Deep Unsupervised Learning for Generalized Assignment Problems: A Case-Study of User-Association in Wireless Networks

There exists many resource allocation problems in the field of wireless\ncommunications which can be formulated as the generalized assignment problems\n(GAP). GAP is a generic form of linear sum assignment problem (LSAP) and is\nmore challenging to solve owing to the presence of both equality and inequality\nconstraints. We propose a novel deep unsupervised learning (DUL) approach to\nsolve GAP in a time-efficient manner. More specifically, we propose a new\napproach that facilitates to train a deep neural network (DNN) using a\ncustomized loss function. This customized loss function constitutes the\nobjective function and penalty terms corresponding to both equality and\ninequality constraints. Furthermore, we propose to employ a Softmax activation\nfunction at the output of DNN along with tensor splitting which simplifies the\ncustomized loss function and guarantees to meet the equality constraint. As a\ncase-study, we consider a typical user-association problem in a wireless\nnetwork, formulate it as GAP, and consequently solve it using our proposed DUL\napproach. Numerical results demonstrate that the proposed DUL approach provides\nnear-optimal results with significantly lower time-complexity.\n