In this work, we introduce CAT-V (Caption Anything in Video), a training-free framework for fine-grained object-centric video captioning of user-selected instances. CAT-V combines (i) a SAMURAI-based Segmenter for precise object masks across frames, (ii) a TRACE-Uni Temporal Analyzer for event boundary detection and coarse event descriptions, and (iii) an InternVL-2.5 Captioner that, conditioned on spatiotemporal visual prompts and chain-of-thought (CoT) guidance, produces detailed, temporally coherent captions about object attributes, actions, states, interactions, and context. The system supports point, box, and region prompts and maintains temporal sensitivity by tracking object states across segments. In contrast to vanilla video captioning that is overly abstract and dense video captioning that is often terse, CAT-V enables object-level specificity with spatial accuracy and temporal coherence, without additional training data.

With the rapid growth of online video platforms and the escalating volume of video content, the need for proficient video understanding tools has increased significantly. Given the remarkable capabilities of large language models (LLMs) in language and multimodal tasks, this survey provides a detailed overview of recent advances in video understanding that harness the power of LLMs (Vid-LLMs). The emergent capabilities of Vid-LLMs are surprisingly advanced, particularly their ability for open-ended multi-granularity (abstract, temporal, and spatiotemporal) reasoning combined with common-sense knowledge, suggesting a promising path for future video understanding. We examine the unique characteristics and capabilities of Vid-LLMs, categorizing the approaches into three main types:Video Analyzer × LLM, Video Embedder × LLM, and (Analyzer + Embedder) × LLM. We identify five subtypes based on the functions of LLMs in Vid-LLMs:LLMas Summarizer,LLMas Manager,LLMas Text Decoder,LLMas Regressor, andLLMas Hidden Layer. This survey also presents a comprehensive study of the tasks, datasets, benchmarks, and evaluation methods for Vid-LLMs. Additionally, it explores the extensive applications of Vid-LLMs in various domains, highlighting their remarkable scalability and versatility in real-world video understanding challenges. Additionally, it summarizes the limitations of existing Vid-LLMs and outlines directions for future research. For more information, readers are encouraged to visit the repository at https://github.com/yunlong10/Awesome-LLMs-for-Video-Understanding.

The advancement of Multimodal Large Language Models (MLLMs) has enabled significant progress in multi-modal understanding, expanding their capacity to analyze video content. However, existing evaluation benchmarks for MLLMs primarily focus on abstract video comprehension, lacking a detailed assessment of their ability to understand video compositions, the nuanced interpretation of how visual elements combine and interact within highly compiled video contexts. We introduce VidComposition, a new benchmark specifically designed to evaluate the video composition understanding capabilities of MLLMs using carefully curated compiled videos and cinematic-level annotations. VidComposition includes 982 videos with 1706 multiple-choice questions, covering various compositional aspects such as camera movement, angle, shot size, narrative structure, character actions and emotions, etc. Our comprehensive evaluation of 33 open-source and proprietary MLLMs reveals a significant performance gap between human and model capabilities. This highlights the limitations of current MLLMs in understanding complex, compiled video compositions and offers insights into areas for further improvement. Our benchmark is publicly available at https://yunlong10.github.io/VidComposition/.

Foundation models (FMs) such as CLIP have demonstrated impressive zero-shot performance across various tasks by leveraging large-scale, unsupervised pre-training. However, they often inherit harmful or unwanted knowledge from noisy internet-sourced datasets, compromising their reliability in real-world applications. Existing model unlearning methods either rely on access to pre-trained datasets or focus on coarse-grained unlearning (e.g., entire classes), leaving a critical gap for fine-grained unlearning. In this paper, we address the challenging scenario of selectively forgetting specific portions of knowledge within a class—without access to pre-trained data—while preserving the model’s overall performance. We propose a novel three-stage approach that progressively unlearns targeted knowledge while mitigating over-forgetting. It consists of (1) a forgetting stage to fine-tune the CLIP on samples to be forgotten, (2) a reminding stage to restore performance on retained samples, and (3) a restoring stage to recover zero-shot capabilities using model souping. Additionally, we introduce knowledge distillation to handle the distribution disparity between forgetting/retaining samples and unseen pre-trained data. Extensive experiments on CIFAR-10, ImageNet-1K, and style datasets demonstrate that our approach effectively unlearns specific subgroups while maintaining strong zero-shot performance on semantically similar subgroups and other categories, significantly outperforming baseline unlearning methods, which lose effectiveness under the CLIP unlearning setting.

