| Title | Training and Analysing Recurrent Nets That Learn to Predict Non-Regular Languages |
|---|---|
| Speaker | Dr. Stephan Chalup |
| Date | Thursday, 25th July 2002 |
| Time | 11:10 -- 12:00 |
| Venue | Computer Science - Room 152 |
| Abstract | A simple recurrent neural network can be represented by a small directed cyclic graph. There is a high probability that such a graph is present in large biological neural networks such as the human brain. This study shows how simple recurrent neural networks can learn to predict subsets of non-regular languages and analyses their hidden unit activity. These results can contribute to the discussion whether an innate language module is required to process syntax or if this ability can be acquired after birth. |
| About the speaker | Dr. Stephan K. Chalup was educated in mathematics and neurobiology at the University of Heidelberg. He received his Ph. D. from QUT in 2002. Currently he is lecturer for Computer Science and Software Engineering at the University of Newcastle where he is leader of the Robotics and Brain Architecting Laboratory. For details see http://www.cs.newcastle.edu.au/~chalup |
For more information on this seminar, please email: Dr. Stephen Chalup
For information on our seminar program, suggestions for seminars, or mailing list updates, please email: seminars@cs.adfa.edu.au or see: http://www.cs.adfa.edu.au/seminars/2003/
Copyright
and Disclaimer