Foundations of Unsupervised Deep Learning (Ruslan Salakhutdinov, CMU)

Summary

This talk introduces unsupervised learning, contrasting it with supervised methods like convolutional networks and highlighting its current developmental stage. The primary motivation is the vast amount of unlabeled data available, with a focus on deep learning and hierarchical representations to discover structure. The practical takeaway is the potential of unsupervised methods, demonstrated by an autoencoder on the Reuters dataset, for tasks like visualization and uncovering hidden patterns in data.