Multimodal in-context learning (ICL) is becoming a key capability that allows large vision-language models (LVLMs) to adapt to novel tasks without parameter updates, which expands their usefulness in many real-world applications. However, ICL performance remains unstable even when the in-context demonstrations (ICDs) are well matched, showing that LVLMs still struggle to make full use of the provided context. While existing work mainly focuses on prompt engineering or post-hoc logit calibration, we study the attention mechanisms inside LVLMs to address their inherent limitations. We identify two important weaknesses in their self-attention that hinder effective ICL. To address these weaknesses, we propose Context-Aware Modulated Attention (CAMA), a training-free and plug-and-play method that dynamically adjusts attention logits based on the input in-context sequence. CAMA uses a two-stage modulation process that strengthens attention to semantically important tokens, especially visual ones. Across four LVLMs and seven benchmarks, CAMA consistently outperforms vanilla models and baselines, showing clear effectiveness and generalization. It can also activate the intended benefits of prompt engineering methods and remains robust across different sequence configurations. Therefore, CAMA opens up new directions for improving multimodal reasoning through a deeper understanding of attention dynamics.

Jie Cao, Tianwei Lin, Bo Yuan, Rolan Yan, Hongyang He, Wenqiao Zhang, Juncheng Li, Dongping Zhang, Siliang Tang, Yueting Zhuang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Pulmonary nodules are critical for early lung cancer diagnosis, but traditional CT imaging methods suffer from low detection rates and poor localization. Small nodule detection is challenging due to subtle differences in density and issues like occlusion. Existing methods such as FPN, with its fixed feature fusion and limited receptive field, struggle to effectively overcome these issues. To address these challenges, our paper proposed three key contributions: Firstly, we proposed MSDet, a multiscale attention and receptive field network for detecting tiny pulmonary nodules. Secondly, we proposed the extended receptive domain (ERD) strategy to capture richer contextual information and reduce false positives caused by nodule occlusion. We also proposed the position channel attention mechanism (PCAM) to optimize feature learning and reduce multiscale detection errors, and designed the tiny object detection block (TODB) to enhance the detection of tiny nodules. Experiments on the LUNA16 dataset show an 8.8% improvement in mAP over YOLOv8, achieving state-of-the-art performance. The code is available at https://github.com/CaiGuoHui123/MSDet.

The survival of numerous endangered wildlife species is increasingly jeopardized by drastic climate changes, ecological disturbances, and human activities, leading to rapid population declines. Identifying endangered and rare species using computer vision is an important task that can aid in biodiversity conservation. However, many rare species are recognizable only by specialized biologists, making the labeling process of images both challenging and costly. Moreover, even with labeled data, training models with imbalanced datasets, i.e., datasets with under-represented categories, often results in inherent learning biases. To address high labeling costs and challenges associated with sparse label learning, we propose a novel semi-supervised learning framework, 4S-Classifier, which comprises two key modules: Rare Species Bank (RSBank) and Attention-based Rare Species Embedding (RSEmbed). The RSBank module stores embeddings of rare species across multiple training epochs, using clustering, kernel density estimation, and confidence scores to enhance the learning of under-represented categories. The RSEmbed module acts as a fusion-based augmentation approach that employs embeddings to improve model performance on these sparse and rare species data. By integrating both modules, our framework achieves a classification accuracy of 88.37% on an endangered species dataset (iNaturalist) and 91.56% on a wildlife dataset (Wildlife Insights) with only 25% of the labeled data, demonstrating its outstanding performance. The code is available at https://github.com/SIPLab24/4S-Classifier

Modern text-to-image generation systems have enabled the creation of remarkably realistic and high-quality visuals, yet they often falter when handling the inherent ambiguities in user prompts. In this work, we present Twin-Co, a framework that leverages synchronized, co-adaptive dialogue to progressively refine image generation. Instead of a static generation process, Twin-Co employs a dynamic, iterative workflow where an intelligent dialogue agent continuously interacts with the user. Initially, a base image is generated from the user's prompt. Then, through a series of synchronized dialogue exchanges, the system adapts and optimizes the image according to evolving user feedback. The co-adaptive process allows the system to progressively narrow down ambiguities and better align with user intent. Experiments demonstrate that Twin-Co not only enhances user experience by reducing trial-and-error iterations but also improves the quality of the generated images, streamlining creative process across various applications.

