Kairui Hu, Penghao Wu, Fanyi Pu, Wang Xiao, Xiang Yue, Bo Li, Yuanhan Zhang, Ziwei Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Robust deployment of large multimodal models (LMMs) in real-world scenarios requires access to external knowledge sources, given the complexity and dynamic nature of real-world information. Existing approaches such as retrieval-augmented generation (RAG) and prompt engineered search agents rely on rigid pipelines, often leading to inefficient or excessive search behaviors. We present MMSearch-R1, the first end-to-end reinforcement learning framework that enables LMMs to perform on-demand, multi-turn search in real-world Internet environments. Our framework integrates both image and text search tools, allowing the model to reason about when and how to invoke them guided by an outcome-based reward with a search penalty. To support training, We collect a multimodal search VQA dataset through a semi-automated pipeline that covers diverse visual and textual knowledge needs and curate a search-balanced subset with both search-required and search-free samples, which proves essential for shaping efficient and on-demand search behavior. Extensive experiments on knowledge-intensive and info-seeking VQA tasks show that our model not only outperforms RAG-based baselines of the same model size, but also matches the performance of a larger RAG-based model while reducing search calls by over 30%. We further analyze key empirical findings to offer actionable insights for advancing research in multimodal search.

Jiawei Guo, Tianyu Zheng, Yizhi Li, Yuelin Bai, Bo Li, Yubo Wang, King Zhu, Graham Neubig, Wenhu Chen, Xiang Yue. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

We introduce EgoLife, a project to develop an egocentric life assistant that accompanies and enhances personal efficiency through AI-powered wearable glasses. To lay the foundation for this assistant, we conducted a comprehensive data collection study where six participants lived together for one week, continuously recording their daily activities—including discussions, shopping, cooking, social-izing, and entertainment—using AI glasses for multimodal person-view video references. This effort resulted in EgoLife Dataset, a comprehensive 300-hour egocentric, terpersonal, multiview, and multimodal daily life with intensive annotation. Leveraging this dataset, we troduce EgoLifeQA, a suite of long-context, life-oriented question-answering tasks designed to provide meaningful sistance in daily life by addressing practical questions as recalling past relevant events, monitoring health and offering personalized recommendations.To address the key technical challenges of 1) developing robust visual-audio models for egocentric data, 2) enabling identity recognition, and 3) facilitating long-context question answering over extensive temporal information, we introduce EgoBulter, an integrated system comprising EgoGPT and EgoRAG. EgoGPT is an omni-modal model trained on egocentric datasets, achieving state-of-the-art performance on egocentric video understanding. EgoRAG is a retrieval-based component that supports answering ultra-long-context questions. Our experimental studies verify their working mechanisms and reveal critical factors and bottlenecks, guiding future improvements. By releasing our datasets, models, and benchmarks, we aim to stimulate further research in egocentric AI assistants.

Vision-Language Models have made significant progress on many perception-focused tasks.However, their progress on reasoningfocused tasks remains limited due to the lack of high-quality and diverse training data.In this work, we aim to address the scarcity of reasoning-focused multimodal datasets.We propose VisualWebInstruct, a novel approach that leverages search engines to create a diverse and high-quality dataset spanning multiple disciplines, including mathematics, physics, finance, and chemistry, etc. Starting with a meticulously selected set of 30,000 seed images, we employ Google Image Search to identify websites containing similar images.We collect and process HTML data from over 700K unique URLs.Through a pipeline of content extraction, filtering, and synthesis, we construct a dataset of approximately 900K question-answer (QA) pairs, with 40% consisting of visual QA pairs and the remaining comprising text-based QA pairs.Models finetuned on VisualWebInstruct demonstrate significant performance improvements: (1) finetuning on Llava-OV results in 10-20 absolute points improvement across benchmarks, and (2) fine-tuning from MAmmoTH-VL yields a 5 absolute points gain across benchmarks.Our best model, MAmmoTH-VL2, achieves the best known performance with SFT without RL within the 10B parameter class on MMMU-Pro (40.7),MathVerse (42.6), and DynaMath (55.7).These results highlight the effectiveness of our dataset in enhancing the reasoning capabilities of vision-language models for complex multimodal tasks.

