Search

Recent years have witnessed increasing interest in adversarial attacks on images, while adversarial video attacks have seldom been explored. In this paper, we propose a sparse adversarial attack strategy on videos (DeepSAVA). Our model aims to add a small human-imperceptible perturbation to the key frame of the input video to fool the classifiers. To carry out an effective attack that mirrors real-world scenarios, our algorithm integrates spatial transformation perturbations into the frame. Instead of using the l p norm to gauge the disparity between the perturbed frame and the original frame, we employ the structural similarity index (SSIM), which has been established as a more suitable metric for quantifying image alterations resulting from spatial perturbations. We employ a unified optimisation framework to combine spatial transformation with additive perturbation, thereby attaining a more potent attack. We design an effective and novel optimisation scheme that alternatively utilises Bayesian Optimisation (BO) to identify the most critical frame in a video and stochastic gradient descent (SGD) based optimisation to produce both additive and spatial-transformed perturbations. Doing so enables DeepSAVA to perform a very sparse attack on videos for maintaining human imperceptibility while still achieving state-of-the-art performance in terms of both attack success rate and adversarial transferability. Furthermore, built upon the strong perturbations produced by DeepSAVA, we design a novel adversarial training framework to improve the robustness of video classification models. Our intensive experiments on various types of deep neural networks and video datasets confirm the superiority of DeepSAVA in terms of attacking performance and efficiency. When compared to the baseline techniques, DeepSAVA exhibits the highest level of performance in generating adversarial videos for three distinct video classifiers. Remarkably, it achieves an impressive fooling rate ranging from 99.5% to 100% for the I3D model, with the perturbation of just a single frame. Additionally, DeepSAVA demonstrates favourable transferability across various time series models. The proposed adversarial training strategy is also empirically demonstrated with better performance on training robust video classifiers compared with the state-of-the-art adversarial training with projected gradient descent (PGD) adversary. • Sparse attacks on video models: perturb fewer frames to gain high fooling rate. • Combining additive and spatial perturbations to enhance attacking performance. • Using SSIM instead of l p -norm to maintain the human perception. • Applying Bayesian Optimisation to identify the most critical frame to perturb. • A new adversarial training method based on combination of diverse perturbations. [ABSTRACT FROM AUTHOR]

Deep neural networks are highly vulnerable to adversarial samples. Most existing adversarial defense methods do not consider the distribution of adversarial samples. We argue that very few adversarial samples in the natural sample set prevent the deep neural networks from learning a complete and effective representation of the adversarial samples. This causes the spatial structures between the natural and the adversarial samples to be vastly different from that of the input space, thus making the models vulnerable to adversarial attacks. Based on this viewpoint, this paper proposes an effective deep-learning adversarial defense method, which incorporates information about the spatial structures of the natural and the adversarial samples in the embedding space during the training process. This proposed approach improves the deep learning model's generalization to new adversarial samples and achieves the purpose of defending against adversarial attacks. Four deep neural networks with different scales are used and experimentally verified on four typical publicly available image data. The experimental results show that our method effectively improves the defense ability of deep learning models against adversarial attacks. • Development of an effective adversarial training method using spatial structure constraints in embedding space. • Spatial structure information of the natural and adversarial samples can help improve the adversarial robustness of DNNs. • The spatial structure constrain can be extended to adversarial training of semi-supervised and unsupervised learning. [ABSTRACT FROM AUTHOR]

In the context of Earth observation, change detection boils down to comparing images acquired at different times by sensors of possibly different spatial and/or spectral resolutions or different modalities (e.g., optical or radar). Even when considering only optical images, this task has proven to be challenging as soon as the sensors differ by their spatial and/or spectral resolutions. This paper proposes a novel unsupervised change detection method dedicated to images acquired by such so-called heterogeneous optical sensors. It capitalizes on recent advances which formulate the change detection task into a robust fusion framework. Adopting this formulation, the work reported in this paper shows that any off-the-shelf network trained beforehand to fuse optical images of different spatial and/or spectral resolutions can be easily complemented with a network of the same architecture and embedded into an adversarial framework to perform change detection. A comparison with state-of-the-art change detection methods demonstrates the versatility and the effectiveness of the proposed approach. • Detecting changes between images acquired by heterogeneous sensors is formulated as a robust fusion problem. • The method is able to handle multi-band images with possibly different spatial and/or spectral resolutions. • The method is unsupervised, i.e., it does not need a train set composed of image pairs with associated binary change maps. • Any pretrained deep network designed to fuse heterogeneous images can be reused to perform change detection. • When compared to conventional approaches, the strategy appears to be more flexible while competing favorably. [ABSTRACT FROM AUTHOR]

