Please use this identifier to cite or link to this item:
https://hdl.handle.net/20.500.14279/23149
Title: | Variational Bayesian Sequence-to-Sequence Networks for Memory-Efficient Sign Language Translation |
Authors: | Partaourides, Harris Voskou, Andreas Kosmopoulos, Dimitrios I. Chatzis, Sotirios P. N. Metaxas, Dimitris |
Major Field of Science: | Natural Sciences |
Field Category: | Computer and Information Sciences |
Keywords: | Deep learning;Gloss to Text;Sign Language Translation;Weight compression |
Issue Date: | 2020 |
Source: | 15th International Symposium on Visual Computing, 2020, 5-7 October, San Diego |
Project: | DeepSignNet: Video processing for Sign Language Recognition using Deep Bayesian Recurrent Neural Networks |
Conference: | International Symposium on Visual Computing (ISVC) |
Abstract: | Memory-efficient continuous Sign Language Translation is a significant challenge for the development of assisted technologies with real-time applicability for the deaf. In this work, we introduce a paradigm of designing recurrent deep networks whereby the output of the recurrent layer is derived from appropriate arguments from nonparametric statistics. A novel variational Bayesian sequence-to-sequence network architecture is proposed that consists of a) a full Gaussian posterior distribution for data-driven memory compression and b) a nonparametric Indian Buffet Process prior for regularization applied on the Gated Recurrent Unit non-gate weights. We dub our approach Stick-Breaking Recurrent network and show that it can achieve a substantial weight compression without diminishing modeling performance. |
URI: | https://hdl.handle.net/20.500.14279/23149 |
ISBN: | 9783030645588 |
DOI: | 10.1007/978-3-030-64559-5_19 |
Rights: | © Springer |
Type: | Conference Papers |
Affiliation : | Cyprus University of Technology University of Patras Rutgers University |
Publication Type: | Peer Reviewed |
Appears in Collections: | Δημοσιεύσεις σε συνέδρια /Conference papers or poster or presentation |
CORE Recommender
This item is licensed under a Creative Commons License