Science China Information Sciences最新文献

Robust open-set recognition (OSR) performance has become a prerequisite for pattern recognition systems in real-world applications. However, the existing OSR methods are primarily implemented on the basis of single-modal perception, and their performance is limited when single-modal data fail to provide sufficient descriptions of the objects. Although multimodal data can provide more comprehensive information than single-modal data, the learning of decision boundaries can be affected by the feature representation gap between different modalities. To effectively integrate multimodal data for robust OSR performance, we propose logit prototype learning (LPL) with active multimodal representation. In LPL, the input multimodal data are transformed into the logit space, enabling a direct exploration of intermodal correlations without the impact of scale inconsistency. Then, the fusion weights of each modality are determined using an entropybased uncertainty estimation method. This approach realizes adaptive adjustment of the fusion strategy to provide comprehensive descriptions in the presence of external disturbances. Moreover, the single-modal and multimodal representations are jointly optimized interactively to learn discriminative decision boundaries. Finally, a stepwise recognition rule is employed to reduce the misclassification risk and facilitate the distinction between known and unknown classes. Extensive experiments on three multimodal datasets have been done to demonstrate the effectiveness of the proposed method.

In this study, we present a simple yet effective pyramid-resolution person restoration method for cross-resolution person re-identification. Our method involves a pyramid resolution restoration network that enhances pyramid resolution images, and utilizes feature distance fusion to leverage valuable and complementary information from these pyramid images. Extensive experiments demonstrate the effectiveness of our method on both real-world cross-resolution datasets and simulated datasets.

The robust finite-time synchronization control problem is investigated for master-slave networked nonlinear telerobotics systems (NNTSs) in this article. Although there have been some research achievements on finite-time control for the NNTSs, these studies are based on the strong assumptions of communication time delays or can only achieve finite-time bounded convergence even when the external forces are zero. Accordingly and in view of the importance of these issues, a novel robust composite learning adaptive control scheme rendering the finite-time master-slave synchronization is proposed in this paper. In particular, the influence of time delays on finite-time convergence of the system is analyzed by employing the multi-dimension finite-time small-gain framework. Meanwhile, in order to achieve accurate and fast estimation of uncertain parameters of the system, both the online historical and the instantaneous data of the estimation data are explored to derive the new parameter adaptive law under a more realizable interval-excitation (IE) condition. Therefore, the convergence of the position/force synchronization errors and the adaptive parameter estimation errors is obtained in finite time, and enhanced robustness of the closed-loop system will also be ensured. Finally, the superior performance of the proposed control algorithms is validated by numerical simulations and hardware experiments.

Recently, intensive attempts have been made to design robust models for fine-grained visual recognition, most notably are the impressive gains for training with noisy labels by incorporating a reweighting strategy into a meta-learning framework. However, it is limited to up or downweighting the contribution of an instance for label reweighting approaches in the learning process. To solve this issue, a novel noise-tolerant method with auxiliary web data is proposed. Specifically, first, the associations made from embeddings of well-labeled data with those of web data and back at the same class are measured. Next, its association probability is employed as a weighting fusion strategy into angular margin-based loss, which makes the trained model robust to noisy datasets. To reduce the influence of the gap between the well-labeled and noisy web data, a bridge schema is proposed via the corresponding loss that encourages the learned embeddings to be coherent. Lastly, the formulation is encapsulated into the meta-learning framework, which can reduce the overfitting of models and learn the network parameters to be noise-tolerant. Extensive experiments are performed on benchmark datasets, and the results clearly show the superiority of the proposed method over existing state-of-the-art approaches.

In this paper, we explore state-based potential games using the semi-tensor product of matrices. First, applying the potential equation, we derive both a necessary and sufficient condition as well as a sufficient condition to verify whether a state-based game qualifies as a potential game. Next, we present two static equivalence conditions of state-based potential games. We then delve into dynamic equivalence. We propose a criterion that allows us to identify state-based games that are dynamically equivalent to state-based potential games and share similar dynamic properties. Ultimately, we introduce the concept of state-based networked evolutionary games. We provide a necessary and sufficient condition to ensure that a state-based networked evolutionary game can be classified as a state-based potential game.

Semimetallic bismuth (Bi) is one of the most effective strategies for reducing the contact resistance of two-dimensional transition metal dichalcogenide field effect transistors (FETs). However, the low melting point of Bi contact (271.5° C) limits its reliable applications. In this study, we demonstrated that the temperature stability of Bi-contacted electrodes could be improved by inserting a high-melting point semimetallic antimony (Sb) between the Bi contacting layer and the gold (Au) capping layer. The proposed Bi/Sb/Au contact electrodes tended to form a metal mixture with a continuous surface during the heating process (Voids appeared on the surface of the Bi/Au contact electrodes after heating at 120° C). Because of the improved contacting layer formed by the semimetal Bi/Sb alloy, the fabricated Bi/Sb/Au-contacted molybdenum sulfide (MoS₂) FETs with different gate lengths demonstrated higher on-state current stability after heating treatment than the Bi/Au contact. Because of the Bi/Sb/Au contact and poly (methyl methacrylate) package, the MoS₂ FETs demonstrated time stability of at least two months from the almost unchanged transfer characteristics. The electrical stability indicates that the insertion of semimetallic Sb is a promising technology for reliable Bi-based contact.

