Yiwei Sun

Yiwei Sun, Peiqi Jiang, Chuanbin Liu et al.

While the safety risks of image-based large language models have been extensively studied, their video-based counterparts (Video LLMs) remain critically under-examined. To systematically study this problem, we introduce \textbf{VideoSafetyBench (VSB-77k) - the first large-scale, culturally diverse benchmark for Video LLM safety}, which compromises 77,646 video-query pairs and spans 19 principal risk categories across 10 language communities. \textit{We reveal that integrating video modality degrades safety performance by an average of 42.3\%, exposing systemic risks in multimodal attack exploitation.} To address this vulnerability, we propose \textbf{VideoSafety-R1}, a dual-stage framework achieving unprecedented safety gains through two innovations: (1) Alarm Token-Guided Safety Fine-Tuning (AT-SFT) injects learnable alarm tokens into visual and textual sequences, enabling explicit harm perception across modalities via multitask objectives. (2) Then, Safety-Guided GRPO enhances defensive reasoning through dynamic policy optimization with rule-based rewards derived from dual-modality verification. These components synergize to shift safety alignment from passive harm recognition to active reasoning. The resulting framework achieves a 65.1\% improvement on VSB-Eval-HH, and improves by 59.1\%, 44.3\%, and 15.0\% on the image safety datasets MMBench, VLGuard, and FigStep, respectively. \textit{Our codes are available in the supplementary materials.} \textcolor{red}{Warning: This paper contains examples of harmful language and videos, and reader discretion is recommended.}

Bowei Pu, Chuanbin Liu, Yifan Ge et al.

Sufficient visual perception is the foundation of video reasoning. Nevertheless, existing Video Reasoning LLMs suffer from perception shortcuts, relying on a flawed single-step perception paradigm. This paradigm describes the video and then conducts reasoning, which runs the risk of insufficient evidence and emergent hallucinations. To address these issues, we introduce a new framework that integrates a loop-based paradigm with an anti-hallucination reward. First, to address the insufficient evidence, we introduce the Perception Loop Reasoning (PLR) paradigm. Instead of describing the video at once, each loop requires the model to describe a video segment with precise timestamps, analyze this segment, and decide the next action. Second, for the risk of hallucinations, the Factual-Aware Evaluator (FAE) evaluates each perception result as a reliable anti-hallucination reward. This reward encourages the model to provide sufficient and precise video evidence. Our FAE, which performs comparably to GPT-4o, is tuned on our AnetHallu-117K, a large-scale hallucination judgment preference dataset. Extensive experiments show that our Video-PLR achieves the state-of-the-art in both 3B and 7B parameter scales and has the best data efficiency. Our code, models, and datasets are released on: https://github.com/BoweiPu/VideoPLR.

Yiwei Sun, Zhihang Liu, Chuanbin Liu et al.

Recently, multimodal large language models have made significant advancements in video understanding tasks. However, their ability to understand unprocessed long videos is very limited, primarily due to the difficulty in supporting the enormous memory overhead. Although existing methods achieve a balance between memory and information by aggregating frames, they inevitably introduce the severe hallucination issue. To address this issue, this paper constructs a comprehensive hallucination mitigation pipeline based on existing MLLMs. Specifically, we use the CLIP Score to guide the frame sampling process with questions, selecting key frames relevant to the question. Then, We inject question information into the queries of the image Q-former to obtain more important visual features. Finally, during the answer generation stage, we utilize chain-of-thought and in-context learning techniques to explicitly control the generation of answers. It is worth mentioning that for the breakpoint mode, we found that image understanding models achieved better results than video understanding models. Therefore, we aggregated the answers from both types of models using a comparison mechanism. Ultimately, We achieved 84.2\% and 62.9\% for the global and breakpoint modes respectively on the MovieChat dataset, surpassing the official baseline model by 29.1\% and 24.1\%. Moreover the proposed method won the third place in the CVPR LOVEU 2024 Long-Term Video Question Answering Challenge. The code is avaiable at https://github.com/lntzm/CVPR24Track-LongVideo

Junjie Liang, Dongkuan Xu, Yiwei Sun et al.

