Research director at Meta leading the Adaptive Experimentation team. Bayesian optimization, Bayesian ML, bandits, RL, experiments, causal inference.
san francisco, ca
Joined May 2007
- Tweets795
- Following261
- Followers3.1K
- Likes2.5K
Proud to share the 1.0 release of Ax, our platform for adaptive experimentation platform. We've been using it for the past 6 years for optimizing everything from end-to-end AI systems, recommender systems, AR hardware, and material science applications.
engineering.fb.com/2025/11/1…
In our paper, we demonstrate superior optimization performance on tiny budgets for a variety of tasks, including single- and multi-objective, high-dimensional inputs, discrete inputs+noisy outcomes, including parallel/async and early termination settings. proceedings.mlr.press/v293/o…
Ax 1.0 includes simplified interfaces for efficient optimization and experiment understanding, ax.dev/docs/tutorials/gettin…, as well as APIs for aiding the development, integration, and benchmarking of novel methods for BayesOpt and active learning with BoTorch.
eytan bakshy retweeted
Big thanks to my amazing collaborators: Sebastian Ament, @MaxBalandat , @davidmeriksson , @jmhernandez233 , and @eytan !
If you are attending ICML 2025, feel free to check out our poster and have a chat with us!
Thu 11:00am - 1:30pm in East Exhibition Hall A-B, # E-1308
Did you know that concrete accounts for 8% of global CO2 emissions? I'm delighted to share that after 2 years of development, we've successfully deployed our strong, low-carbon emission "AI concrete mix" at the latest Meta datacenter.
engineering.fb.com/2025/07/1…
This work was a collaboration between the Meta Adaptive Experimentation and Infrastructure Data Science teams, UIUC, and Amrize. This mix was designed using batch multi-objective Bayesian optimization with BoTorch. Source code and data is available at github.com/facebookresearch/…
Doing research on probabilistic modeling, decision-making under uncertainty, or efficient learning and optimization? Come to NeurIPS to present your work! Submissions due August 29th. gp-seminar-series.github.io/…
Interested in AutoML, active learning, Bayesian machine learning, or Bayesian optimization? We are hiring post-docs for the Adaptive Experimentation team! Experience with BoTorch, GenAI, or causal inference is a plus! Position in NYC, SF, and Menlo Park.
metacareers.com/jobs/8316830…
Is there a bug in OpenReview for #UAI2024? We submitted 3 papers, and none of the reviewers updated their responses and there are only reject or accept decisions—no meta-review.
eytan bakshy retweeted
Software package announcement!
The Geometric Kernels package, which implements various geometric Gaussian process models from my papers - on manifolds, graphs, and similar - is now on PyPi!
It can be installed with `pip install geometric_kernels`.
Check it out!
Now hiring research scientists for the AE team!
@overleaf your "green" status is not accurate status.overleaf.com/ Multiple colleagues are having issues. I have also reported this issue via the Overleaf bug report flow.
Very much looking forward to the Adaptive Experimental Design workshop at NeurIPS this Saturday!
🚀 Just one week until the #NeurIPS2023 workshop on Adaptive Experimental Design & Active Learning in the Real World! 🔬 Exciting lineup of speakers awaits you, featuring cutting-edge insights.
Explore our fantastic schedule at:
neurips.cc/virtual/2023/work…
eytan bakshy retweeted
Two fantastic hands-on tutorials on Ax and Pearl from @eytan and @ZheqingZhu’s teams at @AIatMeta!
Definitely check out the colab tutorial file for Ax: tinyurl.com/ax-neurips2023
The Pearl colab file is linked to in their GitHub readme: github.com/facebookresearch/…
We are now accepting applications for internships on our team. If you are interested in Bayesian optimization + active learning, learning with human feedback, or applications of transformers to probabilistic modeling and causal inference please consider this opportunity!
2024 PhD Internship opportunity! Join me and my amazing colleagues on Meta's Adaptive Experimentation team to work on Bayesian optimization, probabilistic modeling / Gaussian Processes, and sample-efficient decision making: metacareers.com/jobs/9056341…
EI+its variants are the most popular acquisition functions in BayesOpt, but is no longer SoTA. We show that this is bc EI is hard to optimize numerically due to vanishing gradients, study consequences of this & propose a drop in replacement. Follow below for a sick BoTorch demo!
This Post is from an account that no longer exists. Learn more
eytan bakshy retweeted
New working paper with a fantastic team of coauthors:
we study when and how to design cluster experiments with network spillovers.
arxiv.org/pdf/2310.14983.pdf
with @lihua_lei_stat , @guido_imbens, Brian Karrer, Okke Schrijvers, Liang Shi
