Jun Sun

Yige Li, Hanxun Huang, Yunhan Zhao et al.

Generative large language models (LLMs) have achieved state-of-the-art results on a wide range of tasks, yet they remain susceptible to backdoor attacks: carefully crafted triggers in the input can manipulate the model to produce adversary-specified outputs. While prior research has predominantly focused on backdoor risks in vision and classification settings, the vulnerability of LLMs in open-ended text generation remains underexplored. To fill this gap, we introduce BackdoorLLM (Our BackdoorLLM benchmark was awarded First Prize in the SafetyBench competition, https://www.mlsafety.org/safebench/winners, organized by the Center for AI Safety, https://safe.ai/.), the first comprehensive benchmark for systematically evaluating backdoor threats in text-generation LLMs. BackdoorLLM provides: (i) a unified repository of benchmarks with a standardized training and evaluation pipeline; (ii) a diverse suite of attack modalities, including data poisoning, weight poisoning, hidden-state manipulation, and chain-of-thought hijacking; (iii) over 200 experiments spanning 8 distinct attack strategies, 7 real-world scenarios, and 6 model architectures; (iv) key insights into the factors that govern backdoor effectiveness and failure modes in LLMs; and (v) a defense toolkit encompassing 7 representative mitigation techniques. Our code and datasets are available at https://github.com/bboylyg/BackdoorLLM. We will continuously incorporate emerging attack and defense methodologies to support the research in advancing the safety and reliability of LLMs.

Ming Cheng, Yiling Xu, Wang Shen et al.

High-speed, high-resolution stereoscopic (H2-Stereo) video allows us to perceive dynamic 3D content at fine granularity. The acquisition of H2-Stereo video, however, remains challenging with commodity cameras. Existing spatial super-resolution or temporal frame interpolation methods provide compromised solutions that lack temporal or spatial details, respectively. To alleviate this problem, we propose a dual camera system, in which one camera captures high-spatial-resolution low-frame-rate (HSR-LFR) videos with rich spatial details, and the other captures low-spatial-resolution high-frame-rate (LSR-HFR) videos with smooth temporal details. We then devise a Learned Information Fusion network (LIFnet) that exploits the cross-camera redundancies to enhance both camera views to high spatiotemporal resolution (HSTR) for reconstructing the H2-Stereo video effectively. We utilize a disparity network to transfer spatiotemporal information across views even in large disparity scenes, based on which, we propose disparity-guided flow-based warping for LSR-HFR view and complementary warping for HSR-LFR view. A multi-scale fusion method in feature domain is proposed to minimize occlusion-induced warping ghosts and holes in HSR-LFR view. The LIFnet is trained in an end-to-end manner using our collected high-quality Stereo Video dataset from YouTube. Extensive experiments demonstrate that our model outperforms existing state-of-the-art methods for both views on synthetic data and camera-captured real data with large disparity. Ablation studies explore various aspects, including spatiotemporal resolution, camera baseline, camera desynchronization, long/short exposures and applications, of our system to fully understand its capability for potential applications.

Simiao Zhang, Jitao Bai, Menghong Guan et al.

Machine learning is widely used to make decisions with societal impact such as bank loan approving, criminal sentencing, and resume filtering. How to ensure its fairness while maintaining utility is a challenging but crucial issue. Fairness is a complex and context-dependent concept with over 70 different measurement metrics. Since existing regulations are often vague in terms of which metric to use and different organizations may prefer different fairness metrics, it is important to have means of improving fairness comprehensively. Existing mitigation techniques often target at one specific fairness metric and have limitations in improving multiple notions of fairness simultaneously. In this work, we propose CFU (Comprehensive Fairness-Utility), a reinforcement learning-based framework, to efficiently improve the fairness-utility trade-off in machine learning classifiers. A comprehensive measurement that can simultaneously consider multiple fairness notions as well as utility is established, and new metrics are proposed based on an in-depth analysis of the relationship between different fairness metrics. The reward function of CFU is constructed with comprehensive measurement and new metrics. We conduct extensive experiments to evaluate CFU on 6 tasks, 3 machine learning models, and 15 fairness-utility measurements. The results demonstrate that CFU can improve the classifier on multiple fairness metrics without sacrificing its utility. It outperforms all state-of-the-art techniques and has witnessed a 37.5% improvement on average.

Nay Myat Min, Long H. Pham, Yige Li et al.