Visual instruction tuning has made considerable strides in enhancing the capabilities of Large Multimodal Models (LMMs). However, existing open LMMs largely focus on single-image tasks, their applications to multi-image scenarios remains less explored. Additionally, prior LMM research separately tackles different scenarios, leaving it impossible to generalize cross scenarios with new emerging capabilities. To this end, we introduce LLaVA-Interleave, which simultaneously tackles Multi-image, Multi-frame (video), Multi-view (3D), and Multi-patch (single-image) scenarios in LMMs. To enable these capabilities, we regard the interleaved data format as a general template and compile the M4-Instruct dataset with 1,177.6k samples, spanning 4 primary domains with 14 tasks and 41 datasets. We also curate the LLaVA-Interleave Bench to comprehensively evaluate the multi-image performance of LMMs. Through extensive experiments, LLaVA-Interleave achieves leading results in multi-image, video, and 3D benchmarks, while maintaining the performance of single-image tasks. Besides, our model also exhibits several emerging capabilities, e.g., transferring tasks across different settings and modalities.

Perceiving and generating diverse modalities are crucial for AI models to effectively learn from and engage with real-world signals, necessitating reliable evaluations for their development. We identify two major issues in current evaluations: (1) inconsistent standards, shaped by different communities with varying protocols and maturity levels; and (2) significant query, grading, and generalization biases. To address these, we introduce MixEval-X, the first any-to-any, real-world benchmark designed to optimize and standardize evaluations across diverse input and output modalities. We propose multi-modal benchmark mixture and adaptation-rectification pipelines to reconstruct real-world task distributions, ensuring evaluations generalize effectively to real-world use cases. Extensive meta-evaluations show our approach effectively aligns benchmark samples with real-world task distributions. Meanwhile, MixEval-X's model rankings correlate strongly with that of crowd-sourced real-world evaluations (up to 0.98) while being much more efficient. We provide comprehensive leaderboards to rerank existing models and organizations and offer insights to enhance understanding of multi-modal evaluations and inform future research.

Advancements in large pre-trained generative models have expanded their potential as effective data generators in visual recognition. This work delves into the impact of generative images, primarily comparing paradigms that harness external data (i.e. generative vs. retrieval vs. original). Our key contributions are: 1) GenBench Construction: We devise GenBench, a broad benchmark comprising 22 datasets with 2548 categories, to appraise generative data across various visual recognition tasks. 2) CLER Score: To address the insufficient correlation of existing metrics (e.g., FID, CLIP score) with downstream recognition performance, we propose CLER, a training-free metric indicating generative data's efficiency for recognition tasks prior to training. 3) New Baselines: Comparisons of generative data with retrieved data from the same external pool help to elucidate the unique traits of generative data. 4) External Knowledge Injection: By fine-tuning special token embeddings for each category via Textual Inversion, performance improves across 17 datasets, except when dealing with low-resolution reference images. Our exhaustive benchmark and analysis spotlight generative data's promise in visual recognition, while identifying key challenges for future investigation.

Recent advances in Large Multimodal Models (LMMs) have unveiled great potential as visual assistants. However, most existing works focus on responding to individual instructions or using previous dialogues for contextual understanding. There is little discussion on employing both images and text as in-context examples to enhance the instruction following capability. To bridge this gap, we introduce the Otter model to leverage both textual and visual in-context examples for instruction tuning. Specifically, Otter builds upon Flamingo with Perceiver architecture, and has been instruction tuned for general purpose multi-modal assistant. Otter seamlessly processes multi-modal inputs, supporting modalities including text, multiple images, and dynamic video content. To support the training of Otter, we present the MIMIC-IT (MultI-Modal In-Context Instruction Tuning) dataset, which encompasses over 3 million multi-modal instruction-response pairs, including approximately 2.2 million unique instructions across a broad spectrum of images and videos. MIMIC-IT has been carefully curated to feature a diverse array of in-context examples for each entry. Comprehensive evaluations suggest that instruction tuning with these in-context examples substantially enhances model convergence and generalization capabilities. Notably, the extensive scenario coverage provided by the MIMIC-IT dataset empowers the Otter model to excel in tasks involving complex video and multi-image understanding.