We introduce TRiCo, a novel triadic game-theoretic co-training framework that rethinks the structure of semi-supervised learning by incorporating a teacher, two students, and an adversarial generator into a unified training paradigm. Unlike existing co-training or teacher-student approaches, TRiCo formulates SSL as a structured interaction among three roles: (i) two student classifiers trained on frozen, complementary representations, (ii) a meta-learned teacher that adaptively regulates pseudo-label selection and loss balancing via validation-based feedback, and (iii) a non-parametric generator that perturbs embeddings to uncover decision boundary weaknesses. Pseudo-labels are selected based on mutual information rather than confidence, providing a more robust measure of epistemic uncertainty. This triadic interaction is formalized as a Stackelberg game, where the teacher leads strategy optimization and students follow under adversarial perturbations. By addressing key limitations in existing SSL frameworks—such as static view interactions, unreliable pseudo-labels, and lack of hard sample modeling—TRiCo provides a principled and generalizable solution. Extensive experiments on CIFAR-10, SVHN, STL-10, and ImageNet demonstrate that TRiCo consistently achieves state-of-the-art performance in low-label regimes, while remaining architecture-agnostic and compatible with frozen vision backbones.

Although Multimodal Large Language Models (MLLMs) have demonstrated remarkable achievements in recent years, they remain vulnerable to adversarial examples that result in harmful responses. Existing attacks typically focus on optimizing adversarial perturbations for a certain multimodal image-prompt pair or fixed training dataset, which often leads to overfitting. Consequently, these perturbations fail to remain malicious once transferred to attack unseen image-prompt pairs, suffering from significant resource costs to cover the diverse multimodal inputs in complicated real-world scenarios. To alleviate this issue, this paper proposes a novel adversarial attack on MLLMs based on distribution approximation theory, which models the potential image-prompt input distribution and adds the same distribution-fitting adversarial perturbation on multimodal input pairs to achieve effective cross-image/prompt transfer attacks. Specifically, we exploit the Laplace approximation to model the Gaussian distribution of the image and prompt inputs for the MLLM, deriving an estimate of the mean and covariance parameters. By sampling from this approximated distribution with Monte Carlo mechanism, we efficiently optimize and fit a single input‑agnostic perturbation over diverse image‑prompt pairs, yielding strong universality and transferability. Extensive experiments are conducted to verify the strong adversarial capabilities of our proposed attack against prevalent MLLMs spanning a spectrum of images/prompts.

Power grid construction significantly enhances power grid management and risk control. Inconsistent operations at construction sites can jeopardize grid stability and crew safety. Traditional power lines are less favored due to mobility limitations, while batteries add burden and impracticality. To address this, a wireless power transfer and video monitoring system is developed using RF technology and Yolo V3 model. This enables continuous monitoring and employee safety analysis. The system's detection performance is optimized using HPSO, surpassing existing methods in accuracy and speed. It ensures real-time monitoring and improves the detection of potential risk sources, crucial for construction site safety.

With the improvement of resolution and precision of marine gravity anomaly data from satellite altimetry, the gravity matching navigation of underwater vehicles has been supported by background field data. To improve the accuracy, a multi-index model for analyzing the gravity background field characteristics is constructed, and an adaptive matching navigation algorithm is designed. First, the common single-index feature analysis methods are introduced, and their advantages and disadvantages are analyzed. The multi-index average statistical parameter (MASP) model applicable to the characterization of marine gravity anomaly background field is designed, and the reliability is verified by two satellite altimetric gravity anomaly datasets from the northwest Indian Ocean and the western Pacific Ocean. Then, an adaptive matching navigation algorithm is proposed by introducing the MASP model in the filtering process, which can adjust the filter structure in real time. Finally, the performance of the method is verified by about 30 h of field test. The matching results show that the proposed method can adjust the filter structure adaptively and has better comprehensive performance.

The background field of marine gravity anomaly from satellite altimetry is an important data support for gravity-aided navigation of underwater vehicles. To achieve long-endurance and high-precision navigation and positioning, an interpolation algorithm based on ensemble learning is proposed, and an adaptive matching navigation method is designed. First, the common interpolation theories of marine gravity anomaly background field are introduced, and the advantages and disadvantages of these theories are analyzed. The BP-Bagging ensemble learning algorithm for marine gravity anomaly background field interpolation is designed, and the reliability is verified by two satellite altimetric gravity anomaly datasets from the western Pacific Ocean. Then, a matching navigation algorithm based on the calculation method of gravity gradient deviation on survey line is proposed, which can evaluate the matching capability online according to the position of the underwater vehicle and adaptively adjusts the filter feedback coefficients. Finally, the performance of the method is verified by field tests. The matching results show that the proposed method improves navigation accuracy while enhancing the adaptive capability of the matching algorithm and providing better overall performance. Meanwhile, the gravity background field with high resolution yields better matching results.

A wide-band near-field source localization method is presented in this paper. Based on a pre-estimated source location, we form a focusing matrix which is able to compensate the wavefront distortion (with respect to far-field wavefront) due to near-field propagation and frequency dependent phase shift simultaneously. The focused covariance matrix has been proved to have a partially far-field narrow band structure, which allows us to estimate the bearings of the sources by the well studied far-field DOA estimators. The range estimation is then carried out via peak searching of the 1D MUSIC spectral with the estimated bearings. The performance of the algorithm is tested by simulations.