Learning to Branch: Accelerating Resource Allocation in Wireless Networks
share
Resource allocation in wireless networks, such as device-to-device (D2D) communications, is usually formulated as mixed integer nonlinear programming (MINLP) problems, which are generally NP-hard and difficult to get the optimal solutions. Traditional methods to solve these MINLP problems are all based on mathematical optimization techniques, which can only get sub-optimal solution and usually have forbidding complexity for real-time implementation. In this paper, we introduce a machine leaning (ML) technique to address the MINLP problems in wireless networks. Different from the existing end-to-end learning paradigm, the proposed learning method exploits the specific algorithm structures. Specifically, we adopt the imitation learning method to accelerate the branch-and-bound (B&B) algorithm, a globally optimal algorithm for MINLP problems. It is achieved by learning a good prune policy to speed up the most time-consuming branch process of the B&B algorithm. With appropriate feature selection, the imitation learning can be further converted into a binary classification problem, which can be solved by the classical support vector machine (SVM). Moreover, we develop a mixed training strategy to further reinforce the generalization ability and the soft-decision algorithm to achieve dynamic control over optimality and computational complexity. By using resource allocation in D2D communications as an example, extensive simulation demonstrates that the proposed method can achieve good optimality and reduce computational complexity simultaneously. Since the proposed method only needs hundreds of training samples, it can be potentially generalized into larger scale problems and applied to MINLP problems in other wireless communication networks.
READ FULL TEXT