FnS - Study Independently: Self-supervised Learning and Its Applications

Add to your list(s) Download to your calendar using vCal

• Jianbo Jiao
• Thursday 28 April 2022, 16:00-16:30
• LG23.

If you have a question about this talk, please contact Mirco Giacobbe.

With the development of computing resources and the availability of large-scale data, modern machine learning, especially deep learning, has been demonstrated to be super effective in many areas, even performing better than human experts in some cases. However, most existing deep models rely on human-annotated data to train and do not generalise very well. Whereas in many scenarios such human annotation is difficult and even infeasible to acquire. As a result, how to learn representations purely from the data itself is crucial. In this talk, I will give an introduction to the technique to achieve this—self-supervised learning, with corresponding applications. General self-supervised visual representation learning will be introduced with typical approaches including but not limited to pretext-task design and contrastive learning. Self-supervised learning with sequential video data and multi-modal data will be introduced as well. Finally, the ability to achieve equivariance to arbitrary transformations in a novel self-supervised manner will be discussed.

The talk will also be streamed on zoom: https://bham-ac-uk.zoom.us/j/85289214035

This talk is part of the Facts and Snacks series.

This talk is included in these lists:

• Facts and Snacks
• LG23

Note that ex-directory lists are not shown.