The Data Exchange Podcast: Nic Hohn and Max Pumperla on lessons learned from applying RL to industrial problems.
Happy Thanksgiving to listeners who celebrate it! This episode features conversations with two experts who have been applying reinforcement learning to problems in industry. First is an excerpt of my conversation with Nicolas (Nic) Hohn, Chief Data Scientist, McKinsey/QuantumBlack Australia. Nic led a team of data scientists charged with helping America’s Cup winning team, Emirates Team New Zealand, test new designs for hydrofoils – important sailing boat components that could be modified based on rules set forth by race organizers. I also include an excerpt of a conversation with Max Pumperla, Data Science Professor at IU International University of Applied Sciences, who at the time of our conversation, was also the Head of Product Research at Pathmind1, a SaaS that helps businesses use reinforcement learning in real-world applications.
Highlights from our conversation with Nic Hohn on YouTube:
- Introduction to Nic Hohn, Chief Data Scientist at Quantum Black which is part of McKinsey in Australia
“Making Boats Fly with AI on Ray” Keynote from Ray Summit
America’s Cup Project
Development, Physical Construction, and Simulation
Simulator is also a source of IP
Map a problem into the Reinforcement Learning context
How much compute are in these simulations and optimizations? What kind of scale are we talking about?
At what point did you sense this is more than a science experiment? We may have something here?
Reinforcement Learning, Recommenders, and Personalization
The right circumstances for Reinforcement Learning to be the right answer
Traditional techniques limiting, optimization no longer works, need clear idea of what you’re solving for
Offline Reinforcement Learning, Tutorials, Pure Simulation Angle
Can you imagine MLOps for RL (Reinforcement Learning)?
Could it be that people are building simulations and they don’t know that it’s RL?
What is your take on Machine Learning and Data Science in the enterprise in general? Key challenges? Trends?
Do you find you still have to do educational talks on implementing AI?
Resurgence of interest again in data, data pipelines, data quality, data infrastructure, and data engineering
Where do you fall in the centralized versus non-centralized ML?
Do I need more data scientists or more data engineers with Auto ML, the right data, and the MLR pipelines?
In terms of Responsible AI, are there concerns about bias, security, privacy, safety, and reliability?
Experiment Tracking and Management Tools
Scaling AI, MLOps, Monitoring Tools, and Model Management Tools
Traditional companies looking for solutions to buckets of activities to automate and enable AI
Repeatable, scalable, sustainable ML and AI Practice to stay ahead of competitors
- “Applications of Reinforcement Learning: Recent examples from large US companies”
- “Enterprise Applications of Reinforcement Learning: Recommenders and Simulation Modeling”
- “One Simple Chart: Demand for Reinforcement Learning Holds Steady”
- Nicolas Hohn: “Reinforcement Learning For the Win”
- Max Pumperla: “Connecting Reinforcement Learning to Simulation Software”
- Max Pumperla on “Open source Hyperparameter Tuning libraries (Hyperopt, Optuna, and Tune)”
Subscribe to our Newsletter:
We also publish a popular newsletter where we share highlights from recent episodes, trends in AI / machine learning / data, and a collection of recommendations.
 Ben Lorica was an an advisor to Pathmind.
[Image: Logistics, conveyor belt from pxfuel]