The rise of Vision Transformers (ViTs) as cornerstone models in safety-critical applications like autonomous driving and medical diagnosis has shifted the focus from pure accuracy to verifiable trustworthiness. However, the very mechanisms used to explain these models, their internal attention maps, are themselves vulnerable. This creates a critical "trust gap," as the model's apparent reasoning can be maliciously manipulated. To systematically investigate this vulnerability, we introduce A-SAGE (Attention-based Steering Adversarial Generation by Corrupting Explanations), a dual-objective attack framework that forces a model to misclassify an input while simultaneously corrupting its internal attention patterns to generate a misleading explanation. A-SAGE achieves this by optimizing a unified loss that combines a standard classification objective with two explanation-specific terms: an attention entropy loss to diffuse the model's focus and an attention map distortion loss to steer the corrupted explanation towards a desired target. Our primary finding is A-SAGE's exceptional black-box transferability. Using a CaiT-S as a white-box surrogate, adversarial examples generated with imperceptible perturbations achieve attack success rates of 79.4% on ViT-B, 49.7% on ResNet-50, and over 81.5% on other transformers (DeiT-B,TNT-S). Crucially, these successful attacks do not merely destroy the explanation; they generate a coherent but false attention map that deceptively "justifies" the wrong prediction. These results reveal a systemic vulnerability in the core reasoning of modern foundation models, establishing A-SAGE as a critical benchmark for auditing the robustness of AI explainability.

Low probability of intercept (LPI) radar signals have been widely used in modern radars due to the advantages of being hardly intercepted by non-cooperative receivers. Therefore, the waveform recognition of LPI radar signals has recently gained increasing attention. In this letter, an improved LPI radar waveform recognition framework with local dense connection U-net and novel loss function is proposed. Specifically, radar signals are transformed into time-frequency images (TFIs) through time-frequency analysis techniques. Then, the local dense connection U-net is proposed to reduce interference of noise and enhance TFIs features. Finally, the recognition task is implemented through a deep convolutional neural network, which is trained with a noise-robust loss function called SSR-loss. Compared with other existing LPI radar waveform recognition frameworks, our method can obtain a significant improvement in recognition accuracy under noise conditions. When signal-to-noise ratio (SNR) is as low as −10 dB, the overall probability of successful recognition on twelve kinds of typical modulated signals can reach up to 91.17%.

In complex electromagnetic environments, the challenge of signal sorting task for multi-function radars (MFRs) with various work modes has arisen. The previous methods are prone to cause the so-called “increasing batch” problem, which means that the work modes of one MFR may be sorted into multiple emitters. In this letter, a MFR signal sorting method based on complex network is proposed to tackle the problem mentioned above. The novel method utilizes limited penetrable visibility graph to construct the network from interleaved radar pulse sequences, then employs label propagation algorithm and density peak clustering to detect community structures, thus fulfilling deinterleaving of pulses from several MFRs. Simulation results show that the proposed method is effective to alleviate the “increasing batch” problem, and is also robust under non-ideal conditions.

Although deep neural networks (DNNs) have achieved excellent generalization performance on well-labelled datasets, the real-world training sets usually suffer from label noise. Owing to the powerful representational ability, DNNs may be vulnerable to the noisy data thus resulting in a sharp deterioration of test performance. To tackle this problem, we propose a novel cross-training framework to leverage two synergic deep architectures, where each member plays the roles of both teacher of generating curriculums and student of updating model parameters via a novel loss function. It is convenient to train our proposed model directly without any prior information such as noise rate. The state-of-the-art experimental results on several benchmark datasets under different types of label noise have demonstrated the effectiveness and robustness of our method.

We propose a salient-context based semantic matching method to improve relevance ranking in information retrieval. We first propose a new notion of salient context and then define how to measure it. Then we show how the most salient context can be located with a sliding window technique. Finally, we use the semantic similarity between a query term and the most salient context terms in a corpus of documents to rank those documents. Experiments on various TREC collections show the effectiveness of our model compared to the state-of-the-art methods.

Semisupervised incremental learning is the task of classifying data streams with partially labeled data when annotation information is difficult to obtain. Besides the sequential learning manner and lack of label information, multiple novel classes and concept drift may emerge from incremental learning. Most previous studies have only considered these problems in part. To tackle challenges involved in semisupervised incremental learning, an adaptive matrix sketching and clustering method is proposed in this letter, which cohesively and adaptively classifies known classes, identifies multiple novel classes, and updates the learning model. Experiments were conducted to evaluate this method on three benchmark datasets, containing various data types, including network attack analysis, the geospatial information of forests, and images of handwritten numbers. Results validated the effectiveness of our proposed method and its superiority over many previous studies.

Extracting product aspects and their associated sentiments is one of the key tasks in sentiment analysis. Estimating the confidences of extracted aspects is important to ensure the performance. To tackle the issue, this paper proposes a two-step estimation method. Collocations of product features and opinion words are initially extracted through pattern bootstrapping. A criterion synthesizing two measurements, Popularity and Reliability, is novelly exploited to assess both patterns and features. Then the features are further clustered into aspects based on path similarities in the Word Net. Each cluster is assigned a weight based on its Compactness and Texture, and the light ones are filtered out. In addition, this paper also captures global aspect reputations by aggregating sentiment strengths through opinion collocations. Experimental results on a benchmark data set with 5 products demonstrate the effectiveness and reliability of our proposed method.