Deep reinforcement learning models are vulnerable to adversarial attacks that can decrease the cumulative expected reward of a victim by manipulating its observations. Despite the efficiency of previous optimization-based methods for generating adversarial noise in supervised learning, such methods might not achieve the lowest cumulative reward since they do not generally explore the environmental dynamics. Herein, a framework is provided to better understand the existing methods by reformulating the problem of adversarial attacks on reinforcement learning in the function space. The reformulation approach adopted herein generates an optimal adversary in the function space of targeted attacks, repelling them via a generic two-stage framework. In the first stage, a deceptive policy is trained by hacking the environment and discovering a set of trajectories routing to the lowest reward or the worst-case performance. Next, the adversary misleads the victim to imitate the deceptive policy by perturbing the observations. Compared to existing approaches, it is theoretically shown that our adversary is strong under an appropriate noise level. Extensive experiments demonstrate the superiority of the proposed method in terms of efficiency and effectiveness, achieving state-of-the-art performance in both Atari and MuJoCo environments.

Autonomous robots have garnered extensive utilization in diverse fields. Among the critical concerns for autonomous systems, path planning holds paramount importance. Notwithstanding considerable efforts in its development over the years, path planning for autonomous systems continues to grapple with challenges related to low planning efficiency and inadequate obstacle avoidance response in a timely manner. This study proposes a novel and systematic solution to the path planning problem within intricate office buildings. The solution consists of a global planner and a local planner. To handle the global planning aspect, an adaptive clustering-based dynamic programming rapidly exploring random tree (ACDP-RRT) algorithm is proposed. ACDP-RRT effectively identifies obstacles on the map by leveraging geometric features. These obstacles are then represented as a collection of sequentially arranged convex polygons, optimizing the sampling region and significantly enhancing sampling efficiency. For local planning, a network decoupling actor-critic (ND-AC) algorithm is employed. The proposed ND-AC simplifies the local planner design process by integrating planning and control loops into a neural network (NN) trained via an end-to-end model-free deep reinforcement learning (DRL) framework. Moreover, the adoption of network decoupling (ND) techniques leads to an improved obstacle avoidance success rate when compared to conventional actor-critic (AC)-based methods. Extensive simulations and experiments are conducted to demonstrate the effectiveness and robustness of the proposed approach.

Next-generation wireless network aims to support low-latency, high-speed data transmission services by incorporating artificial intelligence (AI) technologies. To fulfill this promise, AI-based network traffic prediction is essential for pre-allocating resources, such as bandwidth and computing power. This can help reduce network congestion and improve the quality of service (QoS) for users. Most studies achieve future traffic prediction by exploiting deep learning and reinforcement learning, to mine spatio-temporal correlated variables. Nevertheless, the prediction results obtained only by the spatio-temporal correlated variables cannot reflect real traffic changes. This phenomenon prevents the true prediction variables from being inferred, making the prediction algorithm perform poorly. Inspired by causal science, we propose a novel network traffic prediction method based on self-supervised spatio-temporal causal discovery (SSTCD). We first introduce the Granger causal discovery algorithm to build a causal graph among prediction variables and obtain spatio-temporal causality in the observed data, which reflects the real reasons affecting traffic changes. Next, a graph neural network (GNN) is adopted to incorporate causality for traffic prediction. Furthermore, we propose a self-supervised method to implement causal discovery to to address the challenge of lacking ground-truth causal graphs in the observed data. Experimental results demonstrate the effectiveness of the SSTCD method.

Optical memory integrates the function of optical sensing in memory devices, remarkably promoting the interconnection between sensory and memory terminals. Silicon charge-coupled photodetectors and floating gate memory have been widely used in imaging and storage technologies, respectively. However, the heterogeneous integration of the two devices requires technological innovation and complex electrical connections. In this work, we adopt a three-dimensional layer stacking method to design a novel optical memory device. On the top of Si charge-coupled photodetectors, we successively deposit two-dimensional graphene, hexagonal boron nitride, and molybdenum disulfide as a floating gate layer, a tunneling layer, and a readout layer, respectively. By applying a gate bias on lightly doped Si, a deep depletion layer is formed with a high voltage potential drop. Under dark conditions, the depletion layer cannot be filled, and the electric field across the h-BN tunnel barrier is relatively small. Under light irradiation, the deep depletion layer is gradually filled, and the h-BN tunneling layer withstands the increasing electric field, resulting in charge storage in the floating gate layer. Based on this mechanism, the device exhibits a gate voltage-dependent operation mode, including an integrated optical sensing-memory mode and an electrically driven storage mode. Under moderate gate voltage, the device can effectively detect the optical information with varied intensity and store the optical information in the floating gate, displaying optically controlled memory characteristics. Our work demonstrates a compact device structure for optical memory and displays excellent optically controlled memory performance, which can be applied in artificial vision systems.