Although Deep Neural Networks (DNNs) have achieved significant success across various industries, their erroneous behaviors have raised widespread concerns. In safety-critical domains where DNNs are deployed, certain types of specific misclassifications (i.e., instances of one specific class being misclassified as another) can lead to catastrophic consequences and urgently need to be corrected. However, existing repair approaches often neglect the adverse impact of general feature extraction patterns on distinguishing specific classes, resulting in limited correction of specific misclassifications. Furthermore, most approaches correct errors by directly adjusting the global decision boundary, which inevitably disturbs the well-trained boundaries and leads to significant accuracy loss. In this article, we propose SMiR, an effective repair framework for addressing specific misclassifications with marginal accuracy loss. Rather than patching the original model, our method localizes and recovers the distorted feature dimensions to mitigate their negative impact on the distinction between specific classes. Moreover, SMiR narrows the scope of repair to the filtered suspect set, protecting other well-trained decision boundaries from disturbance. On average, our approach repairs 27% \(\sim\) 36.2% more specific misclassifications than the baselines, while introducing only 7.8% \(\sim\) 70.9% as many new errors. Besides, SMiR maintains state-of-the-art performance on safety-critical datasets.
While Deep Neural Networks (DNNs) have achieved remarkable progress in cutting-edge domains, their inherent brittleness has become a growing concern. To ensure the reliability and safety of DNN-enabled software, DNN testing has emerged as an indispensable practice. Within this context, test input prioritization is essential for early fault detection and reducing labeling costs. However, it remains challenging to accurately identify failure-inducing inputs. Although decision ambiguity and distributional atypicality are two widely adopted perspectives for characterizing inter-class competition and intra-class typicality respectively, relying on either perspective in isolation inevitably introduces blind spots. In this paper, we propose DuFP (Dual perspective Feature space Prioritization), a KNN density-based test input prioritization approach for DNNs that jointly incorporates both inter-class and intra-class perspectives. The prioritization framework of DuFP is built upon class-conditional density estimation. Based on the estimation results, prediction correctness is characterized by an ambiguity score and an atypicality score, with the former reflecting decision ambiguity and the latter quantifying distributional atypicality. A hybrid uncertainty score is then constructed by integrating both scores to guide the final prioritization. We evaluate DuFP on prioritization and selection tasks across image and text datasets under clean, corrupted, and adversarial scenarios. Experimental results demonstrate that DuFP effectively and efficiently prioritizes fault-inducing inputs and outperforms state-of-the-art approaches.
Adversarial examples have revealed the vulnerability of deep neural networks, and their transferability makes black-box attacks particularly concerning. However, perturbations crafted on a surrogate model often do not remain sufficiently effective on unseen target models. In this paper, we revisit this issue from a frequency-domain perspective and observe that perturbation optimization can become overly dependent on specific spectral patterns, which weakens cross-model transfer. To address this problem, we propose frequency-consistent regularization (FCR), a simple plug-in strategy that can be combined with existing iterative attacks. FCR introduces multiple low-frequency preserving views with randomly sampled frequency ranges at each iteration and optimizes perturbations across these varied views. In this way, the generated perturbations are less tied to a specific frequency configuration and show improved transferability. Experimental results show that FCR consistently improves the transfer performance of various iterative attacks. The improvement is observed not only in standard target models but also in adversarially trained models, where the gain is often more pronounced.
Adversarial examples reveal critical vulnerabilities in deep neural networks, posing significant risks in real-world deployment. In black-box settings, transferable targeted attacks rely on surrogate models but often suffer from low success rates. We argue that this limitation arises not only from surrogate-boundary overfitting but also from insufficient alignment with the target semantic space, which restricts the ability of adversarial examples to encode target-specific characteristics. To address this issue, we propose Cosine-Based Logit Alignment (CBLA), a unified framework for transferable targeted attacks. CBLA replaces the conventional cross-entropy loss with a cosine similarity objective to enhance directional alignment in logit space and alleviate gradient saturation. In addition, a semantic-invariant transformation strategy is introduced to improve structural consistency and cross-model generalization. Experiments on the ImageNet validation set demonstrate that CBLA consistently improves targeted attack success rates, achieving an average gain of 4.55% over strong baselines across multiple architectures.