Discriminatory models of machine learning, models that are able to allocate any properties of objects, are now widely used, but there is also a more interesting class of models: generative. These models allow the creation of objects with given properties: for example, to write chemical formulas for cancer-free drugs previously unknown to mankind or to draw advertising banners that will be called by a particular audience.
Among generative neural networks, two classes are most popular: VAEs and GANs. During the presentation, we will consider variation encoders, while maintaining the balance between the availability of presentation and mathematical rigor. The introduction of the Bayesian approach to the report may also be useful for self-study of such topics as deep reinforcement learning.
Buy this talk
Buy this video
ConferenceCast.tv — conference video talk archive.
With this service you can find interesting talks especially for you!