Deep learning models have revolutionized various domains but have also raised concerns regarding their security and reliability. Adversarial attacks and coverage-based testing have been extensively studied to assess and enhance the dependability of deep neural networks. However, current research in this area has reached a state of stagnation. Adversarial attacks focus on exploiting vulnerabilities in models, while coverage-based testing aims to achieve comprehensive testing but overlooks application scenarios. Moreover, evaluating test cases solely based on their fault-revealing capability is insufficient. To address these limitations, we propose an innovative interdisciplinary framework that incorporates human-computer interaction methods in deep learning security testing. By considering the attributes of model application scenarios, we can design more effective test suites that intend to reveal the model's behavior across various scenarios, aiding in the identification of potential defects. Consequently, the test suite plays a crucial role in the testing process of deep learning models, contributing to the assurance of model robustness and reliability. Additionally, we establish a comprehensive evaluation metric for test suite quality, considering factors such as diversity and naturalness. This framework promotes reliable and secure deployment of deep learning models, fostering interdisciplinary collaboration between artificial intelligence and human-computer interaction.

Deep learning technologies have demonstrated remarkable performance in vulnerability detection. Existing works primarily adopt a uniform and consistent feature learning pattern across the entire target set. While designed for general-purpose detection tasks, they lack sensitivity towards target code comprising multiple functional modules or diverse vulnerability subtypes. In this paper, we present a knowledge fusion-based vulnerability detection method (KF-GVD) that integrates specific vulnerability knowledge into the Graph Neural Network feature learning process. KF-GVD achieves accurate vulnerability detection across different functional modules of the Linux kernel and vulnerability subtypes without compromising general task performance. Extensive experiments demonstrate that KF-GVD outperforms SOTAs on function-level and statement-level vulnerability detection across various target tasks, with an average increase of 40.9% in precision and 26.1% in recall. Notably, KF-GVD discovered 9 undisclosed vulnerabilities when employing on C/C++ open-source projects without ground truth.

In code vulnerability detection tasks, a detector trained on a label-rich source domain fails to provide accurate prediction on new or unseen target domains due to the lack of labeled training data on target domains.Previous studies mainly utilize domain adaptation to perform cross-domain vulnerability detection.But they ignore the negative effect of private semantic characteristics of the target domain for domain alignment, which easily causes the problem of negative transfer.In addition, these methods forcibly reduce the distribution discrepancy between domains and do not take into account the interference of irrelevant target instances for distributional domain alignment, which leads to the problem of excessive alignment.To address the above issues, we propose a novel cross-domain code vulnerability detection framework named MN-CRI.Specifically, we introduce mutual nearest neighbor contrastive learning to align the source domain and target domain geometrically, which could align the common semantic characteristics of two domains and separate out the private semantic characteristics of each domain.Furthermore, we introduce an instance re-weighting scheme to alleviate the problem of excessive alignment.This scheme dynamically assign different weights to instances, reducing the contribution of irrelevant instances so as to achieve better domain alignment.Finally, extensive experiments demonstrate that MNCRI significantly outperforms state-of-the-art cross-domain code vulnerability detection methods by a large margin.

With the development of deep learning, software vulnerability detection methods based on deep learning have achieved great success, which outperform traditional methods in efficiency and precision. At the training stage, all training samples are treated equally and presented in random order. However, in software vulnerability detection tasks, the detection difficulties of different samples vary greatly. Similar to the human learning mechanism following an easy-to-difficult curriculum learning procedure, vulnerability detection models can also benefit from the easy-to-hard curriculums. Motivated by this observation, we introduce curriculum learning for automated software vulnerability detection, which is capable of arranging easy-to-difficult training samples to learn better detection models without any human intervention. Experimental results show that our method achieves obvious performance improvements compared to baseline models.

Deep learning-based autonomous driving systems have been extensively researched due to their superior performance compared to traditional methods. Specifically, end-to-end deep learning systems have been developed, which directly output control signals for vehicles using various sensor inputs. However, deep learning techniques are vulnerable to security issues, generating adversarial examples that can attack the output of the relevant model. This paper proposes an adversarial example generation method that applies a patch to pedestrians' clothing, which can generate dangerous behaviors when the pedestrian appears within the camera lens, thereby attacking the end-to-end autonomous driving system. The proposed method is validated using the CARLA simulator, and the results demonstrate successful attacks in various weather and lighting conditions, exposing the security vulnerabilities of this type of system. This study highlights the need for further research to address these vulnerabilities and ensure the safety of autonomous driving systems.

Universal domain adaptation (UniDA) aims to transfer knowledge learned from a labeled source domain to an unlabeled target domain under domain shift and category shift. Without prior category overlap information, it is challenging to simultaneously align the common categories between two domains and separate their respective private categories. Additionally, previous studies utilize the source classifier's prediction to obtain various known labels and one generic "unknown" label of target samples. However, over-reliance on learned classifier knowledge is inevitably biased to source data, ignoring the intrinsic structure of target domain. Therefore, in this paper, we propose a novel two-stage UniDA framework called MATHS based on the principle of mutual nearest neighbor contrast and hybrid prototype discrimination. In the first stage, we design an efficient mutual nearest neighbor contrastive learning scheme to achieve feature alignment, which exploits the instance-level affinity relationship to uncover the intrinsic structure of two domains. We introduce a bimodality hypothesis for the maximum discriminative probability distribution to detect the possible target private samples, and present a data-based statistical approach to separate the common and private categories. In the second stage, to obtain more reliable label predictions, we propose an incremental pseudo-classifier for target data only, which is driven by the hybrid representative prototypes. A confidence-guided prototype contrastive loss is designed to optimize the category allocation uncertainty via a self-training mechanism. Extensive experiments on three benchmarks demonstrate that MATHS outperforms previous state-of-the-arts on most UniDA settings.

Time series classification is a critical problem in the machine learning field, which spawns numerous research works on it. In this work, we propose AttLSTM-CNNs, an attention-based LSTM network and convolution network that jointly extracts the underlying pattern among the time-series for the classification. The attention-based LSTM automatically captures the long-term temporal dependency among the series, and the CNN describes the spatial sparsity and heterogeneity in the data. The extensive experiments show that the proposed model outperforms the other methods for time-series classification.