The rapid growth of Internet of Things (IoT) devices has heightened hardware security concerns, particularly with the emergence of hardware Trojans (HTs) as malicious components in integrated circuits. These HTs pose significant threats to information privacy and system performance. The early detection methods rely on golden references and domain knowledge, limiting their implementation in large-scale integrated circuits. Traditional machine learning-based HT detections can effectively identify infected circuits without golden chips. However, locating and evaluating the specific behavior of HTs in vast and complex circuits is challenging. To address these challenges, this paper presents an HT Object Detection (HTOD) problem, which involves two primary tasks: identifying the boundaries and sizes of HTs within circuits and distinguishing between different HT behaviors. We have developed a two-stage HT object detection framework based on a bidirectional jumping knowledge network called HTOD-BGNN to solve the HTOD problem, which enables progressive refinement of localized regions. It incorporates enhanced base-type features to ensure scalability and information retention during modeling. Additionally, the implementation of sample augmentation alleviates the challenges of sample imbalance and scarcity. The proposed method enables more intelligent circuit detection, achieving 100% and 87.5% detection rates for Trojan triggering and payload behaviors, respectively. It demonstrates superior localization performance and effective generalization to unseen circuit scenarios via its region refinement mechanism, achieving F1-scores of 54.01% on TrustHub and 90.04% on TRIT datasets.

Decoding gene regulatory networks is critical for understanding cell identity and disease. While single-cell multiomics technologies can measure both gene expression and chromatin accessibility, precisely linking regulatory elements to their target genes remains a great challenge. To this end, we propose MICA, a deep learning interpretable model that treats gene regulation as a bi-directional transition from “regulatory syntax” to “functional language”. The core of MICA is a bilinear attention mechanism that performs end-to-end optimization through a cross-modal bidirectional generative task. The task forces the model to learn attentional weights with predictive power to go beyond simple correlations and capture functional regulatory dependencies. Experimental results show that our approach exhibits outstanding performance in the cross-modal generation task, and in a subsequent interpretability validation we corroborate from multiple perspectives that the resulting attentional mechanism is biologically meaningful. This work provides a powerful and interpretable computational tool for deciphering complex gene regulatory codes at single-cell resolution, offering new perspectives for systematic understanding of cellular heterogeneity.

Crossmodal matching in single-cell omics is essential for explaining biological regulatory mechanisms and enhancing downstream analyses. However, current single-cell crossmodal models often suffer from three limitations: sparse modality signals, underutilization of biological attributes, and insufficient modeling of regulatory interactions. These challenges hinder generalization in data-scarce settings and restrict the ability to uncover fine-grained biologically meaningful crossmodal relationships. Here, we present a novel framework which reformulates crossmodal matching as a graph classification task on Attributed Bipartite Graphs (ABGs). It models single-cell ATAC-RNA data as an ABG, where each expressed ATAC and RNA is treated as a distinct node with unique IDs and biological features. To model crossmodal interaction patterns on the constructed ABG, we propose $\text{Bi}^2\text{Former}$, a **bi**ologically-driven **bi**partite graph trans**former** that learns interpretable attention over ATAC–RNA pairs. This design enables the model to effectively learn and explain biological regulatory relationships between ATAC and RNA modalities. Extensive experiments demonstrate that $\text{Bi}^2\text{Former}$ achieves state-of-the-art performance in crossmodal matching across diverse datasets, remains robust under sparse training data, generalizes to unseen cell types and datasets, and reveals biologically meaningful regulatory patterns. This work pioneers an ABG-based approach for single-cell crossmodal matching, offering a powerful framework for uncovering regulatory interactions at the single-cell omics. Our code is available at: https://github.com/wangxiaotang0906/Bi2Former.

In black-box attack scenarios, enhancing the transferability of adversarial samples for targeted attacks presents significant challenges and is crucial for real-world applications. Most current research focuses on attack strategies that rely on data augmentation or generator training, which tend to perform well in untargeted attacks but exhibit limited effectiveness in targeted attack scenarios. To address this, we propose a novel framework for adversarial sample generation called TADA, designed to modify the sample distribution and improve transferability in both attack scenarios. TADA consists of two key components. First, we change the image distribution in the diffusion model by using the gradient of deviation between images to guide the denoising process. Second, we introduce a fine-tuning strategy that uses the gradient norm of the label loss as a constraint to bootstrap the loss to a flat minimum and regularize the penalty for large weight values. With these two strategies, TADA can more effectively modify the original distribution of images to generate highly transferable adversarial examples. Through extensive experiments with various general-purpose training and defense models, our TADA consistently outperform state-of-the-art attack methods in both scenarios. Notably, the reconstructed images enhance the transferability of existing attack methods. For example, the CFM attack achieves a 12.3% increase in success rate for targeted attacks when transitioning from Res-50 to Inc-Res-v2.