This paper takes a crucial step in the development of energy-aware (EA) NAS methods by offering a benchmark that enhances the reproducibility and accessibility of EA-NAS research. Specifically, we introduce EA-HAS-Bench, the first large-scale energy-aware benchmark designed to enable the study of AutoML methods in achieving improved trade-offs between performance and search energy consumption. EA-HAS-Bench offers a vast architecture/hyperparameter joint search space, encompassing diverse configurations relevant to energy consumption, and proposes a novel surrogate model based on Bézier curves for predicting learning curves with versatile shapes and lengths. On the other hand, recent studies have started integrating large language models (LLMs) into AutoML frameworks to enhance model search efficiency and configuration prediction, yet challenges remain in adapting these methods for energy-efficient searches across vast configuration spaces, as they often neglect energy consumption metrics. As a result, we introduce the Language-Enhanced Shrinkage Search (LESS), a plug-and-play method that utilizes the analytical capabilities of LLMs to enhance the energy efficiency of existing hyperparameter optimization techniques. Moreover, we adapt existing AutoML algorithms to construct baselines. Our experiments demonstrate that these modified energy-aware AutoML methods and LESS achieve an improved balance between energy consumption and model performance.

Abstract To address the issue of low efficiency caused by the repeated use of quantum attack resistant static identity authentication methods in a satellite integrated smart grid, this paper proposes a quantum attack resistant continuous identity authentication protocol. First, in the initial authentication stage, in order to reduce computational complexity, the key encryption mechanism in the CRYSTALS‐Kyber algorithm was improved and combined with the NTRU message recovery digital signature scheme to construct a lattice based explicit AKE (Kyber NTRU. AKE), which achieved mutual authentication and negotiated shared tokens. Second, in the continuous authentication stage, incorporating quantum attack resistant tokens into the current algorithm to improve authentication efficiency. The formal analysis results indicate that compared to the weakly forward secure Kyber.AKE in the CRYSTALS‐Kyber algorithm, Kyber‐NTRU.AKE achieves complete forward secrecy, while the non‐formal analysis results demonstrate the security of the continuous authentication phase. Through theoretical analysis and efficiency comparison with Cyber.AKE, the analysis shows that the Cyber‐NTRU.AKE has higher computational and communication efficiency than Cyber.AKE.

Towards building comprehensive real-world visual perception systems, we propose and study a new problem called panoptic scene graph generation (PVSG). PVSG is related to the existing video scene graph generation (VidSGG) problem, which focuses on temporal interactions between humans and objects localized with bounding boxes in videos. However, the limitation of bounding boxes in detecting non-rigid objects and backgrounds often causes VidSGG systems to miss key details that are crucial for comprehensive video understanding. In contrast, PVSG requires nodes in scene graphs to be grounded by more precise, pixel-level segmentation masks, which facilitate holistic scene understanding. To advance research in this new area, we contribute a high-quality PVSG dataset, which consists of 400 videos (289 third-person + 111 egocentric videos) with totally 150K frames labeled with panoptic segmentation masks as well as fine, temporal scene graphs. We also provide a variety of baseline methods and share useful design practices for future work.

Visual reasoning requires multimodal perception and commonsense cognition of the world. Recently, multiple vision-language models (VLMs) have been proposed with excellent commonsense reasoning ability in various domains. However, how to harness the collective power of these complementary VLMs is rarely explored. Existing methods like ensemble still struggle to aggregate these models with the desired higher-order communications. In this work, we propose Cola, a novel paradigm that coordinates multiple VLMs for visual reasoning. Our key insight is that a large language model (LLM) can efficiently coordinate multiple VLMs by facilitating natural language communication that leverages their distinct and complementary capabilities. Extensive experiments demonstrate that our instruction tuning variant, Cola-FT, achieves state-of-the-art performance on visual question answering (VQA), outside knowledge VQA, visual entailment, and visual spatial reasoning tasks. Moreover, we show that our in-context learning variant, Cola-Zero, exhibits competitive performance in zero and few-shot settings, without finetuning. Through systematic ablation studies and visualizations, we validate that a coordinator LLM indeed comprehends the instruction prompts as well as the separate functionalities of VLMs; it then coordinates them to enable impressive visual reasoning capabilities.

