February 2019
Intermediate to advanced
386 pages
9h 54m
English
Another unsupervised approach is based on generative models. The concept is not very different from what we have already discussed for supervised algorithms, but, in this case, the data generating process doesn't contain any label. Hence the goal is to model a parametrized distribution and optimize the parameters so that the distance between candidate distribution and the data generating process is minimized:

The process is generally based on the Kullback-Leibler divergence or other similar measures:

At the end of the training ...
Read now
Unlock full access