To this end, we propose MVC-Bench, a calibration-centric benchmark for medical image classification with VLMs and Medical-VLMs. MVC-Bench assesses the calibration across three axes: (i) robustness to modality, backbone, and domain shift (ii) effectiveness of calibration strategies and prompt-tuning methods (iii) stability under prompt-template and random-seed variations. The benchmark covers eight different backbones, three medical modalities, including fundus imaging, histopathology, and chest X-ray under in-domain and domain shift settings. It compares post-hoc calibration, train-time calibration, and zero-shot inference methods, together with six prompt-tuning methods. Across more than 1638 controlled experiments, we report accuracy and Expected Calibration Error (ECE) as primary metrics, and further report results with complementary calibration measures, including Maximum Calibration Error (MCE) and Adaptive Calibration Error (ACE). We further investigate the underlying causes of miscalibration in VLMs and Medical-VLMs and propose a simple train-time calibration method, Multi-Class Margin (MCM) regularization, which achieves lowest ECE on 10 out of 12 settings in in-domain and remains competitive under domain shifts. Collectively, MVC-Bench provides a structured evaluation framework and actionable guidance for improving calibration in safety-critical medical workflows.

Recent advances such as Chain-of-Thought prompting have significantly improved large language models (LLMs) in zero-shot medical reasoning. However, prompting-based methods often remain shallow and unstable, while fine-tuned medical LLMs suffer from poor generalization under distribution shifts and limited adaptability to unseen clinical scenarios. To address these limitations, we present TAGS, a test-time framework that combines a broadly capable generalist with a domain-specific specialist to offer complementary perspectives without any model fine-tuning or parameter updates. To support this generalist–specialist reasoning process, we introduce two auxiliary modules: a hierarchical retrieval mechanism that provides multi-scale exemplars by selecting examples based on both semantic and rationale-level similarity, and a reliability scorer that evaluates reasoning consistency to guide final answer aggregation. TAGS achieves strong performance across nine MedQA benchmarks, boosting GPT-4o accuracy by 13.8%, DeepSeek-R1 by 16.8%, and improving a vanilla 7B model from 14.1% to 23.9%. These results surpass several fine-tuned medical LLMs, without any parameter updates.

Text-Attributed Graphs (TAGs), where each node is associated with text descriptions, are ubiquitous in real-world scenarios. They typically exhibit distinctive structure and domain-specific knowledge, motivating the development of Graph Foundation Models (GFMs) that generalize across diverse graphs and tasks. Despite large efforts to integrate Large Language Models (LLMs) and Graph Neural Networks (GNNs) for TAGs, most approaches suffer from decoupled architectures with two-stage alignment, limiting their synergistic potential. Even worse, existing methods assign out-of-vocabulary (OOV) tokens to graph nodes, leading to graph-specific semantics, token explosion, and incompatibility with task-oriented prompt templates, which hinders cross-graph and cross-task transferability. To address these challenges, we propose PromptGFM, a versatile GFM for TAGs grounded in graph vocabulary learning. PromptGFM comprises two key components: (a) a graph understanding module, which explicitly prompts LLMs to replicate the fine-grained GNN workflow within the text space, facilitating seamless GNN-LLM integration and elegant graph-text alignment; and (b) a graph inference module, which establishes a language-based graph vocabulary to ensure expressiveness, transferability, and scalability, enabling readable instructions for joint LLM fine-tuning across graphs and tasks. Extensive experiments demonstrate the superiority of PromptGFM, highlighting its strong transferability across diverse graphs and tasks and its adaptability to auxiliary pre-training tasks. The code is available at https://github.com/agiresearch/PromptGFM.

Medical VLMs frequently hallucinate clinically unsupported findings. We identify a stability gap: visually grounded tokens maintain relatively stable logits under semantics-preserving image perturbations, whereas visually misgrounded tokens exhibit substantially larger shifts. We introduce LENS (Logit-ENhanced Stability), which exploits this signal at both training and inference. During training, Semantics-Preserving Preference Optimization (SPPO) scores preference pairs using clinical faithfulness and cross-view stability; at inference, stability-aware contrastive decoding penalizes tokens whose logits diverge across original and counterfactual views. Constrained counterfactuals use attribution-guided masking and feature-space proximity guards to reduce disruption of lesion- and device-level evidence. Across radiology VQA, chest X-ray reporting, and gastrointestinal endoscopy, LENS consistently outperforms strong decoding-time and preference-optimization baselines. On MIMIC-CXR, entity-level RadGraph recall improves from 21.0 to 22.8 over PCCD, while expert evaluation shows a hallucination-rate reduction from 17.0% to 8.5%.