We consider the problem of learning predictive models from longitudinal data, consisting of irregularly repeated, sparse observations from a set of individuals over time. Such data often exhibit {\em longitudinal correlation} (LC) (correlations among observations for each individual over time), {\em cluster correlation} (CC) (correlations among individuals that have similar characteristics), or both. These correlations are often accounted for using {\em mixed effects models} that include {\em fixed effects} and {\em random effects}, where the fixed effects capture the regression parameters that are shared by all individuals, whereas random effects capture those parameters that vary across individuals. However, the current state-of-the-art methods are unable to select the most predictive fixed effects and random effects from a large number of variables, while accounting for complex correlation structure in the data and non-linear interactions among the variables. We propose Longitudinal Multi-Level Factorization Machine (LMLFM), to the best of our knowledge, the first model to address these challenges in learning predictive models from longitudinal data. We establish the convergence properties, and analyze the computational complexity, of LMLFM. We present results of experiments with both simulated and real-world longitudinal data which show that LMLFM outperforms the state-of-the-art methods in terms of predictive accuracy, variable selection ability, and scalability to data with large number of variables. The code and supplemental material is available at \url{https://github.com/junjieliang672/LMLFM}.

Xianfeng Tang, Yandong Li, Yiwei Sun et al.

Graph neural networks (GNNs) are widely used in many applications. However, their robustness against adversarial attacks is criticized. Prior studies show that using unnoticeable modifications on graph topology or nodal features can significantly reduce the performances of GNNs. It is very challenging to design robust graph neural networks against poisoning attack and several efforts have been taken. Existing work aims at reducing the negative impact from adversarial edges only with the poisoned graph, which is sub-optimal since they fail to discriminate adversarial edges from normal ones. On the other hand, clean graphs from similar domains as the target poisoned graph are usually available in the real world. By perturbing these clean graphs, we create supervised knowledge to train the ability to detect adversarial edges so that the robustness of GNNs is elevated. However, such potential for clean graphs is neglected by existing work. To this end, we investigate a novel problem of improving the robustness of GNNs against poisoning attacks by exploring clean graphs. Specifically, we propose PA-GNN, which relies on a penalized aggregation mechanism that directly restrict the negative impact of adversarial edges by assigning them lower attention coefficients. To optimize PA-GNN for a poisoned graph, we design a meta-optimization algorithm that trains PA-GNN to penalize perturbations using clean graphs and their adversarial counterparts, and transfers such ability to improve the robustness of PA-GNN on the poisoned graph. Experimental results on four real-world datasets demonstrate the robustness of PA-GNN against poisoning attacks on graphs. Code and data are available here: https://github.com/tangxianfeng/PA-GNN.

Yiwei Sun

Educators teaching entry-level university engineering modules face the challenge of identifying which topics students find most difficult and how to support diverse student needs effectively. This study demonstrates a rigorous yet interpretable statistical approach -- hierarchical Bayesian modeling -- that leverages detailed student response data to quantify both skill difficulty and individual student abilities. Using a large-scale dataset from an undergraduate Statics course, we identified clear patterns of skill mastery and uncovered distinct student subgroups based on their learning trajectories. Our analysis reveals that certain concepts consistently present challenges, requiring targeted instructional support, while others are readily mastered and may benefit from enrichment activities. Importantly, the hierarchical Bayesian method provides educators with intuitive, reliable metrics without sacrificing predictive accuracy. This approach allows for data-informed decisions, enabling personalized teaching strategies to improve student engagement and success. By combining robust statistical methods with clear interpretability, this study equips educators with actionable insights to better support diverse learner populations.

Enyan Dai, Kai Shu, Yiwei Sun et al.

The recent advanced deep learning techniques have shown the promising results in various domains such as computer vision and natural language processing. The success of deep neural networks in supervised learning heavily relies on a large amount of labeled data. However, obtaining labeled data with target labels is often challenging due to various reasons such as cost of labeling and privacy issues, which challenges existing deep models. In spite of that, it is relatively easy to obtain data with \textit{inexact supervision}, i.e., having labels/tags related to the target task. For example, social media platforms are overwhelmed with billions of posts and images with self-customized tags, which are not the exact labels for target classification tasks but are usually related to the target labels. It is promising to leverage these tags (inexact supervision) and their relations with target classes to generate labeled data to facilitate the downstream classification tasks. However, the work on this is rather limited. Therefore, we study a novel problem of labeled data generation with inexact supervision. We propose a novel generative framework named as ADDES which can synthesize high-quality labeled data for target classification tasks by learning from data with inexact supervision and the relations between inexact supervision and target classes. Experimental results on image and text datasets demonstrate the effectiveness of the proposed ADDES for generating realistic labeled data from inexact supervision to facilitate the target classification task.

Tsung-Yu Hsieh, Suhang Wang, Yiwei Sun et al.

Time series data is prevalent in a wide variety of real-world applications and it calls for trustworthy and explainable models for people to understand and fully trust decisions made by AI solutions. We consider the problem of building explainable classifiers from multi-variate time series data. A key criterion to understand such predictive models involves elucidating and quantifying the contribution of time varying input variables to the classification. Hence, we introduce a novel, modular, convolution-based feature extraction and attention mechanism that simultaneously identifies the variables as well as time intervals which determine the classifier output. We present results of extensive experiments with several benchmark data sets that show that the proposed method outperforms the state-of-the-art baseline methods on multi-variate time series classification task. The results of our case studies demonstrate that the variables and time intervals identified by the proposed method make sense relative to available domain knowledge.

