Anomaly detection has been widely used in grid operation and maintenance, machine fault detection, and so on. In these applications, the multivariate time-series data from multiple sensors with latent relationships are always high-dimensional, which makes multivariate time-series anomaly detection particularly challenging. In existing unsupervised anomaly detection methods for multivariate time series, it is difficult to capture the complex associations among multiple sensors. Graph neural networks (GNNs) can model complex relations in the form of a graph, but the observed time-series data from multiple sensors lack explicit graph structures. GNNs cannot automatically learn the complex correlations in the multivariate time-series data or make good use of the latent relationships among time-series data. In this paper, we propose a new method—masked graph neural networks for unsupervised anomaly detection (MGUAD). MGUAD can learn the structure of the unobserved causality among sensors to detect anomalies. To robustly learn the temporal context from adjacent time points of time-series data from the same sensor, MGUAD randomly masks some points of the time-series data from the sensor and reconstructs the masked time points. Similarly, to robustly learn the graph-level context from adjacent nodes or edges in the relation graph of multivariate time series, MGUAD masks some nodes or edges in the graph under the framework of a GNN. Comprehensive experiments are conducted on three public datasets. According to the experimental findings, MGUAD outperforms state-of-the-art anomaly detection methods.

To solve collaborative tasks, Multi-Agent Systems (MAS) have been widely used because of their collaboration capabilities and flexibility. They have been applied in various fields, such as robotic collaboration and intelligent transportation systems. In multi-agent collaborative problems, coordinating the collaboration strategies between agents is the most challenging aspect. Traditional Multi-Agent Reinforcement Learning (MARL) methods attempt to transform the team objective into individual objectives for each agent through value decomposition. However, the value allocation mechanisms in existing methods struggle to accurately measure the contribution differences between agents. In this paper, we propose an MARL method based on structured coordination by leveraging the local interaction structure among agents. Depending on the application scenarios, we present explicit and implicit implementation approaches. The explicit method constructs an explicit collaboration graph based on the cooperation relationships between agents, and then uses Shapley values to assess the contribution of each agent in the collaboration graph. Additionally, we propose a method based on self-attention mechanisms to dynamically construct implicit interaction structures. Experimental results demonstrate that, compared to several state-of-the-art multi-agent collaboration algorithms, the two collaboration methods we proposed achieve significant performance advantages across multiple complex collaboration scenarios.

By exploiting the correlation between the structure and the solution of Mixed-Integer Linear Programming (MILP), Machine Learning (ML) has become a promising method for solving large-scale MILP problems. Existing ML-based MILP solvers mainly focus on end-to-end solution learning, which suffers from the scalability issue due to the high dimensionality of the solution space. Instead of directly learning the optimal solution, this paper aims to learn a reduced and equivalent model of the original MILP as an intermediate step. The reduced model often corresponds to interpretable operations and is much simpler, enabling us to solve large-scale MILP problems much faster than existing commercial solvers. However, current approaches rely only on the optimal reduced model, overlooking the significant preference information of all reduced models. To address this issue, this paper proposes a preference-based model reduction learning method, which considers the relative performance (i.e., objective cost and constraint feasibility) of all reduced models on each MILP instance as preferences. We also introduce an attention mechanism to capture and represent preference information, which helps improve the performance of model reduction learning tasks. Moreover, we propose a SetCover based pruning method to control the number of reduced models (i.e., labels), thereby simplifying the learning process. Evaluation on real-world MILP problems shows that 1) compared to the state-of-the-art model reduction ML methods, our method obtains nearly 20% improvement on solution accuracy, and 2) compared to the commercial solver Gurobi, two to four orders of magnitude speedups are achieved.

Due to the rising demand for Quality of Service (QoS) in the emerging network, this paper considers the routing problem of efficiently planning the transmission path of flow demands to balance the load of the network. Existing classical routing algorithms based on static decision-making are unable to adjust to dynamical scenarios with fluctuating flow. Traditional reinforcement learning-based methods face challenges in large-scale scenarios due to the vast state and action spaces. Besides, the neural networks used for policy generation also suffer from issues like rudimentary feature extraction and limited learning capability. In this paper, we first introduce RequestNet, a novel attention-based model designed to better leverage the information of network topology and feature correlations. We also propose a multi-agent modeling approach for the network packet routing problem, named RequestNet-MA, which treats pairs of edge routers as agents, allowing us to allocate traffic from the perspective of flow demands, thus significantly reducing the dimensions of state and action spaces and enhancing inter-agent communication and cooperation. Extensive experiments and ablation studies demonstrate that the proposed RequestNet-MA can reduce maximum link utilization and scale well to a large-scale network.

By exploiting the correlation between the structure and the solution of Mixed-Integer Linear Programming (MILP), Machine Learning (ML) has become a promising method for solving large-scale MILP problems. Existing ML-based MILP solvers mainly focus on end-to-end solution learning, which suffers from the scalability issue due to the high dimensionality of the solution space. Instead of directly learning the optimal solution, this paper aims to learn a reduced and equivalent model of the original MILP as an intermediate step. The reduced model often corresponds to interpretable operations and is much simpler, enabling us to solve large-scale MILP problems much faster than existing commercial solvers. However, current approaches rely only on the optimal reduced model, overlooking the significant preference information of all reduced models. To address this issue, this paper proposes a preference-based model reduction learning method, which considers the relative performance (i.e., objective cost and constraint feasibility) of all reduced models on each MILP instance as preferences. We also introduce an attention mechanism to capture and represent preference information, which helps improve the performance of model reduction learning tasks. Moreover, we propose a SetCover based pruning method to control the number of reduced models (i.e., labels), thereby simplifying the learning process. Evaluation on real-world MILP problems shows that 1) compared to the state-of-the-art model reduction ML methods, our method obtains nearly 20% improvement on solution accuracy, and 2) compared to the commercial solver Gurobi, two to four orders of magnitude speedups are achieved.