Mar 15–18, 2020

Introducing GIG: A new method for explaining any ensemble ML model

Jay Budzik (Zest AI)
2:35pm3:15pm Wednesday, March 18, 2020
Location: LL21 E/F

Who is this presentation for?

Data scientists or analysts




Organizations are increasingly adopting ML to run key business functions. The best-performing models combine diverse model types (neural networks and trees) into stacked ensembles. But explaining these hybrid models with accuracy and precision has been impossible—until now.

Jay Budzik details the new explainability math called GIG that Zest developed. You’ll learn how GIG overcomes the limitations of and extends the commonly used methods shapley additive explanations (SHAP) and integrated gradients. Citing real-world examples, Jay also demonstrates how GIG makes it possible to explain complex ML models so they’re safe to use in high-stakes applications such as finance, healthcare, and defense.

Prerequisite knowledge

  • Familiarity with machine learning, including common techniques, such as neural networks, decision trees, logistic regression, etc.

What you'll learn

  • Understand why using hybrid ML models produces better results than single-model techniques and why allocating credit is difficult when combining discrete and continuous functions
  • Explore the limitations of SHAP and integrated gradients in explaining ensembled models
  • Learn how measure theory resolves the problems of assigning feature importance
Photo of Jay Budzik

Jay Budzik

Zest AI

Jay Budzik is the chief technology officer at ZestFinance, where he oversees Zest’s product and engineering teams. His passion for inventing new technologies—particularly in data mining and AI—has played a central role throughout his career. Previously, he held various positions, including founding an AI enterprise search company, helping major media organizations apply AI and machine learning to expand their audiences and revenue, and developed systems that process tens of trillions of data points. Jay has a PhD in computer science from Northwestern University.

Leave a Comment or Question

Help us make this conference the best it can be for you. Have questions you'd like this speaker to address? Suggestions for issues that deserve extra attention? Feedback that you'd like to share with the speaker and other attendees?

Join the conversation here (requires login)

Contact us

For conference registration information and customer service

For more information on community discounts and trade opportunities with O’Reilly conferences

Become a sponsor

For information on exhibiting or sponsoring a conference

For media/analyst press inquires