Xianfeng Tang, Huaxiu Yao, Yiwei Sun et al.

Graph Convolutional Networks (GCNs) show promising results for semi-supervised learning tasks on graphs, thus become favorable comparing with other approaches. Despite the remarkable success of GCNs, it is difficult to train GCNs with insufficient supervision. When labeled data are limited, the performance of GCNs becomes unsatisfying for low-degree nodes. While some prior work analyze successes and failures of GCNs on the entire model level, profiling GCNs on individual node level is still underexplored. In this paper, we analyze GCNs in regard to the node degree distribution. From empirical observation to theoretical proof, we confirm that GCNs are biased towards nodes with larger degrees with higher accuracy on them, even if high-degree nodes are underrepresented in most graphs. We further develop a novel Self-Supervised-Learning Degree-Specific GCN (SL-DSGC) that mitigate the degree-related biases of GCNs from model and data aspects. Firstly, we propose a degree-specific GCN layer that captures both discrepancies and similarities of nodes with different degrees, which reduces the inner model-aspect biases of GCNs caused by sharing the same parameters with all nodes. Secondly, we design a self-supervised-learning algorithm that creates pseudo labels with uncertainty scores on unlabeled nodes with a Bayesian neural network. Pseudo labels increase the chance of connecting to labeled neighbors for low-degree nodes, thus reducing the biases of GCNs from the data perspective. Uncertainty scores are further exploited to weight pseudo labels dynamically in the stochastic gradient descent for SL-DSGC. Experiments on three benchmark datasets show SL-DSGC not only outperforms state-of-the-art self-training/self-supervised-learning GCN methods, but also improves GCN accuracy dramatically for low-degree nodes.

Yiwei Sun, Shabnam Ghaffarzadegan

Recent advancements in audio event classification often ignore the structure and relation between the label classes available as prior information. This structure can be defined by ontology and augmented in the classifier as a form of domain knowledge. To capture such dependencies between the labels, we propose an ontology-aware neural network containing two components: feed-forward ontology layers and graph convolutional networks (GCN). The feed-forward ontology layers capture the intra-dependencies of labels between different levels of ontology. On the other hand, GCN mainly models inter-dependency structure of labels within an ontology level. The framework is evaluated on two benchmark datasets for single-label and multi-label audio event classification tasks. The results demonstrate the proposed solutions efficacy to capture and explore the ontology relations and improve the classification performance.

Enyan Dai, Yiwei Sun, Suhang Wang

Nowadays, Internet is a primary source of attaining health information. Massive fake health news which is spreading over the Internet, has become a severe threat to public health. Numerous studies and research works have been done in fake news detection domain, however, few of them are designed to cope with the challenges in health news. For instance, the development of explainable is required for fake health news detection. To mitigate these problems, we construct a comprehensive repository, FakeHealth, which includes news contents with rich features, news reviews with detailed explanations, social engagements and a user-user social network. Moreover, exploratory analyses are conducted to understand the characteristics of the datasets, analyze useful patterns and validate the quality of the datasets for health fake news detection. We also discuss the novel and potential future research directions for the health fake news detection.

Xianfeng Tang, Huaxiu Yao, Yiwei Sun et al.

Multivariate time series (MTS) forecasting is widely used in various domains, such as meteorology and traffic. Due to limitations on data collection, transmission, and storage, real-world MTS data usually contains missing values, making it infeasible to apply existing MTS forecasting models such as linear regression and recurrent neural networks. Though many efforts have been devoted to this problem, most of them solely rely on local dependencies for imputing missing values, which ignores global temporal dynamics. Local dependencies/patterns would become less useful when the missing ratio is high, or the data have consecutive missing values; while exploring global patterns can alleviate such problems. Thus, jointly modeling local and global temporal dynamics is very promising for MTS forecasting with missing values. However, work in this direction is rather limited. Therefore, we study a novel problem of MTS forecasting with missing values by jointly exploring local and global temporal dynamics. We propose a new framework LGnet, which leverages memory network to explore global patterns given estimations from local perspectives. We further introduce adversarial training to enhance the modeling of global temporal distribution. Experimental results on real-world datasets show the effectiveness of LGnet for MTS forecasting with missing values and its robustness under various missing ratios.

Yiwei Sun, Suhang Wang, Xianfeng Tang et al.

