The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Causal Models in Practice at Lyft with Sean Taylor - #486
May 24, 2021
Sean Taylor, Staff Data Scientist at Lyft Rideshare Labs, shares his journey from lab director to hands-on innovator. He dives into the moonshot approaches his team takes towards marketplace experimentation and forecasting. The conversation highlights the significance of causality in their modeling efforts and the challenges of balancing supply and demand. Moreover, he discusses the application of neural networks for decision-making, emphasizing collaboration and the transformation of traditional statistical methods to drive business insights.
AI Snips
Chapters
Books
Transcript
Episode notes
Stats-Oriented Approach
- Sean Taylor values the pragmatic, decision-focused approach of statistics.
- He sees AI/ML as tools for improvement, drawing on statistics' rich history of practical application.
Forecasting at Lyft
- Forecasting at Lyft involves a human-in-the-loop system for market management.
- They forecast supply and demand, considering the impact of their decisions, like pricing and incentives.
Causal Models for Planning
- Lyft's forecasting system uses causal models, representing their business with nodes and variables.
- They use differentiable programming to create plans optimizing business objectives directly.
