Swimming Without a Lifeguard in the Ocean of Ideas - Abstraction and Unsupervised Deep Learning

Dive into a comprehensive lecture on unsupervised deep learning and abstraction presented by Kiel Howe from Minerva Schools at KGI. Explore the fascinating world of deep learning models and their ability to discover useful abstractions in input data. Gain insights into unsupervised techniques for uncovering latent features in unlabeled data and learn about the progress made in developing hierarchical latent feature spaces. Understand how these hierarchical representations can enhance interpretability and performance in supervised learning tasks, as well as their role in deep generative models. Discover a new approach to finding hierarchical latent feature spaces and consider the potential connections between neural and symbolic learning systems. Benefit from the speaker's expertise as complex machine learning concepts are explained in accessible terms, with analogies drawn from statistical physics, information theory, and tensor networks.