Maruan Al-Shedivat

Maruan Al-Shedivat is a machine learning researcher whose work runs from meta-learning and federated learning to protein-ligand structure prediction. Raised in Moscow, he studied physics at Lomonosov Moscow State University and computer science at KAUST, then completed a PhD in Carnegie Mellon's Machine Learning Department under Eric Xing. His 2017 paper Continuous Adaptation via Meta-Learning in Nonstationary and Competitive Environments, which introduced the RoboSumo environment, won one of the best paper awards at ICLR 2018. He held a Google PhD Fellowship from 2019, proposed federated posterior averaging in 2020, and as Head of Machine Learning Research at Genesis Molecular AI led the team behind Pearl, a protein-ligand structure model whose 2025 technical report showed it surpassing AlphaFold 3 on public benchmarks. As of October 2026, he is building in stealth.

Timeline

2016–2017

Deep kernels with recurrent structure

In October 2016, he posts Learning Scalable Deep Kernels with Recurrent Structure with Andrew Gordon Wilson, Yunus Saatchi, Zhiting Hu and Eric P. Xing. The paper proposes GP-LSTM, a Gaussian process model whose kernels carry the inductive biases of LSTM recurrent networks, and appears in the Journal of Machine Learning Research in 2017.
2017

RoboSumo and meta-learning

He spends time at OpenAI in 2017. In October 2017, he posts Continuous Adaptation via Meta-Learning in Nonstationary and Competitive Environments with Trapit Bansal, Yuri Burda, Ilya Sutskever, Igor Mordatch and Pieter Abbeel; the paper develops a gradient-based meta-learning algorithm for changing and adversarial settings and designs the RoboSumo multi-agent competitive environment.
2018

An ICLR 2018 best paper

In 2018, his work on meta-learning for continuous adaptation wins one of the best paper awards at ICLR.
2018–2019

CMLH fellow at Carnegie Mellon

Raised in Moscow, he holds a bachelor's degree in physics from Lomonosov Moscow State University and a master's degree in computer science from KAUST in Saudi Arabia. As a PhD student in Carnegie Mellon's Machine Learning Department, advised by Eric Xing, he works on probabilistic approaches to multitask learning and is a 2018 fellow of the Center for Machine Learning and Health, with research on interpretable machine learning for digital healthcare.
2019–2021

A Google PhD Fellowship

In 2019, he is one of 54 students worldwide chosen as a Google Ph.D. fellow; the two-year award includes tuition, a yearly stipend and a Google Research mentor. He also spends time at Google Research from 2018 to 2020.
2020–2021

Federated posterior averaging

In October 2020, he posts Federated Learning via Posterior Averaging with Jennifer Gillenwater, Eric Xing and Afshin Rostamizadeh. The paper formulates federated learning as posterior inference, introduces the FedPA algorithm, which generalizes federated averaging (FedAvg), and appears at ICLR 2021.
2025

Pearl at Genesis Molecular AI

As Head of Machine Learning Research at Genesis Molecular AI, he leads the team and research behind Pearl, a foundation model for protein-ligand cofolding. The technical report appears in October 2025, and Pearl surpasses AlphaFold 3 on the Runs N' Poses and PoseBusters benchmarks, with gains of 14.5% and 14.2% over the next best model.
As of October 2026

Building in stealth

After Genesis, he is building in stealth. He describes his scientific interest as autonomous systems that solve problems at the intersection of AI and science.