Adversarial robustness is one of the most challenging problems in Deep Learning and Computer Vision research. State-of-the-art techniques to enforce robustness are based on Adversarial Training, a computationally costly optimization procedure. For this reason, many alternative solutions have been proposed, but none proved effective under stronger or adaptive attacks. This paper presents Image-Graph Extractor (IGE), a new image filtering scheme that extracts the fundamental nodes of an image and their connections through a graph structure. By utilizing the IGE representation, we have developed a new defense technique, Filtering as a Defense, which prevents attackers from creating malicious patterns that can deceive image classifiers. Moreover, we show that data augmentation with filtered images effectively improves the model's robustness to data corruptions. We validate our techniques on Convolutional Neural Networks on CIFAR-10, CIFAR-100, and ImageNet. [ABSTRACT FROM AUTHOR]

Link prediction is a fundamental problem in complex network analysis, aimed at predicting missing or forthcoming connections. Recent research investigates the potential of Nonnegative Matrix Factorization (NMF) models in reconstructing sparse networks and using deep NMF to uncover the hierarchical structure. Deep models have demonstrated remarkable performance in various domains, but they are susceptible to overfitting, especially on the limited training data. This paper proposes a novel Link Prediction using Adversarial Deep NMF (LPADNMF) to enhance the generalization of network reconstruction in sparse graphs. The main contribution is the introduction of an adversarial training that incorporates a bounded attack on the input, leveraging the L 2 , 1 norm to generate diverse perturbations. This adversarial training aims to improve the model's robustness and prevent overfitting, particularly in scenarios with limited training data. Additionally, the proposed method incorporates first and second-order affinities as input to capture higher-order dependencies and encourage the extraction of informative features from the network structure. To further mitigate overfitting, a smooth L 2 regularization is applied to the model parameters. To optimize the proposed model effectively, we utilize a majorization-minimization algorithm that efficiently updates the perturbation and latent factors in an iterative manner. Our findings demonstrate that the proposed model not only has the ability to uncover and learn complex structures but also possesses generalization capabilities. This method demonstrates superior performance compared to state-of-the-art methods across eight networks. In comparison to the best-performing approach, LPADNMF exhibited improvements of 2.36% in AUC, 4.67% in Precision, 2.21% in Recall, and 3.73% in F-Measure. [ABSTRACT FROM AUTHOR]

Breast cancer is the most commonly diagnosed cancer worldwide, and early detection is essential for reducing mortality rates. Digital mammography is currently the best standard for early detection, as it can assist physicians in treating the disease. However, inaccurate diagnoses from mammography are common and can lead to patients undergoing unnecessary tests and treatments. To address this challenge, deep-learning techniques have shown promising results in improving the accuracy and reliability of breast cancer detection. However, existing methods face two primary challenges: the lack of the annotated data, and the inability to adapt to new data domains. In this paper, we propose SelfAdaptNet to address these issues. Specifically, SelfAdaptNet employs self-supervised learning techniques, such as Bootstrap Your Own Latent (BYOL) and Simple Framework for Learning of Visual Representations (SimCLR), to tackle the problem of limited annotated data. Additionally, the adversarial technique is used to address the problem of domain shift. By successfully reducing domain disparities, this strategy enhances the model's adaptability and robustness across a variety of clinical scenarios. Overall, our contributions offer a more effective and flexible approach for early breast cancer detection, and experimental results demonstrate that SelfAdaptNet can produce promising results as compared with other methods. [ABSTRACT FROM AUTHOR]