Retrieval-augmented generation over knowledge graphs (KGs) can improve the factual grounding of large language models, but efficient retrieval remains challenging when the underlying graph is large and multi-hop evidence is distributed across different regions. We present **SPLIT-RAG**, an efficiency-oriented multi-agent KG-RAG framework that organises a KG into reusable, question-aligned partitions using reasoning patterns observed in training questions. At inference time, SPLIT-RAG forms a query-specific retrieval plan, activates only a small set of relevant partition-specialised agents, and aggregates their retrieved evidence before final answer generation. To improve reliability when evidence from different agents is combined, the framework applies compatibility-based filtering over retrieved triples prior to synthesis. Experiments on WebQSP, CWQ, and MetaQA show that SPLIT-RAG achieves strong performance against recent KG+LLM baselines while reducing end-to-end latency under matched budgets. Component ablations and model-tier analyses further show that selective multi-agent retrieval is central to the performance gain, and that lightweight retrieval agents paired with a stronger head agent provide an effective accuracy–latency trade-off.

Medical benchmarks are dominated by single-turn, multiple-choice clinical cases that poorly reflect real consultations. Practically, clinicians elicit evidence interactively, and patient communication varies widely. We introduce MEDROUNDSQA, a multi-turn diagnostic benchmark derived from 1,387 clinical cases across 17 specialties. Each case is converted into a structured 24-slot clinical record and then instantiated as controlled doctor–patient dual-agent dialogues under varying patient personas, with the underlying clinical content held fixed. We further classify cases by difficulty using model-based uncertainty to enable easy-to-hard analysis. Evaluations of fifteen LLM doctor agents show that (i) moving from a single-turn diagnosis on the standardized records to multi-turn consultations causes large degradations of roughly 13–39 points; (ii) more turns reliably improve question relevance, but diagnostic accuracy exhibits diminishing returns and typically plateaus after 6–12 turns; and (iii) patient persona differences can shift diagnosis accuracy by about 7–8 points from the lowest to highest education levels, highlighting equity risks that single-turn benchmarks miss.

Effective long-term memory management is crucial for language models handling extended contexts. We introduce the Enhanced Ranked Memory Augmented Retrieval ERMAR framework, which dynamically ranks memory entries based on relevance. Unlike prior models, ERMAR employs a novel relevance scoring mechanism and a pointwise re-ranking model for key-value embeddings, inspired by learning-to-rank techniques in information retrieval. By integrating historical usage patterns and adaptive retrieval, ERMAR achieves state-of-the-art results on standard benchmarks, demonstrating superior scalability and performance in long-context tasks.

Infodemics and health misinformation have significant negative impact on individuals and society, exacerbating confusion and increasing hesitancy in adopting recommended health measures. Recent advancements in generative AI, capable of producing realistic, human-like text and images, have significantly accelerated the spread and expanded the reach of health misinformation, resulting in an alarming surge in its dissemination. To combat the infodemics, most existing work has focused on developing misinformation datasets from social media and fact-checking platforms, but has faced limitations in topical coverage, inclusion of AI-generation, and accessibility of raw content. To address these gaps, we present MM-Health, a large scale multimodal misinformation dataset in the health domain consisting of 34,746 news article encompassing both textual and visual information. MM-Health includes human-generated multimodal information (5,776 articles) and AI-generated multimodal information (28,880 articles) from various SOTA generative AI models. Additionally, We benchmarked our dataset against three tasks—reliability checks, originality checks, and fine-grained AI detection—demonstrating that existing SOTA models struggle to accurately distinguish the reliability and origin of information. Our dataset aims to support the development of misinformation detection across various health scenarios, facilitating the detection of human and machine-generated content at multimodal levels.

Medical decision-making often involves integrating knowledge from multiple clinical specialties, typically achieved through multidisciplinary teams. Inspired by this collaborative process, recent work has leveraged large language models (LLMs) in multi-agent collaboration frameworks to emulate expert teamwork. While these approaches improve reasoning through agent interaction, they are limited by static, pre-assigned roles, which hinder adaptability and dynamic knowledge integration. To address these limitations, we propose KAMAC, a Knowledge-driven Adaptive Multi-Agent Collaboration framework that enables LLM agents to dynamically form and expand expert teams based on the evolving diagnostic context. KAMAC begins with one or more expert agents and then conducts a knowledge-driven discussion to identify and fill knowledge gaps by recruiting additional specialists as needed. This supports flexible, scalable collaboration in complex clinical scenarios, with decisions finalized through reviewing updated agent comments. Experiments on two real-world medical benchmarks demonstrate that KAMAC significantly outperforms both single-agent and advanced multi-agent methods, particularly in complex clinical scenarios (i.e., cancer prognosis) requiring dynamic, cross-specialty expertise. Our code is publicly available at: https://github.com/XiaoXiao-Woo/KAMAC.

NLP models often face challenges with under-represented languages due to a lack of sufficient training data and language complexities. This can result in inaccurate predictions and a failure to capture the inherent uncertainties within these languages. This paper introduces a new method for modelling uncertainty in under-represented languages by employing deep Bayesian Gaussian Processes. We develop a novel framework that integrates prior knowledge and leverages kernel functions. This helps enable the quantification of uncertainty in predictions to overcome the data limitations in under-represented languages. The efficacy of our approach is validated through various experiments, and the results are benchmarked against existing methods to highlight the enhancements in prediction accuracy and measurement of uncertainty.

