Unsupervised learning for parametric optimization
This letter proposes the unsupervised training of a feedforward neural network to solve parametric optimization problems involving large numbers of parameters. Such unsupervised training, which consists in repeatedly sampling parameter values and performing stochastic gradient descent, foregoes the...
| Autores: | , , |
|---|---|
| Formato: | artículo |
| Estado: | Versión aceptada para publicación |
| Fecha de publicación: | 2021 |
| País: | España |
| Recursos: | Universitat Pompeu Fabra |
| Repositorio: | Repositorio Digital de la UPF |
| OAI Identifier: | oai:repositori.upf.edu:10230/47546 |
| Acesso em linha: | http://hdl.handle.net/10230/47546 http://dx.doi.org/10.1109/LCOMM.2020.3027981 |
| Access Level: | acceso abierto |
| Palavra-chave: | Machine learning Neural networks Unsupervised learning Parametric optimization Convex optimization Quadratic program |
| Resumo: | This letter proposes the unsupervised training of a feedforward neural network to solve parametric optimization problems involving large numbers of parameters. Such unsupervised training, which consists in repeatedly sampling parameter values and performing stochastic gradient descent, foregoes the taxing precomputation of labeled training data that supervised learning necessitates. As an example of application, we put this technique to use on a rather general constrained quadratic program. Follow-up letters subsequently apply it to more specialized wireless communication problems, some of them nonconvex in nature. In all cases, the performance of the proposed procedure is very satisfactory and, in terms of computational cost, its scalability with the problem dimensionality is superior to that of convex solvers. |
|---|