The performance of object detection networks is often limited by the depth of the feature extraction network. Increasing network parameters may yield limited improvements in the detection system's performance. Additional careful designs of network details are necessary, but they can significantly increase training difficulty. This paper introduces a novel object detection method that utilizes generative adversarial training. Our approach focuses on minimizing the EM distance (Wasserstein distance) of the feature distribution as the primary training objective. We enhance the image features so that training of GAN (Generative Adversarial Networks) yields a feature distribution that exceeds that of the original dataset, obtaining a better feature extraction network. A new loss function is also added to the adversarial training process to ensure stable improvement of the detector. A comparative experiment conducted with the original CenterNet network on MS COCO (Microsoft Common Objects in COntext) 2017 reveals that the generative adversarial training method significantly improves the average precision for most of the examined backbone networks. Among the four backbone networks employed in the experiments, the mean improvement in network AP (Average Precision) values ranged from 0.3 to 0.9, demonstrating their success with minimal training efforts. Moreover, none of the four backbone networks experienced an increase in network parameters during inference. Experimental results indicate that the proposed architecture effectively enhances the network's feature extraction capability without compromising speed during inference. [ABSTRACT FROM AUTHOR]

Adversarial learning has become an effective paradigm for learning transferable features in domain adaptation. However, many previous adversarial domain adaptation methods inevitably damage the discriminative information contained in transferable features, which limits the potential of adversarial learning. In this paper, we explore the reason for this phenomenon and find that the model pays more attention to the alignment of feature norms than the learning of domain-invariant features during adversarial adaptation. Moreover, we observe that the feature norms contain some crucial category information, which is ignored in previous studies. To achieve better adversarial adaptation, we propose two novel feature norms alignment strategies: Histogram-guided Norms Alignment (HNA) and Transport-guided Norms Alignment (TNA). Both strategies model the feature norms from the distribution perspective, which not only facilitates the reduction of the norms discrepancy but also makes full use of discriminative information contained in the norms. Extensive experiments demonstrate that progressively aligning the feature norms distributions of two domains can effectively promote the capture of semantically rich shared features and significantly boost the model's transfer performance. We hope our findings can shed some light on future research of adversarial domain adaptation. [ABSTRACT FROM AUTHOR]

Adversarial training has been considered to be one of the most effective strategies to defend against adversarial attacks. Most existing adversarial training methods have shown a trade-off between training cost and robustness. This paper explores a new optimization direction from training data to reduce the computational cost of adversarial training without scarifying robustness. First, we show that some adversarial examples are less important, meaning that removing them does not hurt the robustness. Second, we propose a method to identify insignificant adversarial examples at a minimal cost. Third, we demonstrate that our approach can be integrated with other adversarial training frameworks with few modifications. The experimental results show that combined with previous works, our approach not only reduces about 20% of computational cost on the CIFAR10 and CIFAR100 datasets but also improves about 1.5% natural accuracy. With less computational cost, it achieves 58.22%, 30.68%, and 41.92% robust accuracy on CIFAR10, CIFAR100, and ImageNet datasets respectively, which are higher than those of the original methods. • Analyze the effectiveness of adversarial examples for adversarial training. • Design a method to remove inefficient adversarial examples but maintain the accuracy. • Design an algorithm to detect inefficient adversarial examples without extra cost. • Conduct experiments to demonstrate the compatibility with other frameworks. [ABSTRACT FROM AUTHOR]

