As the brain-like intelligence develops rapidly, it is urgent to design a more convenient and efficient control framework to cope with the challenge of processing multisensory signals in parallel. Therefore, a multisensory memristive circuit with dual adaptive, parallel processing, and multilevel reinforcement features is proposed. The circuit is mainly composed of modules for receptors, STM and LTM, attention, environmental monitoring and mutual associative memory. Automatic encoding of different sensorial signals is realised by the receptor modules. Dual adaptive regulation of the internal associative memory and external environmental changes on the circuit is implemented by modules of attention and environmental monitoring. Multilevel reinforcement memory is achieved through the interconnection of multiple dimensional features of the same objects. The process of encoding transformation of stimuli, experience memory, and feedback learning is automatically achieved in the brain-inspired neural network structure, which avoids the problems such as encoding difficulties during the conversion of the operating objects, and enables the realization of more brain-like intelligence. The circuit is applied to gripping and recognizing in robotic arms and the scenario memory of different production lines is simulated, which is promising for application in automated factories.

Spatiotemporal Graph Learning (SGL) under Zero-Inflated Distribution (ZID) is crucial for urban risk management tasks, including crime prediction and traffic accident profiling. However, SGL models are vulnerable to adversarial attacks, compromising their practical utility. While adversarial training (AT) has been widely used to bolster model robustness, our study finds that traditional AT exacerbates performance disparities between majority and minority classes under ZID, potentially leading to irreparable losses due to underreporting critical risk events. In this paper, we first demonstrate the smaller top-k gradients and lower separability of minority class are key factors contributing to this disparity. To address these issues, we propose MinGRE, a framework for Minority Class Gradients and Representations Enhancement. MinGRE employs a multi-dimensional attention mechanism to reweight spatiotemporal gradients, minimizing the gradient distribution discrepancies across classes. Additionally, we introduce an uncertainty-guided contrastive loss to improve the inter-class separability and intra-class compactness of minority representations with higher uncertainty. Extensive experiments demonstrate that the MinGRE framework not only significantly reduces the performance disparity across classes but also achieves enhanced robustness compared to existing baselines. These findings underscore the potential of our method in fostering the development of more equitable and robust models.

Vision-Language Models (VLMs) play a crucial role in the advancement of Artificial General Intelligence (AGI). As AGI rapidly evolves, addressing security concerns has emerged as one of the most significant challenges for VLMs. In this paper, we present extensive experiments that expose the vulnerabilities of conventional adaptation methods for VLMs, highlighting significant security risks. Moreover, as VLMs grow in size, the application of traditional adversarial adaptation techniques incurs substantial computational costs. To address these issues, we propose a parameter-efficient adversarial adaptation method called AdvLoRA based on Low-Rank Adaptation. We investigate and reveal the inherent low-rank properties involved in adversarial adaptation for VLMs. Different from LoRA, we enhance the efficiency and robustness of adversarial adaptation by introducing a novel reparameterization method that leverages parameter clustering and alignment. Additionally, we propose an adaptive parameter update strategy to further bolster robustness. These innovations enable our AdvLoRA to mitigate issues related to model security and resource wastage. Extensive experiments confirm the effectiveness and efficiency of AdvLoRA.

Knowledge distillation is widely used as an effective model compression technique to improve the performance of small models. Most of the current researches on knowledge distillation focus on the algorithmic level and ignore the potential benefits of hardware implementation. In this paper, a multi-loss knowledge distillation online training circuit based on memristor crossbar array is designed, which can improve the inference efficiency and reduce the power consumption of deep learning models on edge devices. The circuit is able to process data in real time, and it can be used to handle stuck-at-faults (SAF) caused by factors such as manufacturing defects in the memristor. Moreover, a fault detection scheme with low time cost is proposed in order to address the low efficiency of stuck-at-fault detection in memristor crossbar arrays. The scheme is combined with a self-compensating pruning method and knowledge distillation online training mechanism, which significantly improves the model training and inference capability of the circuit under fault conditions. Experimental results show that the multi-loss knowledge distillation online training improves the accuracy by 4.15% and 63.48% respectively in two models compared with traditional training schemes. The fault-tolerance scheme reduces the power consumption of the memristor crossbar arrays by 41.2% and 72.6% respectively on the two models, demonstrating its potential and advantages in edge computing.

Accurate stock market prediction is crucial for investors seeking significant profits. With increased economic activity, various interrelations between listed companies have become important for accurate predictions. These relations can be represented as complex financial networks, aiding the development of effective graph neural network (GNN) prediction methods. However, current GNN-based methods for stock prediction typically rely on a single static network representation, which fails to capture the dynamic and multifaceted relationships inherent in financial markets. In this article, we propose the multilayer spatial–temporal graph neural network (MST-GNN) to model the complex and evolving interactions between stocks. The MST-GNN framework incorporates a novel spatial–temporal cross-layer high-order fusion mechanism, which includes two key components: spatial–temporal neighborhood aggregation and cross-layer high-order feature fusion. These components enable the model to effectively capture both the temporal evolution and cross-network feature interactions of stocks. Our extensive experiments on four stock networks from the China A-share market demonstrate that MST-GNN significantly outperforms existing GNN-based methods on stock price trend classification and return ranking tasks.

