Dr Josephine Sullivan (Associate Professor)KTH Royal Institute of Technology, Stockholm

Self-supervised and transformers deep learning within computer vision:Deep learning has progressed in recent years, especially within the application areas of computer vision and NLP, from supervised learning to learning without explicit labels. Frequently, these labels for training have been obtained by time consuming manual effort. In this class I will present the techniques of self-supervised learning which has allowed label free learning. In addition I will present the new Transformer architecture which has brought astonishing success to NLP, increasingly computer vision and also allowed seamless multi-modal learning. Finally, I will touch on the on-going mystery of generalization deep learning that is its ability to train very high capacity networks on relatively limited training data and still generalize to unseen example.

Schedule:

10am – Open

11- 11.30am – Coffee break

12.45-2.00pm – Lunch

3.00pm – Coffee break

4.30pm- Close

Location: TSI Building (Lecture Theatre 1), Maynooth University. Virtual attendance will be made available.

Speaker biography:

Josephine Sullivan is an associate professor at KTH Royal Institute of Technology, Stockholm within the school of Electrical Engineering and Computer Science. She has worked for several decades within the field of computer vision and seen the field transformed by deep learning. Her initial research focus was on object and multi-target tracking in visual data and 3D human pose estimation. More recently she has performed research on representation and transfer learning and multi-modal learning between vision and text. She received her PhD from Oxford University and before that a BA in Mathematics from Trinity College Dublin.