Neural Networks

[Massimo I]

Taylor (16.4.2) mentions that Neural Networks can perform better than Decision Trees, while also having their own shortcomings (which are not specified). I seem to understand that NN would be better because (a) their output is not always going to be the prototype of a feature combination, but rather a vector (is it really bad to always use a prototype?) and (b) they can take into account missing features by generating new clusters (I don’t really get how – does this mean they don’t require as much feature selection?).

[Simon]

All good questions, but we’re going to talk about NNs for synthesis a bit later in the course, so make sure to ask them again at that point.

At this stage, we can state that a NN is just a non-linear regression model, and so replacing a regression tree with a NN is not a big conceptual leap. That should be much clearer after we have covered HMM-based speech synthesis and the way that it uses regression trees.

[Author]

Posts