E-mail: benus@ii.fmph.uniba.sk
We are interested in processing of complex symbolic time series by
means of RNNs. In particular, we investigate the organization of RNNs
state space before and after traing to gain an insight how RNNs encode
the history of inputs to use it for prediction. We are also interested
in exploring novel methods of RNN training (like for instance some
unsupervised methods). We work with time series of different origin
and complexity, but in particular we are interested in processing
of artificial and natural languages.
See
Publications - artificial neural networks