Bergmair, Richard. 2004c. “Some Experimental Results on Feed-Forward Networks for Text Classification.”
One well-known design principle guiding the construction of neural networks is that a feed-forward network will be able to filter noise when few hidden units are used. Another design principle states that a network will be able to solve more complex problems when many hidden units are used. The experimental setup presented herein aims at analyzing this tradeoff in the situation where both the noise in the input increases and the problem grows more complex, at least in the problem domain of text classification by unigram frequencies.
==> as submitted
[PDF]
==>
at ResearchGate
==>
at archive.org