Given the limitations of backpropagation, perturbation-based gradient computation methods have recently gained focus for learning with only forward passes, also referred to as queries. Conventional forward learning consumes enormous queries on each data point for accurate gradient estimation through Monte Carlo sampling, which hinders the scalability of those algorithms. However, not all data points deserve equal queries for gradient estimation. In this paper, we study the problem of improving the forward learning efficiency from a novel perspective: how to reduce the gradient estimation variance with minimum cost? For this, we allocate the optimal number of queries within a set budget during training to balance estimation accuracy and computational efficiency. Specifically, with a simplified proxy objective and a reparameterization technique, we derive a novel plug-and-play query allocator with minimal parameters. Theoretical results are carried out to verify its optimality. We conduct extensive experiments for fine-tuning Vision Transformers on various datasets and further deploy the allocator to two black-box applications: prompt tuning and multimodal alignment for foundation models. All findings demonstrate that our proposed allocator significantly enhances the scalability of forward-learning algorithms, paving the way for real-world applications. The implementation is available at https://github.com/RTkenny/FLOPS-Forward-Learning-with-OPtimal-Sampling.

While audio-visual learning equips models with a richer understanding of the real world by leveraging multiple sensory modalities, this integration also introduces new vulnerabilities to adversarial attacks. In this paper, we present a comprehensive study of the adversarial robustness of audio-visual models, considering both temporal and modality-specific vulnerabilities. We propose two powerful adversarial attacks: 1) a temporal invariance attack that exploits the inherent temporal redundancy across consecutive time segments and 2) a modality misalignment attack that introduces incongruence between the audio and visual modalities. These attacks are designed to thoroughly assess the robustness of audio-visual models against diverse threats. Furthermore, to defend against such attacks, we introduce a novel audio-visual adversarial training framework. This framework addresses key challenges in vanilla adversarial training by incorporating efficient adversarial perturbation crafting tailored to multi-modal data and an adversarial curriculum strategy. Extensive experiments in the Kinetics-Sounds dataset demonstrate that our proposed temporal and modality-based attacks in degrading model performance can achieve state-of-the-art performance, while our adversarial training defense largely improves the adversarial robustness as well as the adversarial training efficiency.

Model ensemble adversarial attack has become a powerful method for generating transferable adversarial examples that can target even unknown models, but its theoretical foundation remains underexplored. To address this gap, we provide early theoretical insights that serve as a roadmap for advancing model ensemble adversarial attack. We first define transferability error to measure the error in adversarial transferability, alongside concepts of diversity and empirical model ensemble Rademacher complexity. We then decompose the transferability error into vulnerability, diversity, and a constant, which rigidly explains the origin of transferability error in model ensemble attack: the vulnerability of an adversarial example to ensemble components, and the diversity of ensemble components. Furthermore, we apply the latest mathematical tools in information theory to bound the transferability error using complexity and generalization terms, validating three practical guidelines for reducing transferability error: (1) incorporating more surrogate models, (2) increasing their diversity, and (3) reducing their complexity in cases of overfitting. Finally, extensive experiments with 54 models validate our theoretical framework, representing a significant step forward in understanding transferable model ensemble adversarial attacks.

Vision Transformers (ViTs) have demonstrated impressive performance across a range of applications, including many safety-critical tasks. Many previous studies have observed that adversarial examples crafted on ViTs exhibit higher transferability than those crafted on CNNs, indicating that ViTs contain structural characteristics favorable for transferable attacks. In this work, we take a further step to deeply investigate the role of computational redundancy brought by its unique characteristics in ViTs and its impact on adversarial transferability. Specifically, we identify two forms of redundancy, including the data-level and model-level, that can be harnessed to amplify attack effectiveness. Building on this insight, we design a suite of techniques, including attention sparsity manipulation, attention head permutation, clean token regularization, ghost MoE diversification, and learn to robustify before the attack. A dynamic online learning strategy is also proposed to fully leverage these operations to enhance the adversarial transferability. Extensive experiments on the ImageNet-1k dataset validate the effectiveness of our approach, showing that our methods significantly outperform existing baselines in both transferability and generality across diverse model architectures, including different variants of ViTs and mainstream Vision Large Language Models (VLLMs).

