Although Vision Language Models (VLMs) have excelled at image and video understanding, applying them to hour-long videos is held back by two interrelated challenges: exorbitant computational expense and a qualitative breakdown in long-term temporal reasoning. Thus, models tend to generate answers based on speculation instead of solid visual facts, causing both factually incorrect and plausible hallucinations. This problem is compounded by current benchmarks that, by only emphasizing final answers, lack an effective mechanism to check whether reasoning is substantiated by specific visual evidence. This makes it hard to differentiate between true understanding and pretend comprehension, inhibiting targeted model refinement. To address these interrelated challenges of model fragility and evaluation weakness, we adopt a twofold strategy. First, we present EV²-Bench, a large-scale benchmark that breaks new ground by an evaluation paradigm built upon spatio-temporal visual evidence, forcing models to justify answers with checkable hints. Second, we put forward DynamicSelect, an adaptive token compression system that efficiently condenses salient information by a dynamic semantic selector and a hierarchical compression strategy. Comprehensive experiments demonstrate that DynamicSelect significantly outperforms the baselines on EV²-Bench as well as other public benchmarks. Our study offers not only a more effective approach to long-video understanding but also a more stringent evaluation paradigm, indicating the way toward more robust models.

Medical large Vision-Language Models (Med-LVLMs) have shown strong potential in multimodal clinical applications such as medical visual question answering and report generation. However, Med-LVLMs remain challenged by hallucinations caused by modality misalignment, where models prioritize textual knowledge over visual evidence and generate outputs that conflict with medical images. To mitigate this issue, recent studies have explored preference optimization to improve image–text alignment, achieving promising results. Despite these advances, existing preference-based methods still face two limitations in medical settings: (1) overfitting to superficial cues, and (2) pseudo convergence of the preference signal. In this paper, we propose Dynamic Evidence-Guided Preference Optimization (DEPO), a new framework that enables evidence-aware and adaptive preference learning for Med-LVLMs. DEPO introduces Multi-Modal Evidence Perturbation (MEP) to suppress non-causal textual and visual shortcuts, and Dispreferred Evidence Resampling (DER) to continuously update dispreferred responses as hallucination patterns evolve. Experiments on multiple medical VQA and report generation benchmarks demonstrate consistent improvements over existing methods, with strong robustness across datasets and architectures. All Codes and data will be released after review.

Visual Entity Linking (VEL) is a crucial task for achieving fine-grained visual understanding, matching objects within images (visual mentions) to entities in a knowledge base. Previous VEL tasks rely on textual inputs, but writing queries for complex scenes can be challenging. Visual inputs like clicks or bounding boxes offer a more convenient alternative. Therefore, we propose a new task, Pixel-Level Visual Entity Linking (PL-VEL), which uses pixel masks from visual inputs to refer to objects, supplementing reference methods for VEL. To facilitate research on this task, we have constructed the MaskOVEN-Wiki dataset through an entirely automatic reverse region-entity annotation framework. This dataset contains over 5 million annotations aligning pixel-level regions with entity-level labels, which will advance visual understanding towards fine-grained. Moreover, as pixel masks correspond to semantic regions in an image, we enhance previous patch-interacted attention with region-interacted attention by a visual semantic tokenization approach. Manual evaluation results indicate that the reverse annotation framework achieved a 94.8% annotation success rate. Experimental results show that models trained on this dataset improved accuracy by 18 points compared to zero-shot models. Additionally, the semantic tokenization method achieved a 5-point accuracy improvement over the trained baseline.

Referring Multi-Object Tracking (RMOT) aims to track multiple objects based on a provided language expression. Although prior studies have sought to accomplish this by integrating an textual module into the multi-object tracker, these methods combine text and image features in a basic way, neglecting the importance of text features. In this study, we propose a Hierarchical Fine-grained text-image Fusion tracker, named HFF-Tracker, which can perform fine-grained fusion of pixel-level visual features and text features across various semantic levels. Specifically, we have devised a Hierarchical Multi-Modal Fusion (HMMF) module to merge text and image features at an early stage in a hierarchical and detailed manner. The Text-Guided Decoder (TGD) is designed to provide the query with prior semantic information during the decoding process. Additionally, we have crafted a Text-Guided Prediction Head (TGPH) that utilizes text information to enhance the performance of the prediction head. Furthermore, we have implemented an adaptive Look-Back training strategy to maximize the utilization of valuable labeled data. Extensive experiments on the Refer-KITTI dataset and the Refer-KITTI-V2 dataset demonstrate that our proposed HFF-Tracker outperforms other state-of-the-art methods with remarkable margins.

