#139 Efficient Bayesian Optimization in PyTorch, with Max Balandat

5 snips

Aug 20, 2025

Max Balandat, who leads the modeling and optimization team at Meta, discusses the fascinating world of Bayesian optimization and the BoTorch library. He shares insights on the seamless integration of BoTorch with PyTorch, enhancing flexibility for researchers. The conversation delves into the significance of adaptive experimentation and the impact of LLMs on optimization. Max emphasizes the importance of effectively communicating uncertainty to stakeholders and reflects on the transition from academia to industry, highlighting collaboration in research.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

INSIGHT

Flexible Research-Focused Optimization

BoTorch targets researchers and advanced users who need modular, flexible Bayesian optimization tools.
Its unique value is full PyTorch differentiability and Monte Carlo acquisition functions for rapid experimentation.

ADVICE

Prefer Differentiable MC Acquisition

Use Monte Carlo acquisition functions when you can sample differentiably from your model posterior.
This lets you swap surrogate models quickly and leverage gradient-based optimization for acquisition tuning.

INSIGHT

Differentiability As A Design Anchor

Writing BoTorch in PyTorch enables end-to-end differentiability and easy integration with other PyTorch modules.
That choice accelerates research and lets users plug in encoders or pretrained networks seamlessly.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

Proudly sponsored by PyMC Labs, the Bayesian Consultancy. Book a call, or get in touch!

Intro to Bayes Course (first 2 lessons free)
Advanced Regression Course (first 2 lessons free)

Our theme music is « Good Bayesian », by Baba Brinkman (feat MC Lars and Mega Ran). Check out his awesome work!

Visit our Patreon page to unlock exclusive Bayesian swag ;)

Takeaways:

BoTorch is designed for researchers who want flexibility in Bayesian optimization.
The integration of BoTorch with PyTorch allows for differentiable programming.
Scalability at Meta involves careful software engineering practices and testing.
Open-source contributions enhance the development and community engagement of BoTorch.
LLMs can help incorporate human knowledge into optimization processes.
Max emphasizes the importance of clear communication of uncertainty to stakeholders.
The role of a researcher in industry is often more application-focused than in academia.
Max's team at Meta works on adaptive experimentation and Bayesian optimization.

Chapters:

08:51 Understanding BoTorch

12:12 Use Cases and Flexibility of BoTorch

15:02 Integration with PyTorch and GPyTorch

17:57 Practical Applications of BoTorch

20:50 Open Source Culture at Meta and BoTorch's Development

43:10 The Power of Open Source Collaboration

47:49 Scalability Challenges at Meta

51:02 Balancing Depth and Breadth in Problem Solving

55:08 Communicating Uncertainty to Stakeholders

01:00:53 Learning from Missteps in Research

01:05:06 Integrating External Contributions into BoTorch

01:08:00 The Future of Optimization with LLMs

Thank you to my Patrons for making this episode possible!

Yusuke Saito, Avi Bryant, Ero Carrera, Giuliano Cruz, James Wade, Tradd Salvo, William Benton, James Ahloy, Robin Taylor,, Chad Scherrer, Zwelithini Tunyiswa, Bertrand Wilden, James Thompson, Stephen Oates, Gian Luca Di Tanna, Jack Wells, Matthew Maldonado, Ian Costley, Ally Salim, Larry Gill, Ian Moran, Paul Oreto, Colin Caprani, Colin Carroll, Nathaniel Burbank, Michael Osthege, Rémi Louf, Clive Edelsten, Henri Wallen, Hugo Botha, Vinh Nguyen, Marcin Elantkowski, Adam C. Smith, Will Kurt, Andrew Moskowitz, Hector Munoz, Marco Gorelli, Simon Kessell, Bradley Rode, Patrick Kelley, Rick Anderson, Casper de Bruin, Philippe Labonde, Michael Hankin, Cameron Smith, Tomáš Frýda, Ryan Wesslen, Andreas Netti, Riley King, Yoshiyuki Hamajima, Sven De Maeyer, Michael DeCrescenzo, Fergal M, Mason Yahr, Naoya Kanai, Aubrey Clayton, Jeannine Sue, Omri Har Shemesh, Scott Anthony Robson, Robert Yolken, Or Duek, Pavel Dusek, Paul Cox, Andreas Kröpelin, Raphaël R, Nicolas Rode, Gabriel Stechschulte, Arkady, Kurt TeKolste, Marcus Nölke, Maggi Mackintosh, Grant Pezzolesi, Joshua Meehl, Javier Sabio, Kristian Higgins, Matt Rosinski, Bart Trudeau, Luis Fonseca, Dante Gates, Matt Niccolls, Maksim Kuznecov, Michael Thomas, Luke Gorrie, Cory Kiser, Julio, Edvin Saveljev, Frederick Ayala, Jeffrey Powell, Gal Kampel, Adan Romero, Will Geary, Blake Walters, Jonathan Morgan, Francesco Madrisotti, Ivy Huang, Gary Clarke, Robert Flannery, Rasmus Hindström, Stefan, Corey Abshire, Mike Loncaric, David McCormick, Ronald Legere, Sergio Dolia, Michael Cao, Yiğit Aşık, Suyog Chandramouli and Adam Tilmar Jakobsen.

Links from the show:

Max on Linkedin: https://www.linkedin.com/in/maximilian-balandat-b5843946/
Max on GitHub: https://github.com/Balandat
BoTorch – Bayesian Optimization in PyTorch: https://botorch.org/
BoTorch – A Framework for Efficient Monte-Carlo Bayesian Optimization: https://arxiv.org/pdf/1910.06403
Ax – A higher level, user-friendly black-box optimization tool that heavily leverages BoTorch: https://ax.dev/
Ax – A Platform for Adaptive Experimentation: https://openreview.net/forum?id=U1f6wHtG1g#discussion
PyTorch: https://docs.pytorch.org/docs/stable/index.html
GPyTorch: https://gpytorch.ai/

Transcript

This is an automatic transcript and may therefore contain errors. Please get in touch if you're willing to correct them.