Structuralism and Self-Supervised Learning

This talk is an update on the previous talk about Word Embeddings that includes, alongside with the history and methods of distributional semantics and word embeddings, describes the thesis that the current trend of Self-Supervised learning in Machine learning is a practical implementation of the basic principles of Structuralism.

I presented this talk for the first time at the Deep Learning Classics and Trends reading group, organized by the amazing Rosanne Liu (although there is not a lot of Deep Learning in the talk, just a dash at the end) and recorded myself during the dry run.

Image: Kubuswoningen (Cube Houses) / Piet Blom – Rotterdam (roof plan). Image © Het Nieuwe Instituut

Download Slides Preview