Moacir Ponti - CMIC/WEISS joint seminar series
17 February 2021, 1:00 pm–2:00 pm
Moacir Ponti a talk as part of the CMIC/WEISS joint seminar series
Event Information
Open to
- All
Organiser
-
UCL Centre for Medical Image Computing and Wellcome/EPSRC Centre for Interventional and Surgical Sciences
Speaker: Moacir Ponti
Title: Improving transfer learning with self-supervision and multiple feature alignment with applications on computer vision and biomedical imaging
Abstract: Using pre-trained models for transfer learning became the standard way to use deep learning for a target task when training a model from scratch is not viable. This lecture will present two possibilities for improving transfer of learning: (1) feature extraction in multiple layers, followed by manifold alignment, and (2) the use of semi-supervised self-supervision tasks in order to allow using unlabelled data to obtain more discriminative embeddings.
Short Bio: Moacir Antonelli Ponti is an associate professor at the Universidade de São Paulo (USP), Brazil, with over 40 peer-reviewed papers published in journals and conferences related to Pattern Recognition, Computer Vision, Image and Signal Processing. At USP he has 10 years experience in teaching, as a principal investigator of many research projects and grants, and in management at the Institute of Mathematical and Computer Sciences, chairing the Outreach committee. He was one of the recipients of Google Latin America Research Award in 2017-2018, and co-authored a book "Machine Learning: a practical approach on statistical learning theory", published by Springer.