The culmination of four years of work! Shanshan defended her thesis on 9th of June. The opponents were Jorma Laaksonen from Aalto University, Finland and Xavier Alameda-Pineda from Université Grenoble Alpes, France.
The thesis focuses on self supervised learning, and goes into details like the effect of using angular loss in learning, different sampling strategies for positive and negative pairs in contrastive learning, or learning spatial alignment in 360-degree videos. The thesis is based on five peer-reviewed papers, four of which are conference publications, and one journal.
It was a long day of presentations, questions, and discussions, but it ended with cake, sparkling wine and a fancy dinner (karonkka).
The thesis is available online: Self-supervised Representation Learning on Audio-Video Data