With the rise and prevalence of social bots, their negative impacts on society are gradually recognized, prompting research attention to effective detection and countermeasures. Recently, graph neural networks (GNNs) have flourished and have been applied to social bot detection research, improving the performance of detection methods effectively. However, existing GNN-based social bot detection methods often fail to account for the heterogeneous associations among users within social media contexts, especially the heterogeneous integration of social bots into human communities within the network. To address this challenge, we propose a heterogeneous compatibility perspective for social bot detection, in which we preserve more detailed information about the varying associations between neighbors in social media contexts. Subsequently, we develop a compatibility-aware graph neural network (CGNN) for social bot detection. CGNN consists of an efficient feature processing module, and a lightweight compatibility-aware GNN encoder, which enhances the model’s capacity to depict heterogeneous neighbor relations by emulating the heterogeneous compatibility function. Through extensive experiments, we showed that our CGNN outperforms the existing state-of-the-art (SOTA) method on three commonly used social bot detection benchmarks while utilizing only about 2% of the parameter size and 10% of the training time compared with the SOTA method. Finally, further experimental analysis indicates that CGNN can identify different edge categories to a significant extent. These findings, along with the ablation study, provide strong evidence supporting the enhancement of GNN’s capacity to depict heterogeneous neighbor associations on social media bot detection tasks.

Due to the implementation of national security-related laws and regulations, data privacy protection and ownership issues have attracted much attention, meanwhile, the rise of technologies such as 5G, IOT, big data, and edge computing has promoted the digital transformation of data as a factor of production to empower social governance. At present, traditional machine learning still uses the method of data-centered large models for training and reasoning. This method brings about data fragmentation and island distribution and other problems, which have become the key problems restricting the popularization of Artificial Intelligence (AI) technology applications. In this paper, we explore a federated learning model for user complaint warning algorithm based on big data and other enterprise side data. The experimental result also shows that the prediction accuracy of the federated learning model and the traditional logistic regression model are within an acceptable range on the premise of ensuring user privacy and data security.

The robustness of machine learning (ML) models has gained much attention along with their wide application on various safety-required Industrial Internet of Things (IIoT) paradigms. Researchers found that some specific attacks added on sensor measurements can maliciously disturb IIoT monitors that are designed using ML architectures. The Traditional detection methods could judge whether the measurements are attacked to prevent the failure of monitors. Unfortunately, recent works argue that the commonly used detection methods could be circumvented through adaptive attacks that could acquire the mechanism of detectors; they could not truly enhance the robustness of ML models. Instead, general robust mechanisms should be performed to authentically enhance the robustness of models against any potential attacks with specific restrictions. On the basis of the above argument, we design a robust condition monitor for predicting the fault condition of IIoT systems using the adversarial training technique called robust temporal convolutional network (RTCN). The model is designed to be formally robust to attacks with restricted magnitude. The temporal convolutional network (TCN) is employed to design the base structure of the monitor. TCN can capture temporal information from sensors to enhance the feature extraction performance of models. We also present a novel false data injection (FDI) attack-generating method that utilizes the conception of adversarial perturbations to disturb well-trained monitors. The experimental results verify the efficiency of feature extraction performance of our model from IIoT systems. Furthermore, adversarial training mechanism through a min–max manner could effectively improve the reliability of ML-based IIoT monitors against strong FDI attacks.

The multimodality nature of web data has necessitated complex multimodal information retrieval for a wide range of web applications. Deep neural networks (DNNs) have been widely employed to extract semantic features from raw samples to improve retrieval accuracy. In addition, hashing is widely used to improve computational and storage efficiency. As such, deep hashing frameworks have been applied for multimodal retrieval tasks. However, there is still a great recognitive gap between primate brain structure-inspired DNNs and humans. On computer vision tasks, well-crafted DNN models can be easily defeated by invisible small attacks, and this phenomenon indicates a large recognition gap between DNN models and humans. Recently, adversarial defense methods have been shown to improve the human–machine recognition alignment in several classification tasks. However, the robustness problem on the retrieval tasks, especially on the deep hashing-based multimodal retrieval models, is still not well studied. Therefore, in this article, we present an adversarially robust training mechanism to improve model robustness for the purpose of human–machine recognition alignment on retrieval tasks. Through extensive experimental results on several social multimodal retrieval benchmarks, we show that the robust training hashing framework proposed can mitigate the recognition gap on retrieval tasks. Our study highlights the necessity of robustness enhancement on deep hashing models.

