The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Optimization, Machine Learning and Intelligent Experimentation with Michael McCourt - #545
Dec 16, 2021
Michael McCourt, Head of Engineering at SigOpt, dives into the world of optimization and its pivotal role in machine learning. He shares his journey from theoretical mathematics to practical applications, emphasizing the importance of collaboration and intelligent experimentation. The conversation touches on the intricacies of optimizing ML models, the synergy between active learning and optimization, and the exciting interdisciplinary work emerging from the latest NeurIPS conference, particularly in areas like drug discovery and climate modeling.
AI Snips
Chapters
Transcript
Episode notes
Optimization in Machine Learning
- Optimization is a part of machine learning, used in learning models.
- It's also a standalone technique for problems with less information, using ML to enhance the optimization process.
Optimization Beyond Hyperparameters
- Optimization is used for hyperparameter tuning in machine learning and in other fields like HPC.
- An example is finding optimal materials for specialized glass in optoelectronic devices.
Problem Formulation
- Define the optimization problem's goals and constraints clearly in natural language with stakeholders.
- Then create a rigorous mathematical formulation for computation and measurement.