Fermilab Computing Division

Recurrent Neural Networks for Time Series Prediction

Full Title: Recurrent Neural Networks for Time Series Prediction
Date & Time: 09 Feb 2016 at 13:00
Event Location: WH1W
Event Topic(s): Computing Techniques Seminar
Event Moderator(s):
Event Info: Speaker:
Florian Metze, Associate Research Professor, Carnegie Mellon’s University

Abstract:
Speech recognition can be seen as a sequence-to-sequence translation task, where a sequence of input features(“frames”) are mapped to a sequence of output symbols, namely phones or words. In the past, work has focused on improving the frame-level predictions, while the accuracy of the actual output sequence was an indirect effect. Recurrent Neural Networks are beginning to change that, and offer computationally tractable ways of predicting entire sequences under a single optimization criterion.

In this talk, I will give a brief introduction to the theory of Recurrent Neural Networks and present their most common and interesting use cases in speech recognition, natural language processing, multi-media analysis, or other “big-data” applications. I will outline tasks and data sets, and attempt to distill the key factors that contributed to making Deep Learning so successful. I will also look to other domains that involve (time) series modeling and speculate about possible uses of Recurrent Neural Networks there.

Bio:
Florian Metze is an Associate Research Professor at Carnegie Mellon’s University, in the Language Technologies Institute. His work is centered around speech and multi-media processing with a focus on low resource and multi-lingual speech processing, large-scale multi-media retrieval and summarization, along with recognition of personality or similar meta-data from speech. Most recently, his group released the “Eesen” toolkit for end-to-end speech recognition using recurrent neural networks and connectionist temporal classification.

No talks in agenda


DocDB Home ]  [ Search ] [ Authors ] [ Events ] [ Topics ]

DocDB Version 8.7.23, contact Document Database Administrators
Execution time: 0 wallclock secs ( 0.22 usr + 0.04 sys = 0.26 CPU)