Zhijun Yang

Monitoring computing and communication systems often involves reconstructing transient queue states from measurements that are sparse, noisy, and only indirectly informative about the underlying service process. This paper studies non-stationary asynchronous multi-server polling systems with gated service and develops a physics-informed grey-box framework for performance reconstruction. The framework couples a bidirectional long short-term memory traffic encoder with a physics-constrained neural approximator, so that observable queue trajectories and latent server-allocation profiles can be recovered jointly. The training objective combines data fidelity with fluid conservation, gated-service consistency, and resource-capacity constraints. On a discrete-event simulation benchmark covering balanced, asymmetric, and bursty traffic regimes, the proposed method achieves strong reconstruction accuracy, and multi-seed reruns indicate that the main scenario-wise trends are reproducible. In densely observed regimes, purely data-driven interpolators can attain lower pointwise error. Under degraded monitoring with sparse high-noise observations, Gaussian process regression remains strongest on pure interpolation metrics, whereas the proposed framework is more stable than the LSTM baseline while retaining explicit dynamical consistency and interpretable allocation dynamics. These results position the method as a performance-evaluation tool for monitored polling systems in which prediction accuracy, physical plausibility, and latent resource interpretation are all relevant.

Knowledge tracing (KT) is a critical task in educational data mining, aiming to infer students’ mastery levels of knowledge points using observable historical interaction data and related exercise information. However most of them only focus on local features, neglecting the utilization of students’ overall learning ability features. To address this issue, we propose a novel KT model: Global Feature-guided Knowledge Tracing (GFKT). This model leverages students’ historical interaction data to extract global features for guiding the training process, thereby improving the model’s predictive capability. Specifically, (i) we design a global feature module to obtain students’ overall learning abilities at the current moment. It constructs a data vector of ability values by calculating the difference between students’ correct and incorrect response rates for each knowledge point, collecting this set of ability values as global features, (ii) we utilize Recurrent Neural Network to extract local features from students’ exercise sequences, and propose a joint loss function that combines these local features with global features to train and optimize the model’s performance. Extensive experiments on multiple real-world public datasets, GFKT demonstrates superior predictive performance compared to state-of-the-art KT methods.

With the Internet of Things (IoT) development, there is an increasing demand for multi-service scheduling for Mobile Edge Computing (MEC). We propose using polling for scheduling in edge computing to accommodate multi-service scheduling methods better. Given the complexity of asymmetric polling systems, we have used an information-theoretic approach to analyse the model. Firstly, we propose an asymmetric two-level scheduling approach with priority based on a polling scheduling approach. Secondly, the mathematical model of the system in the continuous time state is established by using the embedded Markov chain theory and the probability-generating function. By solving for the probability-generating function’s first-order partial and second-order partial derivatives, we calculate the exact expressions of the average queue length, the average polling period, and the average delay with an approximate analysis of periodic query way. Finally, we design a simulation experiment to verify that our derived parameters are correct. Our proposed model can better differentiate priorities in MEC scheduling and meet the needs of IoT multi-service scheduling.

A continuous-time exhaustive-limited (K = 2) two-level polling control system is proposed to address the needs of increasing network scale, service volume and network performance prediction in the Internet of Things (IoT) and the Long Short-Term Memory (LSTM) network and an attention mechanism is used for its predictive analysis. First, the central site uses the exhaustive service policy and the common site uses the Limited K = 2 service policy to establish a continuous-time exhaustive-limited (K = 2) two-level polling control system. Second, the exact expressions for the average queue length, average delay and cycle period are derived using probability generating functions and Markov chains and the MATLAB simulation experiment. Finally, the LSTM neural network and an attention mechanism model is constructed for prediction. The experimental results show that the theoretical and simulated values basically match, verifying the rationality of the theoretical analysis. Not only does it differentiate priorities to ensure that the central site receives a quality service and to ensure fairness to the common site, but it also improves performance by 7.3 and 12.2%, respectively, compared with the one-level exhaustive service and the one-level limited K = 2 service; compared with the two-level gated- exhaustive service model, the central site length and delay of this model are smaller than the length and delay of the gated- exhaustive service, indicating a higher priority for this model. Compared with the exhaustive-limited K = 1 two-level model, it increases the number of information packets sent at once and has better latency performance, providing a stable and reliable guarantee for wireless network services with high latency requirements. Following on from this, a fast evaluation method is proposed: Neural network prediction, which can accurately predict system performance as the system size increases and simplify calculations.