Machine Learning Street Talk (MLST) ICLR 2020: Yann LeCun and Energy-Based Models
32 snips
May 19, 2020 
Yann LeCun, a pioneer in machine learning and AI, discusses the latest in self-supervised learning and energy-based models (EBMs). He compares how humans and machines learn concepts, advocating for methods that mimic human cognition. The conversation dives into EBMs' applications in optimizing labels and addresses challenges in traditional models. LeCun also explores the potential of self-supervised learning techniques for enhancing AI capabilities, such as in natural language processing and image recognition.
AI Snips
Chapters
Books
Transcript
Episode notes
Energy-Based Models: A Unifying Framework
- Energy-based models (EBMs) are a general framework for machine learning problems.
- Almost any machine learning problem can be reframed as an EBM problem.
EBM Example: Shape Learning
- An EBM example learns shapes from demonstrations.
- It replicates shapes by inferring energy functions.
ICLR and LeCun's Keynote
- ICLR is a top machine learning conference, second only to NeurIPS.
- Yann LeCun's keynote focused on self-supervised learning, EBMs, and manifold learning.
