ICLR 2020: Yann LeCun and Energy-Based Models
Machine Learning Street Talk (MLST)
00:00
True Energy Functions and Manifolds in Learning
Learning energy functions and understanding data manifolds are important concepts in deep learning. The true energy function is represented by an ellipse, while the learned energy function is approximated through interpolation. Regularizing the model to only produce ellipses can improve the accuracy of the learned energy function. Data manifolds are paths connecting data points in high dimensional space, allowing for flexible representations of data.
Transcript
Play full episode
