Open Source |

Ax: A Platform for Adaptive Experimentation

Mon, 01 Sep 2025 00:00:00 +0000

📄 One paper accepted to AutoML 2025

Tue, 15 Jul 2025 00:00:00 +0000

Our paper was accepted to the 4th International Conference on Automated Machine Learning (AutoML 2025) in the ABCD Track. This is joint work with the at Meta CAS, with Miles Olson, Elizabeth Santorella, Sait Cakmak, and me as co-first authors.

💼 Joined Meta CAS Adaptive Experimentation

Mon, 05 Aug 2024 00:00:00 +0000

Started as a Research Scientist at Meta on the Adaptive Experimentation (AE) team within Central Applied Science (CAS), based in New York City. The team develops and maintains the open-source and frameworks for Bayesian optimization and adaptive experimentation at scale.

Ax

Thu, 01 Aug 2024 00:00:00 +0000

is an open-source platform for developed by Meta’s Adaptive Experimentation team. It provides a unified interface for , , multi-objective, and constrained optimization, built on top of .

I contribute to Ax as part of my work at Meta, with a particular focus on sample-efficient methods for , capacity management, and scaling-law-based modeling. Co-first author on (AutoML 2025).

GPflux

Wed, 01 Sep 2021 00:00:00 +0000

is a TensorFlow/Keras framework for Deep , developed at Secondmind Labs. It builds on and exposes Deep GP layers as familiar Keras building blocks, making it easier to compose deep .

Contributed during my doctoral student researcher appointment at Secondmind Labs, alongside Vincent Dutordoir, ST John, and other members of the lab.

BORE

Thu, 01 Jul 2021 00:00:00 +0000

is the reference implementation of (Tiao et al., ICML 2021). It recasts the acquisition function in as a probabilistic classification problem via , sidestepping the analytical-tractability constraints of conventional surrogate-based methods.

Developed with Aaron Klein.

AutoGluon

Sun, 01 Sep 2019 00:00:00 +0000

is an open-source toolkit from AWS that automates ML for tabular, image, and text data. During my AWS Berlin internship I was a core developer of the -based searcher module — described in and later forming the basis of .

Aboleth

Thu, 01 Jun 2017 00:00:00 +0000

is a minimalistic TensorFlow framework for scalable and approximation, focused on techniques. Built at CSIRO Data61 with Daniel Steinberg and Lachlan McCalman.

revrand

Sun, 01 Nov 2015 00:00:00 +0000

is a Python library for scalable generalized linear models with random feature approximations and stochastic gradient . Built at NICTA with Daniel Steinberg, Lachlan McCalman, Alistair Reid, and Simon O’Callaghan.