Multi-level Hierarchical Classification (MLHC) tackles the challenge of categorizing items within a complex, multi-layered class structure. However, traditional MLHC classifiers often rely on a backbone model with n independent output layers, which tend to ignore the hierarchical relationships between classes. This oversight can lead to inconsistent predictions that violate the underlying taxonomy. Leveraging Large Language Models (LLMs), we propose novel taxonomy-embedded transitional LLM-agnostic framework for multimodality classification. The cornerstone of this advancement is the ability of models to enforce consistency across hierarchical levels. Our evaluations on the MEP-3M dataset - a Multi-modal E-commerce Product dataset with various hierarchical levels- demonstrated a significant performance improvement compared to conventional LLMs structure.

Understanding and interpreting culturally specific language remains a significant challenge for multilingual natural language processing (NLP) systems, particularly for less-resourced languages. To address this problem, this paper introduces PRONE, a novel dataset of 2,830 Nepali proverbs, and evaluates the performance of various language models (LMs) in two tasks: (i) identifying the correct meaning of a proverb from multiple choices, and (ii) categorizing proverbs into predefined thematic categories. The models, including both open-source and proprietary, were tested in zero-shot and few-shot settings with prompts in English and Nepali. While models like GPT-4o demonstrated promising results and achieved the highest performance among LMs, they still fall short of human-level accuracy in understanding and categorizing culturally nuanced content, highlighting the need for more inclusive NLP.

Recent multimodal large language models (MLLMs) have demonstrated significant potential in open-ended conversation, generating more accurate and personalized responses. However, their abilities to memorize, recall, and reason in sustained interactions within real-world scenarios remain underexplored. This paper introduces MMRC, a Multi-Modal Real-world Conversation benchmark for evaluating six core open-ended abilities of MLLMs: information extraction, multi-turn reasoning, information update, image management, memory recall, and answer refusal. With data collected from real-world scenarios, MMRC comprises 5,120 conversations and 28,720 corresponding manually labeled questions, posing a significant challenge to existing MLLMs. Evaluations on 20 MLLMs in MMRC indicate an accuracy drop during open-ended interactions. We identify four common failure patterns: long-term memory degradation, inadequacies in updating factual knowledge, accumulated assumption of error propagation, and reluctance to “say no.” To mitigate these issues, we propose a simple yet effective NOTE-TAKING strategy, which can record key information from the conversation and remind the model during its responses, enhancing conversational capabilities. Experiments across six MLLMs demonstrate significant performance improvements.

While the Large Language Models (LLMs) have been popular in different tasks, their capability to handle health-related claims in diverse linguistic and cultural contexts, such as Arabic dialects, Saudi, Egyptian, Lebanese, and Moroccan has not been thoroughly explored. To this end, we develop a comprehensive evaluation framework to assess how LLMs particularly GPT-4 respond to health-related claims. Our framework focuses on measuring factual accuracy, consistency, and cultural adaptability. It introduces a new metric, the “Cultural Sensitivity Score”, to evaluate the model’s ability to adjust responses based on dialectal differences. Additionally, the reasoning patterns used by the models are analyzed to assess their effectiveness in engaging with claims across these dialects. Our findings highlight that while LLMs excel in recognizing true claims, they encounter difficulties with mixed and ambiguous claims, especially in underrepresented dialects. This work underscores the importance of dialect-specific evaluations to ensure accurate, contextually appropriate, and culturally sensitive responses from LLMs in real-world applications.

Despite the remarkable performances in various applications, machine learning (ML) models could potentially discriminate. They may result in biasness in decision-making, leading to an impact negatively on individuals and society. Recently, various methods have been developed to mitigate biasness and achieve significant performance. Attention mechanisms are a fundamental component of many state-of-the-art ML models and may potentially impact the fairness of ML models. However, how they explicitly influence fairness has yet to be thoroughly explored. In this paper, we investigate how different attention mechanisms affect the fairness of ML models, focusing on models used in Natural Language Processing (NLP) models. We evaluate the performance of fairness of several models with and without different attention mechanisms on widely used benchmark datasets. Our results indicate that the majority of attention mechanisms that have been assessed can improve the fairness performance of Bidirectional Gated Recurrent Unit (BiGRU) and Bidirectional Long Short-Term Memory (BiLSTM) in all three datasets regarding religious and gender-sensitive groups, however, with varying degrees of trade-offs in accuracy measures. Our findings highlight the possibility of fairness being affected by adopting specific attention mechanisms in machine learning models for certain datasets

This paper presents a new multi-modal dataset for identifying hateful content on social media, consisting of 5,680 text-image pairs collected from Twitter, labeled across two labels. Experimental analysis of the presented dataset has shown that understanding both modalities is essential for detecting these techniques. It is confirmed in our experiments with several state-of-the-art multi-modal models. In future work, we plan to extend the dataset in size. We further plan to develop new multi-modal models tailored explicitly to hate-speech detection, aiming for a deeper understanding of the text and image relation. It would also be interesting to perform experiments in a direction that explores what social entities the given hate speech tweet targets.