Xiangyu Liu

Many real-world problems involve hierarchical multi-objective optimization over coupled sequential decisions. A common strategy is to combine heuristic planning with deep reinforcement learning (DRL). While existing hybrid methods often show strong empirical performance, the coupled closed-loop learning dynamics between upper-tier planning and lower-tier control are typically analyzed only empirically, and explicit convergence guarantees for the integrated scheme remain limited. To address this gap, we propose a heuristic-supervised-DRL (HSD) framework that tightly couples (i) a heuristic planner for upper-tier decision-making, (ii) a DRL agent for lower-tier execution, and (iii) an online supervised predictor that serves as an adaptive bridge between planning and execution. The key novelty of HSD lies in this closed-loop architecture and its accompanying theoretical treatment. By formulating the coupled updates as a two-timescale stochastic approximation process, we show that, under standard conditions, the supervised predictor tracks a quasi-stationary regression target and the overall joint process converges almost surely to an asymptotically stable equilibrium. We further analyze robustness under approximate planning errors. As a case study, we instantiate HSD in a multi-UAV-assisted mobile edge computing system. Experimental results show that the proposed framework consistently outperforms representative baselines in key performance metrics, demonstrating both its practical effectiveness and its value as a principled framework for hierarchical optimization in dynamic environments.

Distributed Denial of Service (DDoS) attacks have always been a major concern in the security field. With the release of malware source codes such as BASHLITE and Mirai, Internet of Things (IoT) devices have become the new source of DDoS attacks against many Internet applications. Although there are many datasets in the field of IoT intrusion detection, such as Bot-IoT, Constrained Application Protocol–Denial of Service (CoAP-DoS), and LATAM-DDoS-IoT (some of the names of DDoS datasets), which mainly focus on DDoS attacks, the datasets describing new IoT DDoS attack scenarios are extremely rare, and only N-BaIoT and IoT-23 datasets used IoT devices as DDoS attackers in the construction process, while they did not use Internet applications as victims either. To supplement the description of the new trend of DDoS attacks in the dataset, we built an IoT environment with mainstream DDoS attack tools such as Mirai and BASHLITE being used to infect IoT devices and implement DDoS attacks against WEB servers. Then, data aggregated into a dataset named MBB-IoT were captured at WEB servers and IoT nodes. After the MBB-IoT dataset was split into a training set and a test set, it was applied to the training and testing of the Random Forests classification algorithm. The multi-class classification metrics were good and all above 90%. Secondly, in a cross-evaluation experiment based on Support Vector Machine (SVM), Light Gradient Boosting Machine (LightGBM), and Long Short Term Memory networks (LSTM) classification algorithms, the training set and test set were derived from different datasets (MBB-IoT or IoT-23), and the test performance is better when MBB-IoT is used as the training set.

With the large-scale use of the Internet of Things, security issues have become increasingly prominent. The accurate detection of network attacks in the IoT environment with limited resources is a key problem that urgently needs to be solved. The intrusion detection system based on network traffic characteristics is one of the solutions for IoT security. However, the intrusion detection system has the problem of a large number of traffic features, which makes training and detection slow. Aiming at this problem, this work proposes a feature selection method based on a genetic algorithm. The experiments performed on the Bot-IoT botnet detection dataset show that this method successfully selects 6 features from the original 40 features, with a detection accuracy of 99.98% and an F1-score of 99.63%. Compared with other methods and without feature selection, this method has advantages in training time and detection accuracy.

With the large-scale application of the Internet of Things (IoT), security issues have become increasingly prominent. Device identification is an effective way to secure IoT environment by quickly identifying the category or model of devices in the network. Currently, the passive fingerprinting method used for IoT device identification based on network traffic flow mostly focuses on protocol features in packet headers but does not consider the direction and length of packet sequences. This paper proposes a device identification method for the IoT based on directional packet length sequences in network flows and a deep convolutional neural network. Each value in a packet length sequence represents the size and transmission direction of the corresponding packet. This method constructs device fingerprints from packet length sequences and uses convolutional layers to extract deep features from the device fingerprints. Experimental results show that this method can effectively recognize device identity with accuracy, recall, precision, and f1-score over 99%. Compared with methods using traditional machine learning and feature extraction techniques, our feature representation is more intuitive, and the classification model is effective.

With the increasing integration of mobile Internet into people’s daily life, generally a user will register several different network applications at the same time. Therefore, there are many virtual identities belonging to one person on the Internet, and the similarity analysis of cross-platform network identities is of great significance in the field of network security. This paper studied the Chinese user nicknames and virtual identity recognition in domestic social networking. Considering that account nicknames, to some extent, can reflect the characteristics of the account owners’ naming habits and preferences, and that the information of nickname is easier to obtain than other registration information, we collected the nickname information of users who registered on three application platforms: WeChat, Weibo and Alipay. At the same time, according to the characteristics of account nicknames and their probability distribution, we determined the characteristic indicators that can be used to calculate the similarity of nicknames. Finally, this paper optimized the Jaro distance and Jaro-Winkler distance algorithms, and proposed an identity algorithm suitable for domestic social networks, especially calculating the similarity of Chinese nicknames, and verified the effectiveness of the algorithm on the basis of large-scale real data.