Search for contacts, projects,
courses and publications

Long Short Term Memory

People

 

Schmidhuber J.

(Responsible)

Abstract

The human brain is biological recurrent neural network (RNN). Goal of this project is to further extend, analyze, and apply our state-of-the-art artificial RNNs based on the "Long Short Term Memory" (LSTM) architecture.

While our previous research has focused on demonstrating LSTM's advantages on artificial tasks, we have started to apply LSTM to real world sequences such as speech signals, with promising results. We want to further improve LSTM, in particular, create variants of bi-directional LSTM that use both left and right context to classify sequence elements somewhere within sequences. Intended applications: challenging real world tasks such as speech processing and protein prediction, also with HMM/LSTM hybrids.

Additional information

Start date
01.07.2005
End date
01.04.2006
Duration
9 Months
Funding sources
SNSF
Status
Ended
Category
Swiss National Science Foundation / Project Funding / Division II - Mathematics, Natural and Engineering Sciences