Journal of Information and Intelligence最新文献

The development of information technology has propelled technological reform in artificial intelligence (AI). To address the needs of diversified and complex applications, AI has been increasingly trending towards intelligent, collaborative, and systematized development across different levels and tasks. Research on intelligent, collaborative and systematized AI can be divided into three levels: micro, meso, and macro. Firstly, the micro-level collaboration is illustrated through the introduction of swarm intelligence collaborative methods related to individuals collaboration and decision variables collaboration. Secondly, the meso-level collaboration is discussed in terms of multi-task collaboration and multi-party collaboration. Thirdly, the macro-level collaboration is primarily in the context of intelligent collaborative systems, such as terrestrial-satellite collaboration, space-air-ground collaboration, space-air-ground-air collaboration, vehicle-road-cloud collaboration and end-edge-cloud collaboration. Finally, this paper provides prospects on the future development of relevant fields from the perspectives of the micro, meso, and macro levels.

In this paper, we study turbo codes from the digital signal processing point of view by defining turbo codes over the complex field. It is known that iterative decoding and interleaving between concatenated parallel codes are two key elements that make turbo codes perform significantly better than the conventional error control codes. This is analytically illustrated in this paper. We show that the decoded noise mean power in the iterative decoding decreases when the number of iterations increases, as long as the interleaving decorrelates the noise after each iterative decoding step. An analytic decreasing rate and the limit of the decoded noise mean power are given. The limit of the decoded noise mean power of the iterative decoding of a turbo code with two parallel codes with their rates less than 1/2 is one third of the noise power before the decoding, which can not be achieved by any non-turbo codes with the same rate. From this study, the role of designing a good interleaver can also be clearly seen.

Autoencoder-based rating prediction methods with external attributes have received wide attention due to their ability to accurately capture users' preferences. However, existing methods still have two significant limitations: i) External attributes are often unavailable in the real world due to privacy issues, leading to low quality of representations; and ii) existing methods lack considering complex associations in users' rating behaviors during the encoding process. To meet these challenges, this paper innovatively proposes an inherent-attribute-aware dual-graph autoencoder, named IADGAE, for rating prediction. To address the low quality of representations due to the unavailability of external attributes, we propose an inherent attribute perception module that mines inductive user active patterns and item popularity patterns from users' rating behaviors to strengthen user and item representations. To exploit the complex associations hidden in users’ rating behaviors, we design an encoder on the item-item co-occurrence graph to capture the co-occurrence frequency features among items. Moreover, we propose a dual-graph feature encoder framework to simultaneously encode and fuse the high-order representations learned from the user-item rating graph and item-item co-occurrence graph. Extensive experiments on three real datasets demonstrate that IADGAE is effective and outperforms existing rating prediction methods, which achieves a significant improvement of 4.51%∼41.63 ​% in the RMSE metric.

Radio frequency (RF)-based drone identification technologies have the advantages of long effective distances and low environmental dependence, which has become indispensable for drone surveillance systems. However, since drones operate in unlicensed frequency bands, a large number of co-frequency devices exist in these bands, which brings a great challenge to traditional signal identification methods. Deep learning techniques provide a new approach to complete end-to-end signal identification by directly learning the distribution of RF data. In such scenarios, due to the complexity and high dynamics of the electromagnetic environments, a massive amount of data that can reflect the various propagation conditions of drone signals is necessary for a robust neural network (NN) for identifying drones. In reality, signal acquisition and labeling that meet the above requirements are too costly to implement. Therefore, we propose a virtual electromagnetic environment modeling based data augmentation (DA) method to improve the diversity of drone signal data. The DA method focuses on simulating the spectrograms of drone signals transmitted in real-world environments and randomly generates extra training data in each training epoch. Furthermore, considering the limited processing capability of RF receivers, we modify the original YOLOv5s model to a more lightweight version. Without losing the identification performance, more hardware-friendly designs are applied and the number of parameters decreases about 10-fold. For performance evaluation, we utilized a universal software radio peripheral (USRP) X310 platform to collect RF signals of four drones in an anechoic chamber and a practical wireless scenario. Experiment results reveal that the NN trained with augmented data performs as well as that trained with practical data in the complex electromagnetic environment.