Multivariate time series (MTS), whose patterns change dynamically, often have complex temporal and dimensional dependence. Most existing reconstruction-based MTS anomaly detection methods only learn the point-wise information while ignoring the overall trend of time series, resulting in their incompetence in extracting high-level semantic information. Although a few contrastive learning-based approaches have been proposed recently to solve this problem, they forcibly increase the difference between the features of normal data, leading to the loss of useful information. This paper proposes an adversarial contrastive autoencoder (ACAE) for MTS anomaly detection. ACAE conducts feature combination and decomposition as the contrastive learning proxy task, which introduces adversarial training to learn the transformation-invariant representation of data, achieving a robust representation of MTS. Firstly, ACAE constructs positive and negative sample pairs through the multi-scale timestamp mask and random sampling. Secondly, the features of the original samples are combined with those of the positive and negative samples to generate the positive and negative composite features. Finally, ACAE trains the encoder and discriminator to decompose the negative composite features cooperatively to decrease the similarity between the features of negative pairs. In contrast, it adversarially decomposes the positive composite features to increase the similarity between the features of positive pairs. Experimental results show that ACAE outperforms 14 state-of-the-art baselines on five real-world datasets from different fields. • Adversarial contrastive autoencoder for multivariate time series anomaly detection. • Novel proxy task for contrastive learning by feature combination and decomposition. • Multi-scale timestamp mask for time series data augmentation. [ABSTRACT FROM AUTHOR]

Rib cross-sectional shapes (characterized by the outer contour and cortical bone thickness) affect the rib mechanical response under impact loading, thereby influence the rib injury pattern and risk. A statistical description of the rib shapes or their correlations to anthropometrics is a prerequisite to the development of numerical human body models representing target demographics. Variational autoencoders (VAE) as anatomical shape generators remain to be explored in terms of utilizing the latent vectors to control or interpret the representativeness of the generated results. In this paper, we propose a pipeline for developing a multi-rib cross-sectional shape generative model from CT images, which consists of the achievement of rib cross-sectional shape data from CT images using an anatomical indexing system and regular grids, and a unified framework to fit shape distributions and associate shapes to anthropometrics for different rib categories. Specifically, we collected CT images including 3193 ribs, surface regular grid is generated for each rib based on anatomical coordinates, the rib cross-sectional shapes are characterized by nodal coordinates and cortical bone thickness. The tensor structure of shape data based on regular grids enable the implementation of CNNs in the conditional variational autoencoder (CVAE). The CVAE is trained against an auxiliary classifier to decouple the low-dimensional representations of the inter- and intra- variations and fit each intra-variation by a Gaussian distribution simultaneously. Random tree regressors are further leveraged to associate each continuous intra-class space with the corresponding anthropometrics of the subjects, i.e., age, height and weight. As a result, with the rib class labels and the latent vectors sampled from Gaussian distributions or predicted from anthropometrics as the inputs, the decoder can generate valid rib cross-sectional shapes of given class labels (male/female, 2nd to 11th ribs) for arbitrary populational percentiles or specific age, height and weight, which paves the road for future biomedical and biomechanical studies considering the diversity of rib shapes across the population. • A deep learning-based pipeline is proposed for developing multi-rib shape generative model. • Rib cross-sectional shape data are achieved from CT images using an indexing system and regular grids. • A unified framework is proposed to fit shape distributions and associate shapes to anthropometrics for different ribs. • Valid rib cross-sectional shapes can be generated for arbitrary population percentiles or anthropometrics. [ABSTRACT FROM AUTHOR]

Adversarial training has become the mainstream method to boost adversarial robustness of deep models. However, it often suffers from the trade-off dilemma, where the use of adversarial examples hurts the standard generalization of models on natural data. To study this phenomenon, we investigate it from the perspective of spatial attention. In brief, standard training typically encourages a model to conduct a comprehensive check to input space. But adversarial training often causes a model to overly concentrate on sparse spatial regions. This reduced tendency is beneficial to avoid adversarial accumulation but easily makes the model ignore abundant discriminative information, thereby resulting in weak generalization. To address this issue, this paper introduces an Attention-Enhanced Learning Framework (AELF) for robustness training. The main idea is to enable the model to inherit the attention pattern of standard pre-trained model through an embedding-level regularization. To be specific, given a teacher model built on natural examples, the embedding distribution of teacher model is used as a static constraint to regulate the embedding outputs of the objective model. This design is mainly supported with that the embedding feature of standard model is usually recognized as a rich semantic integration of input. For implementation, we present a simplified AELFs that can achieve the regularization with single cross entropy loss via the parameter initialization and parameter update strategy. This avoids the extra consistency comparison operation between embedding vectors. Experimental observations verify the rationality of our argument, and experimental results demonstrate that it can achieve remarkable improvements in generalization under the high-level robustness. [ABSTRACT FROM AUTHOR]