Chain-of-Thought (CoT) prompting combined with large language models (LLM) has shown great potential in improving performance on challenging reasoning tasks. While understanding why CoT prompting is effective is crucial for the application and improvement of CoT prompting, few studies have addressed this issue. Besides, almost no prior work has conducted theoretical analysis on CoT prompting in the context of black-box models. In this paper, we approach the analysis of CoT prompting in black-box LLMs from an information-theoretic perspective. Specifically, we propose a new metric, EPVI (Estimated Pointwise V-Information), which extends the concept of pointwise V-information to black-box models, quantifying the label-relevant new information introduced by CoT prompting beyond the pre-existing information in the input. Based on this, we conduct a series of experiments at both the task and instance levels to analyze CoT prompting, demonstrating that the effectiveness of CoT prompting can be attributed to its capacity to influence the difficulty of model inference by augmenting or reducing the model-usable information. Furthermore, we show that selecting high-quality demonstrations of CoT reasoning based on EPVI can improve the downstream performance of reasoning tasks.

Deyuan Liu, Zhanyue Qin, Hairu Wang, Zhao Yang, Zecheng Wang, Fangying Rong, Qingbin Liu, Yanchao Hao, Bo Li, Xi Chen, Cunhang Fan, Zhao Lv, Dianhui Chu, Zhiying Tu, Dianbo Sui. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Large language models (LLMs) have acquired the ability to solve general tasks by utilizing instruction finetuning (IFT). However, IFT still relies heavily on instance training of extensive task data, which greatly limits the adaptability of LLMs to real-world scenarios where labeled task instances are scarce and broader task generalization becomes paramount. Contrary to LLMs, humans acquire skills and complete tasks not merely through repeated practice but also by understanding and following instructional guidelines. This paper is dedicated to simulating human learning to address the shortcomings of instance training, focusing on instruction learning to enhance cross-task generalization. Within this context, we introduce Task Adapters Generation from Instructions (TAGI), which automatically constructs the task-specific model in a parameter generation manner based on the given task instructions without retraining for unseen tasks. Specifically, we utilize knowledge distillation to enhance the consistency between TAGI developed through Learning with Instruction and task-specific models developed through Training with Instance, by aligning the labels, output logits, and adapter parameters between them. TAGI is endowed with cross-task generalization capabilities through a two-stage training process that includes hypernetwork pretraining and finetuning. We evaluate TAGI on the Super-Natural Instructions and P3 datasets. The experimental results demonstrate that TAGI can match or even outperform traditional meta-trained models and other hypernetwork models, while significantly reducing computational requirements. Our code will be available at https://github.com/Xnhyacinth/TAGI.

Question Answering over Knowledge Bases (KB-QA), which automatically answer natural language questions based on the facts contained by a knowledge base, is one of the most important natural language processing (NLP) tasks. Simple questions constitute a large part of questions queried on the web, still being a challenge to QA systems. In this work, we propose to conduct pattern extraction and entity linking first, and put forward pattern revising procedure to mitigate the error propagation problem. In order to learn to rank candidate subject-predicate pairs to enable the relevant facts retrieval given a question, we propose to do joint fact selection enhanced by relation detection. Multi-level encodings and multi-dimension information are leveraged to strengthen the whole procedure. The experimental results demonstrate that our approach sets a new record in this task, outperforming the current state-of-the-art by an absolute large margin.

With the rapid growth of knowledge bases (KBs) on the web, how to take full advantage of them becomes increasingly important.Question answering over knowledge base (KB-QA) is one of the promising approaches to access the substantial knowledge.Meanwhile, as the neural networkbased (NN-based) methods develop, NNbased KB-QA has already achieved impressive results.However, previous work did not put more emphasis on question representation, and the question is converted into a fixed vector regardless of its candidate answers.This simple representation strategy is not easy to express the proper information in the question.Hence, we present an end-to-end neural network model to represent the questions and their corresponding scores dynamically according to the various candidate answer aspects via cross-attention mechanism.In addition, we leverage the global knowledge inside the underlying KB, aiming at integrating the rich KB information into the representation of the answers.As a result, it could alleviates the out-of-vocabulary (OOV) problem, which helps the crossattention model to represent the question more precisely.The experimental results on WebQuestions demonstrate the effectiveness of the proposed approach.