An intelligent liquid classification system based on 77 ​GHz ​millimeter wave radar and convolution neural network are proposed in this paper. The data are collected by the AWR1843 radar platform and processed by the neural network on the host PC in real-time. The doppler heatmap generated by radar signal processing is tried for the first time as the input of the system. The information carried by the heatmap in 2 dimensions is analyzed and the reason why the doppler heatmap could be used for classification is explained. The feasible experiment proved that the proposed method can successfully classify 8 kinds of common liquid with high accuracy. The result of the experiment is explained and the limitations of the experiment are discussed. It can be drawn that the combination of FMCW millimeter wave radar and convolution neural network is a method with great potential for liquid classification. The advantages of real time, non-invasive and unilateral measurement can also be used for the detection of dangerous liquids.

To solve the data island problem, federated learning (FL) provides a solution paradigm where each client sends the model parameters but not the data to a server for model aggregation. Peer-to-peer (P2P) federated learning further improves the robustness of the system, in which there is no server and each client communicates directly with the other. For secure aggregation, secure multi-party computing (SMPC) protocols have been utilized in peer-to-peer manner. However, the ideal SMPC protocols could fail when some clients drop out. In this paper, we propose a robust peer-to-peer learning (RP2PL) algorithm via SMPC to resist clients dropping out. We improve the segment-based SMPC protocol by adding a check and designing the generation method of random segments. In RP2PL, each client aggregates their models by the improved robust secure multi-part computation protocol when finishes the local training. Experimental results demonstrate that the RP2PL paradigm can mitigate clients dropping out with no significant degradation in performance.

The joint adoption of sub-6GHz and millimeter wave (mmWave) technology can prevent the blind spots of coverage, enabling comprehensive coverage while realizing high-speed communication rate. According to the sensitivity of mmWave, base stations should be more densely deployed, which is not well described by existing Poisson hole process (PHP) and the Poisson point process (PPP) models. This paper establishes a sub-6GHz and mmWave hybrid heterogeneous cellular network based on the modified Poisson hole process (MPHP). In our proposed model, the sub-6GHz base stations follow the PPP, and the mmWave base stations (MBSs) follow MPHP distribution. The expressions of the coverage probability are derived by using the interference calculation method of integrating the nearest sector exclusion area. Our theoretical analysis has been verified through simulation results, suggesting that the increase in the cell radius decreases the coverage probability of signal-to-interference-plus-noise ratio (SINR), whereas the increase in the sector parameter has the opposite effect. The variation of sub-6GHz base stations (SBSs) density imposes more significant impact than the MBSs on the SINR coverage probability. In addition, the decrease in MBSs density will reduce the average bandwidth allocated to the user equipment (UE), thus reducing the rate coverage probability.

Facial emotion have great significance in human-computer interaction, virtual reality and people's communication. Existing methods for facial emotion privacy mainly concentrate on the perturbation of facial emotion images. However, cryptography-based perturbation algorithms are highly computationally expensive, and transformation-based perturbation algorithms only target specific recognition models. In this paper, we propose a universal feature vector-based privacy-preserving perturbation algorithm for facial emotion. Our method implements privacy-preserving facial emotion images on the feature space by computing tiny perturbations and adding them to the original images. In addition, the proposed algorithm can also enable expression images to be recognized as specific labels. Experiments show that the protection success rate of our method is above 95% and the image quality evaluation degrades no more than 0.003. The quantitative and qualitative results show that our proposed method has a balance between privacy and usability.

Graph convolutional networks (GCNs) have been successfully applied to node representation learning in various real-world applications. However, the performance of GCNs drops rapidly when the labeled data are severely scarce, and the node features are prone to being indistinguishable with stacking more layers, causing over-fitting and over-smoothing problems. In this paper, we propose a simple yet effective contrastive semantic calibration for graph convolution network (CSC-GCN), which integrates stochastic identity aggregation and semantic calibration to overcome these weaknesses. The basic idea is the node features obtained from different aggregation operations should be similar. Toward that end, identity aggregation is utilized to extract semantic features from labeled nodes, while stochastic label noise is adopted to alleviate the over-fitting problem. Then, contrastive learning is employed to improve the discriminative ability of the node features, and the features from different aggregation operations are calibrated according to the class center similarity. In this way, the similarity between unlabeled features and labeled ones from the same class is enhanced while effectively reducing the over-smoothing problem. Experimental results on eight popular datasets show that the proposed CSC-GCN outperforms state-of-the-art methods on various classification tasks.