Large language models (LLMs) demonstrate remarkable performance across myriad language tasks, yet they remain vulnerable to backdoor attacks, where adversaries implant hidden triggers that systematically manipulate model outputs. Traditional defenses focus on explicit token-level anomalies and therefore overlook semantic backdoors-covert triggers embedded at the conceptual level (e.g., ideological stances or cultural references) that rely on meaning-based cues rather than lexical oddities. We first show, in a controlled finetuning setting, that such semantic backdoors can be implanted with only a small poisoned corpus, establishing their practical feasibility. We then formalize the notion of semantic backdoors in LLMs and introduce a black-box detection framework, RAVEN (short for "Response Anomaly Vigilance for uncovering semantic backdoors"), which combines semantic entropy with cross-model consistency analysis. The framework probes multiple models with structured topic-perspective prompts, clusters the sampled responses via bidirectional entailment, and flags anomalously uniform outputs; cross-model comparison isolates model-specific anomalies from corpus-wide biases. Empirical evaluations across diverse LLM families (GPT-4o, Llama, DeepSeek, Mistral) uncover previously undetected semantic backdoors, providing the first proof-of-concept evidence of these hidden vulnerabilities and underscoring the urgent need for concept-level auditing of deployed language models. We open-source our code and data at https://github.com/NayMyatMin/RAVEN.

Zi-Qi Li, Jun Sun, Xiao-Jun Wu et al.

Representation based classification methods have become a hot research topic during the past few years, and the two most prominent approaches are sparse representation based classification (SRC) and collaborative representation based classification (CRC). CRC reveals that it is the collaborative representation rather than the sparsity that makes SRC successful. Nevertheless, the dense representation of CRC may not be discriminative which will degrade its performance for classification tasks. To alleviate this problem to some extent, we propose a new method called sparse and collaborative-competitive representation based classification (SCCRC) for image classification. Firstly, the coefficients of the test sample are obtained by SRC and CCRC, respectively. Then the fused coefficient is derived by multiplying the coefficients of SRC and CCRC. Finally, the test sample is designated to the class that has the minimum residual. Experimental results on several benchmark databases demonstrate the efficacy of our proposed SCCRC. The source code of SCCRC is accessible at https://github.com/li-zi-qi/SCCRC.

Li Sun, Liuan Wang, Jun Sun et al.

Recent advancements in Multimodal Large Language Models (MLLMs) have significantly enhanced the comprehension of multimedia content, bringing together diverse modalities such as text, images, and videos. However, a critical challenge faced by these models, especially when processing video inputs, is the occurrence of hallucinations - erroneous perceptions or interpretations, particularly at the event level. This study introduces an innovative method to address event-level hallucinations in MLLMs, focusing on specific temporal understanding in video content. Our approach leverages a novel framework that extracts and utilizes event-specific information from both the event query and the provided video to refine MLLMs' response. We propose a unique mechanism that decomposes on-demand event queries into iconic actions. Subsequently, we employ models like CLIP and BLIP2 to predict specific timestamps for event occurrences. Our evaluation, conducted using the Charades-STA dataset, demonstrates a significant reduction in temporal hallucinations and an improvement in the quality of event-related responses. This research not only provides a new perspective in addressing a critical limitation of MLLMs but also contributes a quantitatively measurable method for evaluating MLLMs in the context of temporal-related questions.

Yan Luo, Chongyang Zhang, Muming Zhao et al.

Accurate pedestrian classification and localization have received considerable attention due to their wide applications such as security monitoring, autonomous driving, etc. Although pedestrian detectors have made great progress in recent years, the fixed Intersection over Union (IoU) based assignment-regression manner still limits their performance. Two main factors are responsible for this: 1) the IoU threshold faces a dilemma that a lower one will result in more false positives, while a higher one will filter out the matched positives; 2) the IoU-based GT-Proposal assignment suffers from the inconsistent supervision problem that spatially adjacent proposals with similar features are assigned to different ground-truth boxes, which means some very similar proposals may be forced to regress towards different targets, and thus confuses the bounding-box regression when predicting the location results. In this paper, we first put forward the question that \textbf{Regression Direction} would affect the performance for pedestrian detection. Consequently, we address the weakness of IoU by introducing one geometric sensitive search algorithm as a new assignment and regression metric. Different from the previous IoU-based \textbf{one-to-one} assignment manner of one proposal to one ground-truth box, the proposed method attempts to seek a reasonable matching between the sets of proposals and ground-truth boxes. Specifically, we boost the MR-FPPI under R$_{75}$ by 8.8\% on Citypersons dataset. Furthermore, by incorporating this method as a metric into the state-of-the-art pedestrian detectors, we show a consistent improvement.

Yan Luo, Chongyang Zhang, Muming Zhao et al.