The wide application of deep neural networks (DNNs) has significantly improved the performance of hashing models on multimodal retrieval issues. DNN-based deep models can automatically learn semantic features from raw data to make human-level decisions. However, the superior generalization leads to potential privacy leakage risks. Strong DNN-based retrieval models enable malicious crawlers to search for nontag private information based on semantic similarity matching. Hence, executing effective privacy protection mechanisms against those retrieval software is essential for reliable social website construction. In this article, we propose a retrieval task-based adversarial perturbation generation method called Hashing Fake to meet this request. Specifically, DNNs are recently found to be vulnerable to a specific set of attacks called adversarial perturbations, which denote some magnitude-restricted signals added on objective samples to misguide well-crafted DNN models, and perturbations’ magnitudes are small enough that will not induce humans’ perception. Moreover, since existing adversarial perturbation generation methods are designed for supervised tasks, Hashing Fake constructs a differential approximation substitution for perturbation production on unsupervised retrieval tasks. Through extensive experiments on several deep retrieval benchmarks, we demonstrate that well-crafted perturbations using Hashing Fake can effectively misguide objective models’ recognitions to make false predictions. The added norm-restricted perturbations on objective samples will not alter humans’ perception; hence, Hashing Fake can be applied on real-world social websites to protect subscribers’ privacy against malicious retrieval software.

Investment decisions and risk management require understanding the time-varying dependencies between stocks. Graph-based learning systems have emerged as a promising approach for predicting stock prices by leveraging interfirm relationships. However, existing methods rely on a static stock graph predefined from finance domain knowledge and large-scale data engineering, which overlooks the dynamic dependencies between stocks. In this article, we present a novel framework called graph evolution recurrent unit (GERU), which uses a dynamic graph neural network to automatically learn the evolving dependencies from historical stock features, leading to better predictions. Our approach consists of three parts: first, we develop an adaptive dynamic graph learning (ADGL) module to learn latent dynamic dependencies from stock time series. Second, we propose a clustered ADGL (clu-ADGL) to handle large-scale time series by reducing time and memory complexity. Third, we combine the ADGL/clu-ADGL with a graph-gated recurrent unit to model the temporal evolutions of stock networks. Extensive experiments on real-world datasets show that our proposed methods outperform existing methods in predicting stock movements, capturing meaningful dynamic dependencies and temporal evolution patterns from the financial market, and achieving outstanding profitability in portfolio construction.

The aim of cross-modal retrieval is to search for flexible results across different types of multimedia data. However, the labeled data is usually limited and not well paired with different modalities in practical applications. These issues are not well addressed in the existing works, which cannot consider the semantic information about unlabeled and unpaired data, synchronously. Self-training is a well-known strategy to handle semi-supervised problems. Motivated by the self-training, this paper proposes a self-training-based cross-modal hashing framework (STCH) to tackle the semi-supervised and semi-paired challenges. In the framework, graph neural networks are used to capture potential intra-modality and inter-modality similarities to produce pseudo labels. Then the inconsistent pseudo labels of different modalities are refined with a heuristic filter to enhance the model robustness. To train STCH, we propose an alternating learning strategy to conduct the self-train by predicting pseudo labels during the training procedure, which can be seamlessly incorporated into semi-supervised and supervised learning. In this way, the proposed method can leverage sufficient semantic information to enhance the semi-supervised effect and address the semi-paired problem. Experiments on the real-world datasets demonstrate that our approach outperforms related methods on hash cross-modal retrieval.

The rapid development of Internet and online data explosions elicit strong aspirations for users to search for semantic relevant information based on available samples. While the data online are always released with different modalities like images, videos or texts, effective retrieval models should discover latent semantic information with different structures. Recently, the state-of-the-art deep cross modal retrieval frameworks have effectively enhanced the performance on commonly-used platforms using the deep neural networks (DNNs). Yet DNNs have been verified to be easily misguided by small perturbations, and there are already several attack generation methods proposed on DNN-based models for real-world tasks, but they are all focused on supervised tasks like classification or object recognition. To effectively evaluate the robustness of deep cross-modal retrieval frameworks, in this paper, we propose a retrieval-based adversarial perturbation generation method, and demonstrate that our perturbation could effectively attack the state-of-the-art deep cross-modal and single image retrieval hashing models.

Label imbalance and data missing are two major challenges in the problem of credit card fraud detection. However, existing matrix completion algorithms are generally difficult and cannot be easily applied to real-world credit card fraud detection since the scale of the normally used dataset is oversized. In this paper, we develop a spectral regularization algorithm to complete the large-scale sparse matrices, and further utilize an over-sampling algorithm to tackle the problem of the imbalance between positive and negative samples. Experimental results on a real-world dataset demonstrate that our model can outperform the state-of-the-art baseline methods. The proposed method could also be extended to other large-scale scenarios where data is missing or labels are imbalanced.

With the rampaging of Coronavirus disease 2019 (COVID-19) across the world, analyzing the dynamic characteristics and understanding the evolutionary patterns of clusters are becoming even more crucial for people and policymakers to make timely responses for avoiding injury caused by COVID-19. To solve the scarcity of the fine-grained spatiotemporal data, we construct a novel dataset about the spread of patients during the resurgent period of the COVID-19 epidemic at the Xinfadi Market in Beijing. Leveraging our self-build dataset, we analyze the evolutionary characteristics of the cluster of COVID-19 under anti-contagion policies and obtained some remarkable evolution patterns. These findings can provide significant insights for policymakers and researchers to understand the evolutionary characteristics regarding the cluster of COVID-19 and deploy effective anti-contagion policies.