Real-world graph applications, such as advertisements and product recommendations make profits based on accurately classify the label of the nodes. However, in such scenarios, there are high incentives for the adversaries to attack such graph to reduce the node classification performance. Previous work on graph adversarial attacks focus on modifying existing graph structures, which is infeasible in most real-world applications. In contrast, it is more practical to inject adversarial nodes into existing graphs, which can also potentially reduce the performance of the classifier. In this paper, we study the novel node injection poisoning attacks problem which aims to poison the graph. We describe a reinforcement learning based method, namely NIPA, to sequentially modify the adversarial information of the injected nodes. We report the results of experiments using several benchmark data sets that show the superior performance of the proposed method NIPA, relative to the existing state-of-the-art methods.

Yiwei Sun, Suhang Wang, Tsung-Yu Hsieh et al.

Data from many real-world applications can be naturally represented by multi-view networks where the different views encode different types of relationships (e.g., friendship, shared interests in music, etc.) between real-world individuals or entities. There is an urgent need for methods to obtain low-dimensional, information preserving and typically nonlinear embeddings of such multi-view networks. However, most of the work on multi-view learning focuses on data that lack a network structure, and most of the work on network embeddings has focused primarily on single-view networks. Against this background, we consider the multi-view network representation learning problem, i.e., the problem of constructing low-dimensional information preserving embeddings of multi-view networks. Specifically, we investigate a novel Generative Adversarial Network (GAN) framework for Multi-View Network Embedding, namely MEGAN, aimed at preserving the information from the individual network views, while accounting for connectivity across (and hence complementarity of and correlations between) different views. The results of our experiments on two real-world multi-view data sets show that the embeddings obtained using MEGAN outperform the state-of-the-art methods on node classification, link prediction and visualization tasks.

Yimin Zhou, Yiwei Sun, Vasant Honavar

We explore the use of a knowledge graphs, that capture general or commonsense knowledge, to augment the information extracted from images by the state-of-the-art methods for image captioning. The results of our experiments, on several benchmark data sets such as MS COCO, as measured by CIDEr-D, a performance metric for image captioning, show that the variants of the state-of-the-art methods for image captioning that make use of the information extracted from knowledge graphs can substantially outperform those that rely solely on the information extracted from images.

Yiwei Sun, Ngot Bui, Tsung-Yu Hsieh et al.

Real-world social networks and digital platforms are comprised of individuals (nodes) that are linked to other individuals or entities through multiple types of relationships (links). Sub-networks of such a network based on each type of link correspond to distinct views of the underlying network. In real-world applications, each node is typically linked to only a small subset of other nodes. Hence, practical approaches to problems such as node labeling have to cope with the resulting sparse networks. While low-dimensional network embeddings offer a promising approach to this problem, most of the current network embedding methods focus primarily on single view networks. We introduce a novel multi-view network embedding (MVNE) algorithm for constructing low-dimensional node embeddings from multi-view networks. MVNE adapts and extends an approach to single view network embedding (SVNE) using graph factorization clustering (GFC) to the multi-view setting using an objective function that maximizes the agreement between views based on both the local and global structure of the underlying multi-view graph. Our experiments with several benchmark real-world single view networks show that GFC-based SVNE yields network embeddings that are competitive with or superior to those produced by the state-of-the-art single view network embedding methods when the embeddings are used for labeling unlabeled nodes in the networks. Our experiments with several multi-view networks show that MVNE substantially outperforms the single view methods on integrated view and the state-of-the-art multi-view methods. We further show that even when the goal is to predict labels of nodes within a single target view, MVNE outperforms its single-view counterpart suggesting that the MVNE is able to extract the information that is useful for labeling nodes in the target view from the all of the views.

Tsung-Yu Hsieh, Yasser EL-Manzalawy, Yiwei Sun et al.

Many machine learning, statistical inference, and portfolio optimization problems require minimization of a composition of expected value functions (CEVF). Of particular interest is the finite-sum versions of such compositional optimization problems (FS-CEVF). Compositional stochastic variance reduced gradient (C-SVRG) methods that combine stochastic compositional gradient descent (SCGD) and stochastic variance reduced gradient descent (SVRG) methods are the state-of-the-art methods for FS-CEVF problems. We introduce compositional stochastic average gradient descent (C-SAG) a novel extension of the stochastic average gradient method (SAG) to minimize composition of finite-sum functions. C-SAG, like SAG, estimates gradient by incorporating memory of previous gradient information. We present theoretical analyses of C-SAG which show that C-SAG, like SAG, and C-SVRG, achieves a linear convergence rate when the objective function is strongly convex; However, C-CAG achieves lower oracle query complexity per iteration than C-SVRG. Finally, we present results of experiments showing that C-SAG converges substantially faster than full gradient (FG), as well as C-SVRG.