Pedestrian detection benefits greatly from deep convolutional neural networks (CNNs). However, it is inherently hard for CNNs to handle situations in the presence of occlusion and scale variation. In this paper, we propose W$^3$Net, which attempts to address above challenges by decomposing the pedestrian detection task into \textbf{\textit{W}}here, \textbf{\textit{W}}hat and \textbf{\textit{W}}hether problem directing against pedestrian localization, scale prediction and classification correspondingly. Specifically, for a pedestrian instance, we formulate its feature by three steps. i) We generate a bird view map, which is naturally free from occlusion issues, and scan all points on it to look for suitable locations for each pedestrian instance. ii) Instead of utilizing pre-fixed anchors, we model the interdependency between depth and scale aiming at generating depth-guided scales at different locations for better matching instances of different sizes. iii) We learn a latent vector shared by both visual and corpus space, by which false positives with similar vertical structure but lacking human partial features would be filtered out. We achieve state-of-the-art results on widely used datasets (Citypersons and Caltech). In particular. when evaluating on heavy occlusion subset, our results reduce MR$^{-2}$ from 49.3$\%$ to 18.7$\%$ on Citypersons, and from 45.18$\%$ to 28.33$\%$ on Caltech.

Wei Shen, Ziqiang Shi, Jun Sun

Many recent few-shot learning methods concentrate on designing novel model architectures. In this paper, we instead show that with a simple backbone convolutional network we can even surpass state-of-the-art classification accuracy. The essential part that contributes to this superior performance is an adversarial feature learning strategy that improves the generalization capability of our model. In this work, adversarial features are those features that can cause the classifier uncertain about its prediction. In order to generate adversarial features, we firstly locate adversarial regions based on the derivative of the entropy with respect to an averaging mask. Then we use the adversarial region attention to aggregate the feature maps to obtain the adversarial features. In this way, we can explore and exploit the entire spatial area of the feature maps to mine more diverse discriminative knowledge. We perform extensive model evaluations and analyses on miniImageNet and tieredImageNet datasets demonstrating the effectiveness of the proposed method.

Jiangchao Yao, Ya Zhang, Ivor W. Tsang et al.

Learning with noisy labels, which aims to reduce expensive labors on accurate annotations, has become imperative in the Big Data era. Previous noise transition based method has achieved promising results and presented a theoretical guarantee on performance in the case of class-conditional noise. However, this type of approaches critically depend on an accurate pre-estimation of the noise transition, which is usually impractical. Subsequent improvement adapts the pre-estimation along with the training progress via a Softmax layer. However, the parameters in the Softmax layer are highly tweaked for the fragile performance due to the ill-posed stochastic approximation. To address these issues, we propose a Latent Class-Conditional Noise model (LCCN) that naturally embeds the noise transition under a Bayesian framework. By projecting the noise transition into a Dirichlet-distributed space, the learning is constrained on a simplex based on the whole dataset, instead of some ad-hoc parametric space. We then deduce a dynamic label regression method for LCCN to iteratively infer the latent labels, to stochastically train the classifier and to model the noise. Our approach safeguards the bounded update of the noise transition, which avoids previous arbitrarily tuning via a batch of samples. We further generalize LCCN for open-set noisy labels and the semi-supervised setting. We perform extensive experiments with the controllable noise data sets, CIFAR-10 and CIFAR-100, and the agnostic noise data sets, Clothing1M and WebVision17. The experimental results have demonstrated that the proposed model outperforms several state-of-the-art methods.

Jiangchao Yao, Jiajie Wang, Ivor Tsang et al.

There is an emerging trend to leverage noisy image datasets in many visual recognition tasks. However, the label noise among the datasets severely degenerates the \mbox{performance of deep} learning approaches. Recently, one mainstream is to introduce the latent label to handle label noise, which has shown promising improvement in the network designs. Nevertheless, the mismatch between latent labels and noisy labels still affects the predictions in such methods. To address this issue, we propose a quality embedding model, which explicitly introduces a quality variable to represent the trustworthiness of noisy labels. Our key idea is to identify the mismatch between the latent and noisy labels by embedding the quality variables into different subspaces, which effectively minimizes the noise effect. At the same time, the high-quality labels is still able to be applied for training. To instantiate the model, we further propose a Contrastive-Additive Noise network (CAN), which consists of two important layers: (1) the contrastive layer estimates the quality variable in the embedding space to reduce noise effect; and (2) the additive layer aggregates the prior predictions and noisy labels as the posterior to train the classifier. Moreover, to tackle the optimization difficulty, we deduce an SGD algorithm with the reparameterization tricks, which makes our method scalable to big data. We conduct the experimental evaluation of the proposed method over a range of noisy image datasets. Comprehensive results have demonstrated CAN outperforms the state-of-the-art deep learning approaches.

Song Wang, Jun Sun, Satoshi Naoi

The handwritten string recognition is still a challengeable task, though the powerful deep learning tools were introduced. In this paper, based on TAO-FCN, we proposed an end-to-end system for handwritten string recognition. Compared with the conventional methods, there is no preprocess nor manually designed rules employed. With enough labelled data, it is easy to apply the proposed method to different applications. Although the performance of the proposed method may not be comparable with the state-of-the-art approaches, it's usability and robustness are more meaningful for practical applications.

