Details

Autor(en) / Beteiligte

• Pozas-Kerstjens, Alejandro
• Muñoz-Gil, Gorka
• Piñol, Eloy
• García-March, Miguel Ángel
• Acín, Antonio
• Lewenstein, Maciej
• Grzybowski, Przemysław R

Titel

Efficient training of energy-based models via spin-glass control

Ist Teil von

• Machine learning: science and technology, 2021-06, Vol.2 (2), p.25026

Ort / Verlag

Bristol: IOP Publishing

Erscheinungsjahr

2021

Quelle

EZB-FREE-00999 freely available EZB journals

Beschreibungen/Notizen

• Abstract We introduce a new family of energy-based probabilistic graphical models for efficient unsupervised learning. Its definition is motivated by the control of the spin-glass properties of the Ising model described by the weights of Boltzmann machines. We use it to learn the Bars and Stripes dataset of various sizes and the MNIST dataset, and show how they quickly achieve the performance offered by standard methods for unsupervised learning. Our results indicate that the standard initialization of Boltzmann machines with random weights equivalent to spin-glass models is an unnecessary bottleneck in the process of training. Furthermore, this new family allows for very easy access to low-energy configurations, which points to new, efficient training algorithms. The simplest variant of such algorithms approximates the negative phase of the log-likelihood gradient with no Markov chain Monte Carlo sampling costs at all, and with an accuracy sufficient to achieve good learning and generalization.