To improve performance in current computer vision class, neural systems need to take into account spatially remote information.
At the seminar, we will discuss one of the latest NeurIPS papers that proposes a new and straightforward approach to that problem. It utilizes filters with average pooling from rectangular regions. We then can use sizes and margins of those filters as additional features to efficiently calculate distant correlations.
Speaker: Stanislav Belyaev.
Presentation language: Russian.
Date and time: February 20th, 18:30-20:00.
Location: Times, room 204.
Videos from previous seminars are available at http://bit.ly/MLJBSeminars