We will talk about the domain adaptation task. About how the problem can be solved for the unsupervised dataset, for which we know possible set of labels. To obtain this knowledge use another labled dataset.
This approach is applicable, for example, to synthetic data for processing real data. Synthetic data is inherently already labeled.
There are several approaches to the problem. We can generate images so that they look realistic, thereby performing a style transfer, as well as adapt on a deep semantic level. One of the main tools in this task are the generative adversarial neural networks.
At the seminar we will consider both approaches and SOTA algorithms in each of the methods.
Speaker: Alexey Artamonov.
Presentation language: Russian.
Date and time: November 20th, 20:30-22:00.
Location: Times, white boards (4th floor).