Speaker
Moshir Harsh
(ENS, Paris)
Description
Restricted Boltzmann Machines (RBMs) are generative neural networks that can learn and sample from probability distributions over its set of inputs. RBMs are a fundamental building block in deeper neural networks such as Deep belief Networks and can act as feature extractors from high dimensional data sets. However it is not well understood how RBMs learn features in time. We try to understand the important features of the learning process and time dynamics by training RBMs on simple ising model configurations.
Primary authors
Moshir Harsh
(ENS, Paris)
Mr
Jerome Tubiana
Mr
Remi Monasson
