This review examines recent feature-based transfer learning techniques for Cross-Project Software Defect Prediction. We summarize representative approaches in five categories-feature selection, feature mapping/alignment, deep/adversarial learning, semantics-enhanced transfer, and hybrid/multi-source designs-and discuss their reported effectiveness across common benchmark datasets. The review highlights trade-offs between predictive accuracy, computational cost, and model interpretability, and concludes with open challenges and directions for future work.

The convergence of autonomous driving and deep learning technologies has brought unprecedented convenience to future mobility while also introducing new security challenges. Adversarial attacks can exploit minor perturbations to deceive vehicle perception and decision-making processes, thereby posing potential dangers to both passengers and pedestrians. Although considerable progress has been made in developing adversarial detection and defense mechanisms, significant challenges remain, including high computational overhead, limited real-time performance, incomplete multi-modal integration, and insufficient understanding of black-box attacks and cross-scenario transfer. To comprehensively enhance the security and robustness of autonomous driving systems, it is necessary to further expand the data and model scales of adversarial examples, promote multi-modal fusion, improve the generalizability and real-time performance of adversarial defenses, and conduct additional validation under realistic and complex environments. Based on these considerations, this paper systematically reviews recent advances and gaps in adversarial research for autonomous driving. Furthermore, it explore future research directions from the perspectives of multi-modal fusion, dataset scale expansion, black-box defense, and the emergent role of large language models.

Deep neural networks can exhibit defects that lead to erroneous decisions in safety-critical scenarios, potentially causing severe consequences. While various repair techniques exist, methods based on retraining are often costly and imprecise and existing searchbased approaches can suffer from inefficient fault localization and patch generation. This paper introduces an end-to-end multi-objective, search-based framework for repairing DNNs that couples spectrumguided fault localization with an enhanced particle-swarm patch generator, and it directly modifies neural weights to address these challenges. Our approach features a novel, two-stage fault localization technique that first uses spectrum-based analysis to identify suspicious neurons and then applies multi-objective optimization to pinpoint the most critical weights for repair. For patch generation, we employ an enhanced Multi-Objective Particle Swarm Optimization algorithm designed for faster convergence and more effective exploration of the solution space. We evaluated our framework on multiple DNN architectures using four public datasets (Fashion-MNIST, CIFAR-10, LFW, and GTSRB). The proposed fault localization method led to an 18% increase in repair rate on the Fashion-MNIST dataset compared to state-of-the-art techniques. Furthermore, our end-to-end framework successfully reduced the most frequent misclassification type in a pretrained model by 31.3 % and proved effective for the adaptive repair of fully-trained models without causing catastrophic performance degradation. The results demonstrate that our approach offers a more precise and less disruptive alternative to retraining, providing a practical and cost-efficient method for the targeted repair of DNNs.

Multiobjective multitask optimization (MO-MTO) aims to exploit the similarities among different multiobjective optimization tasks through knowledge transfer (KT), facilitating their simultaneous resolution. The effective design of KT techniques embedded in multiobjective evolutionary optimizers is crucial for enhancing the performance of multiobjective multitask evolutionary algorithms (MO-MTEAs). However, a significant limitation of existing KT techniques in MO-MTEAs is their equal treatment of particles/individuals for transferred knowledge reception, which can negatively impact the balance of diversity and convergence in population evolution. To remedy this limitation, this article proposes a new MO-MTEA, named MTEA-DCK, which incorporates diversity-oriented KT (DKT) and convergence-oriented KT (CKT) techniques tailored for different particles in the population. MTEA-DCK utilizes a strength-Pareto-based competitive mechanism to divide particles into winners and losers: 1) for winners, DKT is conducted via an intertask domain alignment approach to enhance population diversity and 2) for losers, CKT is executed within the unified search space to improve convergence. Additionally, to ensure robust performance on complex task combinations, we introduce two automatic parameter control strategies specifically designed for these KT techniques. MTEA-DCK was performed on 39 benchmark MO-MTO problems and demonstrated superior performance compared to eight state-of-the-art MO-MTEAs and six multiobjective evolutionary algorithms. Finally, we present three real-world MO-MTO application cases, where our approach also yielded better results than other algorithms.

Solving multiobjective optimization problems (MOPs) through metaheuristic methods gets considerable attention. Based on the classical variation operators, several enhanced operators, as well as multiobjective optimization evolutionary algorithms, have been developed. Among these operators, the competitive swarm optimizer (CSO) exhibits promising performance. However, it encounters difficulties when tackling constrained MOPs (CMOPs) with large objective spaces or complex infeasible regions. In this article, a competitive and cooperative swarm optimizer is proposed, which contains two particle update strategies: 1) the CSO provides faster convergence speed to accelerate the approximation of the Pareto front and 2) the cooperative swarm optimizer suggests a mutual-learning strategy to enhance the ability to jump out of local feasible regions or local optima. We also present a new algorithm for CMOPs. The results on four benchmark suites with 47 instances demonstrate the superiority of our approach compared with other state-of-the-art methods. Additionally, its effectiveness on large-scale CMOPs has also been verified.