Understanding perspective is fundamental to human visual perception, yet the extent to which multimodal large language models (MLLMs) internalize perspective geometry remains unclear. We introduce MMPerspective, the first benchmark specifically designed to systematically evaluate MLLMs' understanding of perspective through 10 carefully crafted tasks across three complementary dimensions: Perspective Perception, Reasoning, and Robustness. Our benchmark comprises 2,711 real-world and synthetic image instances with 5,083 question-answer pairs that probe key capabilities, such as vanishing point perception and counting, perspective type reasoning, line relationship understanding in 3D space, invariance to perspective-preserving transformations, etc. Through a comprehensive evaluation of 43 state-of-the-art MLLMs, we uncover significant limitations: while models demonstrate competence on surface-level perceptual tasks, they struggle with compositional reasoning and maintaining spatial consistency under perturbations. Our analysis further reveals intriguing patterns between model architecture, scale, and perspective capabilities, highlighting both robustness bottlenecks and the benefits of chain-of-thought prompting. MMPerspective establishes a valuable testbed for diagnosing and advancing spatial understanding in vision-language systems. Resources are available at https://yunlong10.github.io/MMPerspective/

Abstract With the introduction of machine learning in high stakes decision-making, ensuring algorithmic fairness has become an increasingly important task. To this end, many mathematical definitions of fairness have been proposed, and a variety of optimisation techniques have been developed, all designed to maximise a given notion of fairness. Fair solutions, however, tend to rely on the quality of training data, and can be highly sensitive to noise. Recent studies have shown that robustness of many such fairness strategies—i.e., their ability to perform well on unseen data—is not a given and requires careful consideration. To address this challenge, we propose robustness ratio , which is a novel criterion to measure the robustness of diverse fairness optimisation strategies. We support our analysis with multiple extensive experiments on five benchmark fairness data sets, using three prominent fairness strategies, in view of four of the most popular definitions of fairness. Our experiments show that while fairness methods that rely on threshold optimisation (post-processing) mostly outperform other techniques, they are acutely sensitive to noise. This is in contrast to two other methods—correlation remover (pre-processing) and exponentiated gradient descent (in-processing)—which become increasingly fairer as the random noise injected into the data becomes larger. Our findings offer a comprehensive overview of fairness strategies that proves invaluable when tasked with choosing the most suitable method for the task at hand. To the best of our knowledge, we are the first to quantitatively evaluate the robustness of fairness optimisation strategies.

Artificial neural network (ANN) has been widely used in automation. However, the vulnerability of ANN under certain attacks poses a security threat to critical automation systems. Previous research has shown that adding noise to ANNs can enhance robustness. Nonetheless, striking a balance between robustness and task performance remains challenging, as excessive noise improves robustness but hampers performance, while low noise offers minor robustness improvement. In this work, we propose to learn the distribution of optimal injected noise, which improves the robustness as well as maintains the performance. Specifically, we compute the pathwise stochastic gradient estimate with respect to the standard deviation of the Gaussian noise added to each neuron of the ANN and optimize both the noise distribution and model parameters during training with negligible additional computational cost. In numerical experiments, our proposed method can achieve significant performance improvement on the robustness of several popular ANN structures under both black box and white box attacks. We also evaluate the proposed technique on two automation tasks: the classic reinforcement learning task of the cart pole game and a fault detection problem. Our results showed that the proposed technique outperforms a conventional neural network in terms of performance, robustness, and visual explainability.Note to Practitioners—The robustness of artificial neural networks is a critical consideration in automation applications as real-world data is often subject to unforeseen perturbations from the environment, potentially causing AI systems to behave unpredictably and unstably. For example, object detection is a widely employed AI technique in automation applications. However, current object detection systems are vulnerable to noise perturbation. Even small, imperceptible noise can lead the model to malfunction. Our work focuses on improving the robustness of neural networks. We propose a novel technique that can be added to any layer of existing neural networks to enhance robustness. Extensive experiments conducted in various scenarios have verified the effectiveness of the proposed method in enhancing both performance and robustness.