Invariant risk minimization (IRM) has recently emerged as a promising alternative for domain generalization. Nevertheless, the loss function is difficult to optimize for nonlinear classifiers and the original optimization objective could fail when pseudo-invariant features and geometric skews exist. Inspired by IRM, in this paper we propose a novel formulation for domain generalization, dubbed invariant information bottleneck (IIB). IIB aims at minimizing invariant risks for nonlinear classifiers and simultaneously mitigating the impact of pseudo-invariant features and geometric skews. Specifically, we first present a novel formulation for invariant causal prediction via mutual information. Then we adopt the variational formulation of the mutual information to develop a tractable loss function for nonlinear classifiers. To overcome the failure modes of IRM, we propose to minimize the mutual information between the inputs and the corresponding representations. IIB significantly outperforms IRM on synthetic datasets, where the pseudo-invariant features and geometric skews occur, showing the effectiveness of proposed formulation in overcoming failure modes of IRM. Furthermore, experiments on DomainBed show that IIB outperforms 13 baselines by 0.9% on average across 7 real datasets.

Visual correspondence of 2D animation is the core of many applications and deserves careful study. Existing correspondence datasets for 2D cartoon suffer from simple frame composition and monotonic movements, making them insufficient to simulate real animations. In this work, we present a new 2D animation visual correspondence dataset, AnimeRun, by converting open source 3D movies to full scenes in 2D style, including simultaneous moving background and interactions of multiple subjects. Statistics show that our proposed dataset not only resembles real anime more in image composition, but also possesses richer and more complex motion patterns compared to existing datasets. With this dataset, we establish a comprehensive benchmark by evaluating several existing optical flow and segment matching methods, and analyze shortcomings of these methods on animation data. Data are available at https://lisiyao21.github.io/projects/AnimeRun.

Out-of-distribution (OOD) detection is vital to safety-critical machine learning applications and has thus been extensively studied, with a plethora of methods developed in the literature. However, the field currently lacks a unified, strictly formulated, and comprehensive benchmark, which often results in unfair comparisons and inconclusive results. From the problem setting perspective, OOD detection is closely related to neighboring fields including anomaly detection (AD), open set recognition (OSR), and model uncertainty, since methods developed for one domain are often applicable to each other. To help the community to improve the evaluation and advance, we build a unified, well-structured codebase called OpenOOD, which implements over 30 methods developed in relevant fields and provides a comprehensive benchmark under the recently proposed generalized OOD detection framework. With a comprehensive comparison of these methods, we are gratified that the field has progressed significantly over the past few years, where both preprocessing methods and the orthogonal post-hoc methods show strong potential.

While self-supervised pretraining has proven beneficial for many computer vision tasks, it requires expensive and lengthy computation, large amounts of data, and is sensitive to data augmentation. Prior work demonstrates that models pretrained on datasets dissimilar to their target data, such as chest X-ray models trained on ImageNet, underperform models trained from scratch. Users that lack the resources to pretrain must use existing models with lower performance. This paper explores Hierarchical PreTraining (HPT), which decreases convergence time and improves accuracy by initializing the pretraining process with an existing pretrained model. Through experimentation on 16 diverse vision datasets, we show HPT converges up to 80× faster, improves accuracy across tasks, and improves the robustness of the self-supervised pretraining process to changes in the image augmentation policy or amount of pretraining data. Taken together, HPT provides a simple framework for obtaining better pretrained representations with less computational resources.

Large-scale labeled training datasets have enabled deep neural networks to excel across a wide range of benchmark vision tasks. However, in many applications, it is prohibitively expensive and time-consuming to obtain large quantities of labeled data. To cope with limited labeled training data, many have attempted to directly apply models trained on a large-scale labeled source domain to another sparsely labeled or unlabeled target domain. Unfortunately, direct transfer across domains often performs poorly due to the presence of domain shift or dataset bias. Domain adaptation (DA) is a machine learning paradigm that aims to learn a model from a source domain that can perform well on a different (but related) target domain. In this article, we review the latest single-source deep unsupervised DA methods focused on visual tasks and discuss new perspectives for future research. We begin with the definitions of different DA strategies and the descriptions of existing benchmark datasets. We then summarize and compare different categories of single-source unsupervised DA methods, including discrepancy-based methods, adversarial discriminative methods, adversarial generative methods, and self-supervision-based methods. Finally, we discuss future research directions with challenges and possible solutions.

