Strategies Against Mode Collapse
Ishan explains the importance of dissimilarity in models to avoid trivial solutions. Contrastive learning defines positives and negatives, while clustering creates soft clusters in feature space. The concept of prototypes in Swagger reduces the comparison complexity, offering a unique approach to feature similarity.In this clip
From this podcast
Machine Learning Street Talk (MLST)
#55 Self-Supervised Vision Models (Dr. Ishan Misra - FAIR).
Related Questions