Recent studies have revealed that Deep Neural Networks (DNNs) are highly vulnerable to adversarial examples, which are generated by introducing imperceptible perturbations to clean images, leading to misclassification. The existing untargeted attack usually only focuses on weakening the original class when generating adversarial examples, ignoring the model’s prediction distribution for other classes. Based on the analysis of the attention heatmap of model decision and the existing adversarial attack results, we find that the high-confidence negative classes of the images often reflect the natural weak directions in the model decision, and updating the adversarial examples along this direction is more likely to help it deviate from the original class. Therefore, we propose an untargeted adversarial example generation method via Negative Class Guidance (NCG). First, the logits of the clean image are extracted according to the classification confidence. Second, the soft label is generated via smoothing and normalization operations. Finally, a novel loss function is derived that integrates negative class information with the soft label to guide the update direction of adversarial examples. Extensive experiments conducted on the ImageNet dataset demonstrate that NCG substantially enhances the adversarial transferability of state-of-the-art attack methodologies on both Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), highlighting its effectiveness in black-box attack scenarios.

Accurately predicting quarterly or monthly energy consumption remains challenging so far. Despite the abundance of relevant studies, most of them focus on univariate modeling. Moreover, the core of nearly all multivariate forecasting studies is an unstable forecasting system based on a single model. Therefore, there is an urgent need for an efficient and rational prediction method. For the prediction task of quarterly or monthly energy consumption characterized by small samples and nonlinearity, this article develops a new joint forecasting-centered forecasting framework by integrating machine learning and grey system theory. In this forecasting framework, grey relational analysis is used to filter the influencing factors of the study object, a new adaptive weighted least squares support vector regression model is developed to describe the relationship between the study object and the filtered influencing factors, and a new difference equation prediction model is employed to predict the future values of the filtered influencing factors. The joint forecasting task is accomplished by inputting the future values of the filtered influencing factors into the trained adaptive weighted least squares support vector regression model. Experimental simulation results demonstrate that the two prediction models developed in this framework, along with the overall forecasting approach, outperform competing methods. These results confirm the effectiveness of the proposed forecasting framework in accurately predicting quarterly or monthly energy consumption, even in scenarios with limited data and nonlinear relationships.

Traditional fine-grained image classification relies extensively on large-scale training samples with annotated ground truth. However, in the real world, some fine-grained categories are represented by only a few available images, and existing few-shot models struggle to distinguish the subtle differences among them. Furthermore, the challenge is compounded by the fact that intra-class distances for some fine-grained categories can be significantly large, whereas inter-class distances might be minimal, leading to distinct task-specific distinguishing features for each category. To address these challenges, we propose a novel network, FicNet, utilizing Multi-Frequency Neighborhood (MFN) and Double-Cross Modulation (DCM). MFN captures a multi-frequency structure representation that is independent of the background by integrating spatial and frequency domain information, reducing intra-class distances. Concurrently, DCM modulates the representation through global context and inter-class relationships, enabling both support and query features to align with complete targets and respond to identical parts. This approach facilitates the accurate identification of subtle inter-class differences. Comprehensive experiments conducted on three fine-grained benchmark datasets for two few-shot tasks have verified that FicNet exhibits exceptional performance compared to state-of-the-art methods. Notably, it achieves classification accuracies of 93.17% and 95.36% on the “Caltech-UCSD Birds” and “Stanford Cars” datasets, respectively, surpassing the benchmarks set by general fine-grained image classification methods.

Plentiful adversarial attack researches have revealed the fragility of deep neural networks (DNNs), where the imperceptible perturbations can cause drastic changes in the output. Among the diverse types of attack methods, gradient-based attacks are powerful and easy to implement, arousing wide concern for the security problem of DNNs. However, under the black-box setting, the existing gradient-based attacks have much trouble in breaking through DNN models with defense technologies, especially those adversarially trained models. To make adversarial examples more transferable, in this paper, we explore the fluctuation phenomenon on the plus-minus sign of the adversarial perturbations’ pixels during the generation of adversarial examples, and propose an ingenious Gradient Relevance Attack (GRA). Specifically, two gradient relevance frameworks are presented to better utilize the information in the neighbor-hood of the input, which can correct the update direction adaptively. Then we adjust the update step at each iteration with a decay indicator to counter the fluctuation. Experiment results on a subset of the ILSVRC 2012 validation set forcefully verify the effectiveness of GRA. Furthermore, the attack success rates of 68.7% and 64.8% on Tencent Cloud and Baidu AI Cloud further indicate that GRA can craft adversarial examples with the ability to transfer across both datasets and model architectures. Code is released at https://github.com/RYC-98/GRA.

Deep neural networks are highly susceptible to imperceptible noise, even to the human eye. While high attack success rate has been achieved in white-box setting, the attack performance tends to decline in black-box environments. To address this challenge, this study introduces a novel adaptive points selecting iterative fast gradient sign method, named AI-FGSM, which leverages temporal neighborhoods to enhance transferability performance in black-box environments. AI-FGSM is unique in that it extracts gradients from additional points in the temporal neighborhood and adjusts the current gradient using previous gradients to maintain the updated trend's stability. This gradient correction enables faster optimization and restricts falling into local optima. Experimental results on the ImageNet dataset show that AI-FGSM outperforms advanced gradient-based attack methods such as MI-FGSM and NI-FGSM in terms of attack success rate and transferability. Moreover, in black-box environments, AI-FGSM achieves an average attack performance improvement of 44.2% for some models with adversarial training and 32.9% for several models without adversarial training. These results confirm the performance and effectiveness of AI-FGSM, underscoring its potential as a powerful tool for improving the transferability of adversarial attacks.