Presentation | 2014-11-18 Denoising High-dimensional Sequences with the Bidirectional Recurrent Restricted Boltzmann Machine Shoken KANEKO, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | We propose a probabilistic neural network for modeling high-dimensional sequences with complex non-linearities. Our model is an extension of the previously introduced Recurrent Neural Network-Restricted Boltzmann Machine. We extend the model by adding a backward recurrent chain, which makes bidirectional propagation of information possible and allowing our model to incorporate knowledge of future observations. Our model can be readily applied for tasks such as denoising of high-dimensional sequences. We show that our model outperforms the unidirectional model in the task of denoising moving pictures of balls bouncing in a box, reconstructing much smoother sequences. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Machine Learning / Neural Networks / Recurrent Neural Network-Restricted Boltzmann Machines |
Paper # | IBISML2014-62 |
Date of Issue |
Conference Information | |
Committee | IBISML |
---|---|
Conference Date | 2014/11/10(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | |
Vice Chair | |
Secretary | |
Assistant |
Paper Information | |
Registration To | Information-Based Induction Sciences and Machine Learning (IBISML) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Denoising High-dimensional Sequences with the Bidirectional Recurrent Restricted Boltzmann Machine |
Sub Title (in English) | |
Keyword(1) | Machine Learning |
Keyword(2) | Neural Networks |
Keyword(3) | Recurrent Neural Network-Restricted Boltzmann Machines |
1st Author's Name | Shoken KANEKO |
1st Author's Affiliation | Research and Development Division, Yamaha Corporation() |
Date | 2014-11-18 |
Paper # | IBISML2014-62 |
Volume (vol) | vol.114 |
Number (no) | 306 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |