Sara Beery lectures on similarity-based representation learning as part of MIT's 6.7960 Deep Learning course. She covers metric learning and contrastive learning, comparing self-supervised and supervised approaches to building useful embeddings. The lecture works through the InfoNCE loss function and the alignment and uniformity principles used to evaluate whether a learned representation space is well structured. Beery also addresses the practical problem of hard negatives, explaining why the choice of negative examples during training strongly affects what a model actually learns. Running about 76 minutes, the session fits into the course's broader unit on representation learning and assumes familiarity with basic neural network training concepts. It is a technical, board-and-slides style lecture aimed at students already working through the course's deep learning curriculum.