Most existing voice conversion methods focus primarily on separating speech content from speaker information while overlooking the decoupling of pitch information. Additionally, the quality of converted speech significantly degrades when the speech of the target speaker is contaminated by noises. To address these issues, this paper proposes a noise-robust voice conversion model with multi-feature decoupling based on adversarial training. The proposed framework utilizes three distinct encoders to encode speech content, speaker identity, and pitch information independently, which aims to enhance the performance of decoupling by minimizing their mutual information and reduce the correlations between feature vectors. Moreover, a gradient reversal layer and a noise decoupling discriminator are incorporated into the framework, which extracts noise-resistant speaker representations and content representations through adversarial training to facilitate the synthesis of high-quality speech. In order to optimize the learning process, a training strategy is developed which involves alternating between clean and noisy data during the training of the encoder. This strategy effectively guides and expedites the convergence of the model. Experimental results demonstrate that compared to the state-of-the-art baselines of noise-robust voice conversion, the proposed model achieves improvements around 0.31 and 0.39 in terms of speech naturalness and speaker similarity evaluation metrics, respectively. [ABSTRACT FROM AUTHOR]

Accurate forecasting of wind power is crucial for efficiently scheduling power grids that incorporate wind energy. However, distribution instability of time-varying data can undermine the generalization capability of forecasting models in future periods. This paper introduces a transfer learning approach incorporating adversarial training and temporal convolution to address wind power forecasting from the distribution-centric perspective. The proposed approach can be divided into two joint modules: the temporal domain split and domain-adversarial temporal convolutional network. The former segregates training data into temporal domains characterized by the most significant distribution differences, while the latter aims is designed to learn shared knowledge from the segmented temporal domains. The shared knowledge is independent of distribution shifts, which can be generalized to future testing well. Specifically, the domain-adversarial temporal convolutional network is composed of a base model incorporating temporal convolutional network and a domain classifier. Both components are jointly optimized by minimizing power prediction loss and maximizing temporal domain classification loss. The data from actual wind turbines are employed to validate the proposed approach. Experimental results show that the superior performance of the proposed approach compared to benchmark models, achieving a remarkable 24.56% improvement in accuracy by introducing adversarial training. [ABSTRACT FROM AUTHOR]

• Fault diagnosis is extended to single domain generalization scenario by the MSG-ACN. • Multi-scale style generation strategy is established to generate diverse samples. • Adversarial contrastive learning strategy is designed to train the MSG-ACN. • Class-wise domain-invariant features with good generalization ability are learned. • The proposed MSG-ACN model outperforms the state-of-the-art fault diagnosis models. Domain generalization methods can effectively identify machinery faults under unseen new target working conditions. Nevertheless, most of them rely on data from multiple source domains that are available for model training. However, it is laborious difficult to collect complete monitoring data of machinery under multiple working conditions. Confronting the scenario that only one working condition is available, this paper proposes a novel single domain generalization model, termed multi-scale style generative and adversarial contrastive networks (MSG-ACN), which learns diagnosis knowledge from the single working condition and generalizes it to new working conditions. The main idea of the MSG-ACN model is to generate diverse samples in an extended domain via a domain generation module, and extract domain-invariant features from the source and extended domains via a diagnosis task module. A multi-scale style generation strategy is established to ensure that the generated samples contain abundant state information with the aids of multi-scale convolutional kernels and style learning. Furthermore, an adversarial contrastive learning strategy is designed to promote the learning of class-wise domain-invariant representations while maintaining the diversity of the generated samples. Extensive generalization diagnosis experiments on two datasets verify the superiority of the proposed method over the state-of-the-art fault diagnosis methods. [ABSTRACT FROM AUTHOR]

