next up previous
Next: THE EASY TASK Up: NEURAL PREDICTORS FOR DETECTING Previous: REDUNDANCY REDUCTION: HOW?

EXAMPLE 1: Sequence Classification

Outline. This section starts by describing a sequence classification task. The training sequences consist of relatively nonredundant components. It is shown how to solve the task with a conventional recurrent neural net. Then it is shown that the recurrent net fails to learn a very similar task involving training sequences conveying a lot of redundant information. Finally it is demonstrated how unsupervised adaptive redundancy reduction allows for learning the second task, too.



Subsections

Juergen Schmidhuber 2003-02-19