Please use this identifier to cite or link to this item:
|Title:||Echo-state conditional restricted boltzmann machines||Authors:||Chatzis, Sotirios P.||Keywords:||Restricted boltzmann machines;Εcho-state||Category:||Electrical Engineering - Electronic Engineering - Information Engineering||Field:||Engineering and Technology||Issue Date:||2014||Source:||28th AAAI Conference on Artificial Intelligence, 2014, Québec, Canada, 27–31 July||Abstract:||Restricted Boltzmann machines (RBMs) are a powerful generative modeling technique, based on a complex graphical model of hidden (latent) variables. Conditional RBMs (CRBMs) are an extension of RBMs tailored to modeling temporal data. A drawback of CRBMs is their consideration of linear temporal dependencies, which limits their capability to capture complex temporal structure. They also require many variables to model long temporal dependencies, a fact that might provoke overfitting proneness. To resolve these issues, in this paper we propose the echo-state CRBM (ESCRBM): our model uses an echo-state network reservoir in the context of CRBMs to efficiently capture long and complex temporal dynamics, with much fewer trainable parameters compared to conventional CRBMs. In addition, we introduce an (implicit) mixture of ES-CRBM experts (im-ESCRBM) to enhance even further the capabilities of our ES-RBM model. The introduced im-ES-CRBM allows for better modeling temporal observations which might comprise a number of latent or observable subpatterns that alternate in a dynamic fashion. It also allows for performing sequence segmentation using our framework. We apply our methods to sequential data modeling and classification experiments using public datasets.||URI:||http://ktisis.cut.ac.cy/handle/10488/8203||Type:||Conference Papers|
|Appears in Collections:||Δημοσιεύσεις σε συνέδρια/Conference papers|
Show full item record
Page view(s) 5072
checked on Dec 19, 2018
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.