Convolutional Neural Networks (CNNs) have been widely used in various fields. However, it is essential to perform sufficient testing to detect internal defects before deploying CNNs, especially in security-sensitive scenarios. Generating error-inducing inputs to trigger erroneous behavior is the primary way to detect CNN model defects. However, in practice, when the model under test is a black-box CNN model without accessible internal information, in some scenarios it is still necessary to generate high-quality test inputs within a limited testing budget. In such a new scenario, a potential approach is to generate transferable test inputs by analyzing the internal knowledge of other white-box CNN models similar to the model under test, and then use transferable test inputs to test the black-box CNN model. The main challenge in generating transferable test inputs is how to improve their error-inducing capability for different CNN models without changing the test oracle. We found that different CNN models make predictions based on features of similar important regions in images. Adding targeted perturbations to important regions will generate transferable test inputs with high realism. Therefore, we propose the Interpretable Analysis-based Transferable Test (IATT) Generation method for CNNs, which employs interpretation methods of CNN models to explain and localize important regions in test inputs, using backpropagation optimizer and perturbation mask process to add targeted perturbations to these important regions, thereby generating transferable test inputs. This process is repeated to iteratively optimize the transferability and realism of the test inputs. To verify the effectiveness of IATT, we perform experimental studies on nine deep learning models, including ResNet-50 and Vit-B/16, and commercial computer vision system Google Cloud Vision , and compared our method with four state-of-the-art baseline methods. Experimental results show that transferable test inputs generated by IATT can effectively cause black-box target models to output incorrect results. Compared to existing testing and adversarial attack methods, the average Error-inducing Success Rate (ESR) in different testing scenarios is 18.1%–52.7% greater than the baseline methods. Additionally, the test inputs generated by IATT achieve high ESR while maintaining high realism.

In the field of Natural Language Processing (NLP), Language Models (LMs) are widely applied in tasks such as text classification, machine translation, and knowledge reasoning. However, the defects of LMs make them vulnerable to adversarial attacks, resulting in substantial economic losses. Adversarial examples can effectively expose vulnerabilities of LMs and be used for adversarial training to improve the robustness of the models. Existing methods mostly generate adversarial examples by first selecting important tokens and then adding perturbations to them. Such methods require a large number of queries to the victim model, which is not applicable in scenarios where the query budget is limited. To address the imperative demands for more query-efficient adversarial example generation, this paper presents CBAPB, a Classification Boundary Adjacent Perturbation and Back-track based textual adversarial attack method, which initially introduces coarse-grained perturbations at random positions while preserving the original semantics of input examples until they reach the similarity threshold. Subsequently, fine-grained perturbation backtracking is conducted on all successfully misclassified examples to minimize perturbation magnitudes. We conduct multiple experiments on the Yelp Reviews, AG News, and DBpedia datasets by employing BERT as the victim model. Comparative analysis against baselines reveals that CBAPB requires merely 3.2% of the average query times of these baselines, while increasing the attack success rate by 7.6%, with only a slight decrease of 1.5% in textual similarity. Experimental results demonstrate the effectiveness of CBAPB, which is not only a query-efficient method but also with greater attack success rates.

Abstract Image denoising aims to remove noise from images and improve the quality of images. However, most image denoising methods heavily rely on pairwise training strategies and strict prior knowledge about image structure or noise distribution. While these methods exhibit significant results when handling known types of noise, their generalization performance diminishes when confronted with images containing unknown noise distributions. To address this issue, a two‐stage approach is introduced for enhancing the generalizability of image denoising. The proposed method does not rely on a large amount of paired data or prior knowledge of the noise type and level. Instead, it constructs a denoising pipeline with improved generalizability through an MLP‐based denoiser and generative diffusion prior. Specifically, in the first stage, an initial denoised image is predicted with a structure resembling that of the underlying clean image by introducing an MLP‐based U‐shaped denoising network aided by an implicit structural prior. In the second stage, the generalizability and quality of the denoiser are further enhanced by conditioning the result obtained from the previous stage on the pretrained denoising diffusion null‐space model. Extensive experimentation on multiple datasets demonstrates that this method exhibits better denoising performance and generalizability than other image denoising methods.

Recently, deep neural networks (DNN) have been widely applied in various fields, such as image classification, even replace humans to make decisions in some specific tasks. However, like traditional software, DNNs inevitably contain defects. If defective DNN models are applied in safety-critical fields, such as autonomous driving and medical diagnosis, it may cause disastrous consequences. Therefore, effective testing methods are urgently needed to improve the reliability of DNNs. The existing DNN testing methods typically generate test data by either globally modifying the original data or taking adversarial approaches. The generated test data typically struggle to simultaneously achieve good performance in both the degree of difference from the original data and the Error-inducing Success Rate (ESR) with respect to the target DNN model. Moreover, the perturbation-based methods are difficult to be understood by humans. To address the above issue, this paper proposes DeepIA, an interpretability analysis based test data generation method for DNN. DeepIA analyzes the interpretability of decision-making behaviors for DNN. According to the interpretability analysis results, the original training data is split into different regions to evaluate their influences on decision-making results of the DNN. After that, the most significant regions of the original test data are transformed to generate new test data. Experimental results show that the interpretability method effectively enhances the misleading ability of DeepIA for the DNN model under test. Compared with DeepTest and DeepSearch, DeepIA can generate test data with minor permutations and greater ESR.

Natural language processing (NLP) models are vulnerable to adversarial examples. Generating high-quality adversarial examples, which expose the vulnerability of NLP models and can be used to evaluate and improve their robustness, deserves further research. Existing techniques of generating adversarial examples in the NLP field are typically based on greedy synonym replacements, which may result in out-of-context and unnatural perturbations, and are easily identifiable by humans. In this paper, we present MOB-TAG, a Multi-objective Optimization based Textual Adversarial Example Generation method, which includes three types of perturbations, and utilizes pre-trained models such as BERT and RoBERTa to generate high-quality adversarial examples. MOBTAG generates fluent and grammatical output through a mask-then-infill procedure, with introducing multi-objective optimization and genetic algorithm to pursue a high attack success rate while maintaining a high level of similarity and readability. Experimental results show that compared with methods such as TextFooler, BERTAttack, and CLARE, MOBTAG improves the attack success rate and the textual similarity by at least 11.8% and 0.09 on average, respectively.

High quality code comments are of great value for program maintenance. However, during the development process, developers often neglect to update corresponding comments when changing code. In such case, inconsistent comments are introduced which affect the maintainability of the software. In previous work, code changes are usually performed by treating the code as ordinary text and the structural information of the code are ignored. In this paper, we propose an approach named SICUP (Structural Information based Comment UPdater) to provide a new solution for comment updating tasks. SICUP uses the structural information of the code to help updating comments by constructing different sequences of ASTs. Experiments on a popular dataset demonstrates that SICUP outperforms CUP, which is an effective deep learning-based approach in terms of accuracy and recall.

In this paper, we present a novel method to simulate the influences of small social group on pedestrian's behaviors under emergency situations. Our method is built on an important observation that the relationships between group members are usually different and even mutual relationships between group members might be asymmetric. Two phenomena can be produced by our method based on this observation. The first is group aggregation phenomenon which means that pedestrians tend to stay closer to their socially close group members than socially distant members. The second phenomenon is the complicated process of searching for lost members which is modeled as a cost-based function in our method. This function will guide pedestrians to make many decisions like whether to look for the lost members, who will be searched for, who will conduct the search as well as when to abort the search. The experimental results show that our method can produce very real social group behaviors under emergency situations.