Information bottleneck
Add to your list(s)
Download to your calendar using vCal
If you have a question about this talk, please contact Konstantina Palla.
Please note the change in time. This RCC will take place at 15:00-16.30pm.
The “information bottleneck” provides an information theoretic perspective on both supervised and unsupervised learning. We start with a joint distribution over two variables p(x,y). Examples of x & y might include distributions over documents and associated word counts, visual input and spike count data, or speech spectrograms and phoneme labels. The goal is to extract the meaningful information from this joint distribution in the form of a new variable t, which is produced according to a (possibly probabilistic) mapping p(t|x). The information bottleneck achieves this by demanding the new variable t to be as informative about y as possible, whilst also compressing x as much as possible. We will describe several variants of this approach, the connections to other learning approaches, and we will finish by evaluating the method in this context.
This talk is part of the Machine Learning Reading Group @ CUED series.
This talk is included in these lists:
Note that ex-directory lists are not shown.
|