Machine learning models can perform well on indistribution data but often fail on biased subgroups that are underrepresented in the training data, hindering the robustness of models for reliable applications. Such subgroups are typically unknown due to the absence of subgroup labels. Discovering biased subgroups is the key to understanding models' failure modes and further improving models' robustness. Most previous works of subgroup discovery make an implicit assumption that models only underperform on a single biased subgroup, which does not hold on in-the-wild data where multiple biased subgroups exist. In this work, we propose Decomposition, Interpretation, and Mitigation (DIM), a novel method to address a more challenging but also more practical problem of discovering multiple biased subgroups in image classifiers. Our approach decomposes the image features into multiple components that represent multiple subgroups. This decomposition is achieved via a bilinear dimension reduction method, Partial Least Square (PLS), guided by useful supervision from the image classifier. We further interpret the semantic meaning of each subgroup component by generating natural language descriptions using vision-language foundation models. Finally, DIM mitigates multiple biased subgroups simultaneously via two strategies, including the data and model-centric strategies. Extensive experiments on CIFAR-100 and Breeds datasets demonstrate the effectiveness of DIM in discovering and mitigating multiple biased subgroups. Furthermore, DIM uncovers the failure modes of the classifier on Hard ImageNet, showcasing its broader applicability to understanding model bias in image classifiers. The code is available at https://github.com/ZhangAIPI/DIM.

Adversarial examples, crafted by adding perturbations imperceptible to humans, can deceive neural networks. Re-cent studies identify the adversarial transferability across various models, i.e., the cross-model attack ability of adversarial samples. To enhance such adversarial transferability, existing input transformation-based methods diversify input data with transformation augmentation. However, their effectiveness is limited by the finite number of available transformations. In our study, we introduce a novel approach named Learning to Transform (L2T). L2T increases the diversity of transformed images by selecting the optimal combination of operations from a pool of candidates, consequently im-proving adversarial transferability. We conceptualize the selection of optimal transformation combinations as a tra-jectory optimization problem and employ a reinforcement learning strategy to effectively solve the problem. Comprehensive experiments on the ImageNet dataset, as well as practical tests with Google Vision and GPT-4V, reveal that L2T surpasses current methodologies in enhancing adversarial transferability, thereby confirming its effectiveness and practical significance. The code is available at https://github.com/ZhangAIPI/TransferAttack.

While backpropagation (BP) is the mainstream approach for gradient computation in neural network training, its heavy reliance on the chain rule of differentiation constrains the designing flexibility of network architecture and training pipelines. We avoid the recursive computation in BP and develop a unified likelihood ratio (ULR) method for gradient estimation with only one forward propagation. Not only can ULR be extended to train a wide variety of neural network architectures, but the computation flow in BP can also be rearranged by ULR for better device adaptation. Moreover, we propose several variance reduction techniques to further accelerate the training process. Our experiments offer numerical results across diverse aspects, including various neural network training scenarios, computation flow rearrangement, and fine-tuning of pre-trained models. All findings demonstrate that ULR effectively enhances the flexibility of neural network training by permitting localized module training without compromising the global objective and significantly boosts the network robustness.

Accurate understanding of the biological functions of enzymes is vital for various tasks in both pathologies and industrial biotechnology. However, the existing methods are usually not fast enough and lack explanations on the prediction results, which severely limits their real-world applications. Following our previous work, DEEPre, we propose a new interpretable and fast version (ifDEEPre) by designing novel self-guided attention and incorporating biological knowledge learned via large protein language models to accurately predict the commission numbers of enzymes and confirm their functions. Novel self-guided attention is designed to optimize the unique contributions of representations, automatically detecting key protein motifs to provide meaningful interpretations. Representations learned from raw protein sequences are strictly screened to improve the running speed of the framework, 50 times faster than DEEPre while requiring 12.89 times smaller storage space. Large language modules are incorporated to learn physical properties from hundreds of millions of proteins, extending biological knowledge of the whole network. Extensive experiments indicate that ifDEEPre outperforms all the current methods, achieving more than 14.22% larger F1-score on the NEW dataset. Furthermore, the trained ifDEEPre models accurately capture multi-level protein biological patterns and infer evolutionary trends of enzymes by taking only raw sequences without label information. Meanwhile, ifDEEPre predicts the evolutionary relationships between different yeast sub-species, which are highly consistent with the ground truth. Case studies indicate that ifDEEPre can detect key amino acid motifs, which have important implications for designing novel enzymes. A web server running ifDEEPre is available at https://proj.cse.cuhk.edu.hk/aihlab/ifdeepre/ to provide convenient services to the public. Meanwhile, ifDEEPre is freely available on GitHub at https://github.com/ml4bio/ifDEEPre/.