Due to its robust and precise distance measurements, LiDAR plays an important role in scene understanding for autonomous driving. Training deep neural networks (DNNs) on LiDAR data requires large-scale point-wise annotations, which are time-consuming and expensive to obtain. Instead, simulation-to-real domain adaptation (SRDA) trains a DNN using unlimited synthetic data with automatically generated labels and transfers the learned model to real scenarios. Existing SRDA methods for LiDAR point cloud segmentation mainly employ a multi-stage pipeline and focus on feature-level alignment. They require prior knowledge of real-world statistics and ignore the pixel-level dropout noise gap and the spatial feature gap between different domains. In this paper, we propose a novel end-to-end framework, named ePointDA, to address the above issues. Specifically, ePointDA consists of three modules: self-supervised dropout noise rendering, statistics-invariant and spatially-adaptive feature alignment, and transferable segmentation learning. The joint optimization enables ePointDA to bridge the domain shift at the pixel-level by explicitly rendering dropout noise for synthetic LiDAR and at the feature-level by spatially aligning the features between different domains, without requiring the real-world statistics. Extensive experiments adapting from synthetic GTA-LiDAR to real KITTI and SemanticKITTI demonstrate the superiority of ePointDA for LiDAR point cloud segmentation.

The success of supervised learning hinges on the assumption that the training and test data come from the same underlying distribution, which is often not valid in practice due to potential distribution shift. In light of this, most existing methods for unsupervised domain adaptation focus on achieving domain-invariant representations and small source domain error. However, recent works have shown that this is not sufficient to guarantee good generalization on the target domain, and in fact, is provably detrimental under label distribution shift. Furthermore, in many real-world applications it is often feasible to obtain a small amount of labeled data from the target domain and use them to facilitate model training with source data. Inspired by the above observations, in this paper we propose the first method that aims to simultaneously learn invariant representations and risks under the setting of semi-supervised domain adaptation (Semi-DA). First, we provide a finite sample bound for both classification and regression problems under Semi-DA. The bound suggests a principled way to obtain target generalization, i.e., by aligning both the marginal and conditional distributions across domains in feature space. Motivated by this, we then introduce the LIRR algorithm for jointly Learning Invariant Representations and Risks. Finally, extensive experiments are conducted on both classification and regression tasks, which demonstrate that LIRR consistently achieves state-of-the-art performance and significant improvements compared with the methods that only learn invariant representations or invariant risks. Our code will be released at LIRR@github

Confidence calibration is of great importance to the reliability of decisions made by machine learning systems. However, discriminative classifiers based on deep neural networks are often criticized for producing overconfident predictions that fail to reflect the true correctness likelihood of classification accuracy. We argue that such an inability to model uncertainty is mainly caused by the closed-world nature in softmax: a model trained by the cross-entropy loss will be forced to classify input into one of K pre-defined categories with high probability. To address this problem, we for the first time propose a novel K+1-way softmax formulation, which incorporates the modeling of open-world uncertainty as the extra dimension. To unify the learning of the original K-way classification task and the extra dimension that models uncertainty, we 1) propose a novel energy-based objective function, and moreover, 2) theoretically prove that optimizing such an objective essentially forces the extra dimension to capture the marginal data distribution. Extensive experiments show that our approach, Energy-based Open-World Softmax (EOW-Softmax), is superior to existing state-of-the-art methods in improving confidence calibration.

Simulation-to-real domain adaptation for semantic segmentation has been actively studied for various applications such as autonomous driving. Existing methods mainly focus on a single-source setting, which cannot easily handle a more practical scenario of multiple sources with different distributions. In this paper, we propose to investigate multi-source domain adaptation for semantic segmentation. Specifically, we design a novel framework, termed Multi-source Adversarial Domain Aggregation Network (MADAN), which can be trained in an end-to-end manner. First, we generate an adapted domain for each source with dynamic semantic consistency while aligning at the pixel-level cycle-consistently towards the target. Second, we propose sub-domain aggregation discriminator and cross-domain cycle discriminator to make different adapted domains more closely aggregated. Finally, feature-level alignment is performed between the aggregated domain and target domain while training the segmentation network. Extensive experiments from synthetic GTA and SYNTHIA to real Cityscapes and BDDS datasets demonstrate that the proposed MADAN model outperforms state-of-the-art approaches. Our source code is released at: https://github.com/Luodian/MADAN.