Federated Learning (FL) is a distributed learning approach that allows each client to retain its original data locally and share only the parameters of the local updates with the server. While FL can mitigate the problem of "data islands", the training process involving non-independent and identically distributed (non-IID) data still faces the formidable challenge of model performance degradation due to "client drift" in practical applications. To address this challenge, in this paper, we design a novel approach termed " Tw o-fold Knowl e dg e Dis t illation for non-IID Fed erated Learning" (FedTweet), meticulously designed for the personalized training of both local and global models within various heterogeneous data contexts. Specifically, the server employs global pseudo-data for fine-tuning the initial aggregated model through knowledge distillation and adopts dynamic aggregation weights for local generators based on model similarity to ensure diversity in global pseudo-data. Clients freeze the received global model as a teacher model and conduct adversarial training between the local model and local generator, thus preserving the personalized information in the local updates while correcting their directions. FedTweet enables both global and local models to serve as teacher models for each other, ensuring bidirectional guarantees for personalization and generalization. Finally, extensive experiments conducted on benchmark datasets demonstrate that FedTweet outperforms several previous FL methods on heterogeneous datasets. [ABSTRACT FROM AUTHOR]

Data-Free Knowledge Distillation (DFKD) aims to craft a customized student model from a pre-trained teacher model by synthesizing surrogate training images. However, a seldom-investigated scenario is to distill the knowledge to multiple heterogeneous students simultaneously. In this paper, we aim to study how to improve the performance by coevolving peer students, termed Data-Free Multi-Student Coevolved Distillation (DF-MSCD). Based on previous DFKD methods, we advance DF-MSCD by improving the data quality from the perspective of synthesizing unbiased, informative and diverse surrogate samples: 1) Unbiased. The disconnection of image synthesis among different timestamps during DFKD will lead to an unnoticed class imbalance problem. To tackle this problem, we reform the prior art into an unbiased variant by bridging the label distribution of the synthesized data among different timestamps. 2) Informative. Different from single-student DFKD, we encourage the interactions not only between teacher–student pairs, but also within peer students, driving a more comprehensive knowledge distillation. To this end, we devise a novel Inter-Student Adversarial Learning method to coevolve peer students with mutual benefits. 3) Diverse. To further promote Inter-Student Adversarial Learning, we develop Mixture-of-Generators, in which multiple generators are optimized to synthesize different yet complementary samples by playing min–max games with multiple students. Experiments are conducted to validate the effectiveness and efficiency of the proposed DF-MSCD, surpassing the existing state-of-the-arts on multiple popular benchmarks. To emphasize, our method can obtain heterogeneous students by training once, which is superior to single-student DFKD methods in terms of both training time and testing accuracy. [ABSTRACT FROM AUTHOR]

Continual learning is a high-potential technique that enables intelligent motor fault diagnosis models to extend new diagnosable fault classes without costly training from scratch. However, existing continual learning methods have the following limitations. (1) They manually detect new faults, which is labor-intensive, untimely, and more importantly, may lead to mistaken diagnosis results. (2) They adopt the traditional knowledge distillation to align the absolute responses of old and new models, which alleviates catastrophic forgetting but restricts flexible learning from incremental datasets. To overcome the above limitations, this paper proposes a novel self-driven continual learning framework for class-added motor fault diagnosis, which can spontaneously detect unseen faults and perform more flexible continual learning from incremental datasets. For the automatic detection of unseen faults, after collecting online samples, adversarial training with exemplars of each seen class is conducted to measure the class separability. The truth fault classes that are unseen for diagnosis models can be clearly distinguished from all seen classes, and correspondingly missed diagnosis or misdiagnosing can be avoided effectively and incremental samples with new fault types can be collected quickly. For the flexible continual learning strategy, a more flexible knowledge distillation is proposed to preserve the prediction propensity rather than the absolute response. This strategy not only keeps the recognition performance of old classes but also loosens unnecessary constraints and increases the diagnosis model plasticity to learn new knowledge from incremental datasets, thus improving the accuracy of motor fault diagnosis during continual learning. The effectiveness of the proposed method is verified by conducting fault simulation experiments of three-phase motors and its superiority is also demonstrated by comparing it with some state-of-the-art diagnosis methods. [ABSTRACT FROM AUTHOR]