Active learning (AL) tries to maximize the model’s performance when the labeled data set is limited, and the annotation cost is high. Although it can be efficiently implemented in deep neural networks (DNNs), it is questionable whether the model can maintain the ability to generalize well when there are significant distributional deviations between the labeled and unlabeled data sets. In this article, we consider introducing adversarial training and adversarial samples into AL to mitigate the problem of degraded generalization performance due to different data distributions. In particular, our proposed adversarial training AL (ATAL) has two advantages, one is that adversarial training by different networks enables the network to have better prediction performance and robustness with limited labeled samples. The other is that the adversarial samples generated by the adversarial training can effectively expand the labeled data set so that the designed query function can efficiently select the most informative unlabeled samples based on the expanded labeled data set. Extensive experiments have been performed to verify the feasibility and efficiency of our proposed method, i.e., CIFAR-10 demonstrates the effectiveness of our method—new state-of-the-art robustness and accuracy are achieved.

Centralized learning now faces data mapping and security constraints that make it difficult to carry out. Federated learning with a distributed learning architecture has changed this situation. By restricting the training process to participants’ local, federated learning addresses the model training needs of multiple data sources while better protecting data privacy. However, in real-world application scenarios, federated learning faces the need to achieve fairness in addition to privacy protection. In practice, it could happen that some federated learning participants with specific motives may short join the training process to obtain the current global model with only limited contribution to the federated learning whole, resulting in unfairness to the participants who previously participated in federated learning. We propose the FedACC framework with a server-initiated global model accuracy control method to address this issue. Besides measuring the accumulative contributions of newly joined participants and providing participants with a model with an accuracy that matches their contributions, the FedACC still guarantees the validity of participant gradients based on the accuracy-decayed model. Under the FedACC framework, users do not have access to the full version of the current global model early in their training participation. However, they must produce a certain amount of contributions before seeing the full-accuracy model. We introduce an additional differential privacy mechanism to protect clients’ privacy further. Experiments demonstrate that the FedACC could obtain about 10%–20% accuracy gain compared to the state-of-the-art methods while balancing the fairness, performance, and security of federated learning.

With the future 6G era, spiking neural networks (SNNs) can be powerful processing tools in various areas due to their strong artificial intelligence (AI) processing capabilities, such as biometric recognition, AI robotics, autonomous drive, and healthcare. However, within Cyber Physical System (CPS), SNNs are surprisingly vulnerable to adversarial examples generated by benign samples with human-imperceptible noise, this will lead to serious consequences such as face recognition anomalies, autonomous drive-out of control, and wrong medical diagnosis. Only by fully understanding the principles of adversarial attacks with adversarial samples can we defend against them. Nowadays, most existing adversarial attacks result in a severe accuracy degradation to trained SNNs. Still, the critical issue is that they only generate adversarial samples by randomly adding, deleting, and flipping spike trains, making them easy to identify by filters, even by human eyes. Besides, the attack performance and speed also can be improved further. Hence, Spike Probabilistic Attack (SPA) is presented in this paper and aims to generate adversarial samples with more minor perturbations, greater model accuracy degradation, and faster iteration. SPA uses Poisson coding to generate spikes as probabilities, directly converting input data into spikes for faster speed and generating uniformly distributed perturbation for better attack performance. Moreover, an objective function is constructed for minor perturbations and keeping attack success rate, which speeds up the convergence by adjusting parameters. Both white-box and black-box settings are conducted to evaluate the merits of SPA. Experimental results show the model's accuracy under white-box attack decreases by 9.2S%~31.1S% better than others, and average success rates are 74.87% under the black-box setting. The experimental results indicate that SPA has better attack performance than other existing attacks in the white-box and better transferability performance in the black-box setting,