The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Gauge Equivariant CNNs, Generative Models, and the Future of AI with Max Welling - TWiML Talk #267
May 20, 2019
In this enlightening discussion, Max Welling, a research chair in machine learning at the University of Amsterdam and Qualcomm's VP of Technologies, dives into groundbreaking topics. He reveals his work on Bayesian deep learning, Gauge Equivariant CNNs, and innovations in AI for improved computing efficiency. Max also shares his insights on the evolution of AI, emphasizing the balance between models and data, and explores the exciting possibilities of integrating generative models with rule-based systems to pave the way for artificial general intelligence.
AI Snips
Chapters
Transcript
Episode notes
Cypher's Unexpected Win
- Max Welling's company, Cypher, outperformed large consulting firms in a bank competition.
- Their success with a single GPU led to securing the project and the eventual acquisition by Qualcomm.
Active Learning at Tata Steel
- Cypher developed an active learning product for Tata Steel to detect defects in steel slabs.
- The algorithm identified images needing expert labeling, improving its performance over time.
Compute Efficiency's Importance
- Deep learning's increasing model sizes highlight the importance of compute efficiency.
- Making algorithms run efficiently is crucial for progress, similar to how the brain operates efficiently.