Xuefeng Xiao, Lianwen Jin, Yafeng Yang et al.

Like other problems in computer vision, offline handwritten Chinese character recognition (HCCR) has achieved impressive results using convolutional neural network (CNN)-based methods. However, larger and deeper networks are needed to deliver state-of-the-art results in this domain. Such networks intuitively appear to incur high computational cost, and require the storage of a large number of parameters, which renders them unfeasible for deployment in portable devices. To solve this problem, we propose a Global Supervised Low-rank Expansion (GSLRE) method and an Adaptive Drop-weight (ADW) technique to solve the problems of speed and storage capacity. We design a nine-layer CNN for HCCR consisting of 3,755 classes, and devise an algorithm that can reduce the networks computational cost by nine times and compress the network to 1/18 of the original size of the baseline model, with only a 0.21% drop in accuracy. In tests, the proposed algorithm surpassed the best single-network performance reported thus far in the literature while requiring only 2.3 MB for storage. Furthermore, when integrated with our effective forward implementation, the recognition of an offline character image took only 9.7 ms on a CPU. Compared with the state-of-the-art CNN model for HCCR, our approach is approximately 30 times faster, yet 10 times more cost efficient.

Song Wang, Li Sun, Wei Fan et al.

Nowadays the CNN is widely used in practical applications for image classification task. However the design of the CNN model is very professional work and which is very difficult for ordinary users. Besides, even for experts of CNN, to select an optimal model for specific task may still need a lot of time (to train many different models). In order to solve this problem, we proposed an automated CNN recommendation system for image classification task. Our system is able to evaluate the complexity of the classification task and the classification ability of the CNN model precisely. By using the evaluation results, the system can recommend the optimal CNN model and which can match the task perfectly. The recommendation process of the system is very fast since we don't need any model training. The experiment results proved that the evaluation methods are very accurate and reliable.

Xiaohang Ren, Kai Chen, Jun Sun

Candidate text region extraction plays a critical role in convolutional neural network (CNN) based text detection from natural images. In this paper, we propose a CNN based scene text detection algorithm with a new text region extractor. The so called candidate text region extractor I-MSER is based on Maximally Stable Extremal Region (MSER), which can improve the independency and completeness of the extracted candidate text regions. Design of I-MSER is motivated by the observation that text MSERs have high similarity and are close to each other. The independency of candidate text regions obtained by I-MSER is guaranteed by selecting the most representative regions from a MSER tree which is generated according to the spatial overlapping relationship among the MSERs. A multi-layer CNN model is trained to score the confidence value of the extracted regions extracted by the I-MSER for text detection. The new text detection algorithm based on I-MSER is evaluated with wide-used ICDAR 2011 and 2013 datasets and shows improved detection performance compared to the existing algorithms.

Xiaohang Ren, Kai Chen, Jun Sun

Scene text recognition plays an important role in many computer vision applications. The small size of available public available scene text datasets is the main challenge when training a text recognition CNN model. In this paper, we propose a CNN based Chinese text recognition algorithm. To enlarge the dataset for training the CNN model, we design a synthetic data engine for Chinese scene character generation, which generates representative character images according to the fonts use frequency of Chinese texts. As the Chinese text is more complex, the English text recognition CNN architecture is modified for Chinese text. To ensure the small size nature character dataset and the large size artificial character dataset are comparable in training, the CNN model are trained progressively. The proposed Chinese text recognition algorithm is evaluated with two Chinese text datasets. The algorithm achieves better recognize accuracy compared to the baseline methods.

Song Wang, Dongchun Ren, Li Chen et al.

Recently, the deep neural network (derived from the artificial neural network) has attracted many researchers' attention by its outstanding performance. However, since this network requires high-performance GPUs and large storage, it is very hard to use it on individual devices. In order to improve the deep neural network, many trials have been made by refining the network structure or training strategy. Unlike those trials, in this paper, we focused on the basic propagation function of the artificial neural network and proposed the binarized deep neural network. This network is a pure binary system, in which all the values and calculations are binarized. As a result, our network can save a lot of computational resource and storage. Therefore, it is possible to use it on various devices. Moreover, the experimental results proved the feasibility of the proposed network.

Shuangyong Song, Yao Meng, Zhongguang Zheng et al.

Our FRDC_QA team participated in the QA-Lab English subtask of the NTCIR-11. In this paper, we describe our system for solving real-world university entrance exam questions, which are related to world history. Wikipedia is used as the main external resource for our system. Since problems with choosing right/wrong sentence from multiple sentence choices account for about two-thirds of the total, we individually design a classification based model for solving this type of questions. For other types of questions, we also design some simple methods.