Given the severe vulnerability of Deep Neural Networks (DNNs) against adversarial examples, there is an urgent need for an effective adversarial attack to identify the deficiencies of DNNs in security-sensitive applications. As one of the prevalent black-box adversarial attacks, the existing transfer-based attacks still cannot achieve comparable performance with the white-box attacks. Among these, input transformation based attacks have shown remarkable effectiveness in boosting transferability. In this work, we find that the existing input transformation based attacks transform the input image globally, resulting in limited diversity of the transformed images. We postulate that the more diverse transformed images result in better transferability. Thus, we investigate how to locally apply various transformations onto the input image to improve such diversity while preserving the structure of image. To this end, we propose a novel input transformation based attack, called Structure Invariant Transformation (SIA), which applies a random image transformation onto each image block to craft a set of diverse images for gradient calculation. Extensive experiments on the standard ImageNet dataset demonstrate that SIA exhibits much better transferability than the existing SOTA input transformation based attacks on CNN-based and transformer-based models, showing its generality and superiority in boosting transferability. Code is available at https://github.com/xiaosen-wang/SIT.

Google's quantum supremacy announcement has received broad questions from academia and industry due to the debatable estimate of 10,000 years' running time for the classical simulation task on the Summit supercomputer. Has quantum supremacy already come? Or will it come in one or two decades later? To avoid hasty advertisements of quantum supremacy by tech giants or quantum startups and eliminate the cost of dedicating a team to the classical simulation task, we advocate an open-source approach to maintain a trustable benchmark performance. In this paper, we take a reinforcement learning approach for the classical simulation of quantum circuits and demonstrate its great potential by reporting an estimated simulation time of less than 4 days, a speedup of 5.40x over the state-of-the-art method. Specifically, we formulate the classical simulation task as a tensor network contraction ordering problem using the K-spin Ising model and employ a novel Hamiltonina-based reinforcement learning algorithm. Then, we establish standard criteria to evaluate the performance of classical simulation of quantum circuits. We develop a dozen of massively parallel environments to simulate quantum circuits. We open-source our parallel gym environments and benchmarks. We hope the AI/ML community and quantum physics community will collaborate to maintain reference curves for validating an unequivocal first demonstration of empirical quantum supremacy.

Tensor learning is a powerful tool for big data analytics and machine learning, e.g., gene analysis and deep learning. However, tensor learning algorithms are compute-intensive since their time and space complexities grow exponentially with the order of tensors, which hinders their application. In this paper, we exploit the parallelism of tensor learning primitives using GPU tensor cores and develop high-performance tensor learning algorithms. First, we propose novel hardware-oriented optimization strategies for tensor learning primitives on GPU tensor cores. Second, for big data analytics, we employ the optimized tensor learning primitives to accelerate the CP tensor decomposition and then apply it for gene analysis. Third, we optimize the Tucker tensor decomposition and propose a novel Tucker tensor layer to compress deep neural networks. We employ natural gradients to train the neural networks, which only involve a forward pass without backpropagation and thus are suitable for GPU computations. Compared with TensorLab and TensorLy libraries on an A100 GPU, our third-order CP tensor decomposition achieves up to$16.32\times$and$32.25\times$speedups; and$6.09\times$and$6.72\times$speedups for our third-order Tucker tensor decomposition. The proposed fourth-order CP and Tucker tensor decompositions achieve up to$30.65\times$and$5.41\times$speedups over the TensorLab. Our CP tensor decomposition for gene analysis achieves up to$5.88\times$speedup over TensorLy. Compared with a conventional fully connected neural network, our Tucker tensor layer neural network achieves an accuracy of$97.9\%$, a speedup of$4.47\times$, and a compression ratio of$2.92$at the cost of$0.4\%$drop in accuracy.