Self-Supervised Learning

Nicolò Felicioni (web, mail)

Intelligenza artificiale, robotica e computer vision

Deep Learning, Neural Networks, Machine Learning

LaTex, Python, Tensorflow/Pytorch

Self-supervised learning (SSL) is an unsupervised technique for learning representations without human-provided labels. It consists of creating auxiliary tasks on unlabeled input data and learning representations by solving these tasks. SSL has shown great results on various tasks (e.g., SimCLR, GPT). The goal of this project is to provide a literature review, particularly focusing on the theoretical foundations of SSL, and to implement a comparison of the reviewed approaches.