Sheheryar Zaidi

I am a Staff Research Scientist at Google DeepMind.

My research goal is to build AI systems capable of reasoning and pushing the scientific frontier. My current interests include reinforcement learning, large language models, and AI-driven scientific discovery.

Much of my work has focused on fundamentally understanding neural networks (architectures, training techniques, and generalization) and using that understanding to build scalable deep learning systems to solve problems, particularly scientific ones.

I did my PhD in deep learning at Oxford with Yee Whye Teh and Arnaud Doucet, as an Aker Scholar and Google PhD Fellow. Before that, I studied math: a BA at Oxford (top 3, Junior Mathematical Prize) and a master’s at Cambridge (Part III, distinction).

My first name can be tricky to pronounce: the short version is Sheh (pronounced “Sha”). 🦔

---

Publications

Pre-training via Denoising for Molecular Property Prediction  
Sheheryar Zaidi*, Michael Schaarschmidt*, James Martens, Hyunjik Kim, Yee Whye Teh, Alvaro Sanchez-Gonzalez, Peter Battaglia, Razvan Pascanu, Jonathan Godwin
*Equal contribution.
Spotlight presentation at ICLR, 2023
arXiv   GitHub   Poster

When Does Re-initialization Work?  
Sheheryar Zaidi*, Tudor Berariu*, Hyunjik Kim, Jörg Bornschein, Claudia Clopath, Yee Whye Teh, Razvan Pascanu
*Equal contribution.
Spotlight presentation at I Can’t Believe It’s Not Better Workshop at NeurIPS, 2022
Published in PMLR volume 187
arXiv   Poster

LieTransformer: Equivariant Self-Attention for Lie Groups
Michael Hutchinson*, Charline Le Lan*, Sheheryar Zaidi*, Emilien Dupont, Yee Whye Teh, Hyunjik Kim
*Equal contribution.
ICML, 2021
arXiv   GitHub

Provably Strict Generalisation Benefit for Equivariant Models
Bryn Elesedy, Sheheryar Zaidi
ICML, 2021
arXiv

Neural Ensemble Search for Uncertainty Estimation and Dataset Shift
Sheheryar Zaidi*, Arber Zela*, Thomas Elsken, Chris Holmes, Frank Hutter, Yee Whye Teh
*Equal contribution.
NeurIPS, 2021
Oral presentation at ICML 2020 Worshop on Uncertainty & Robustness in Deep Learning
arXiv   GitHub

Effectiveness and resource requirements of test, trace and isolate strategies for COVID in the UK
Bobby He*, Sheheryar Zaidi*, Bryn Elesedy*, Michael Hutchinson*, Andrei Paleyes*, Guy Harling, Anne M Johnson, Yee Whye Teh
*Equal contribution.
Royal Society Open Science, 2021
DOI

Efficient Bayesian Inference of Instantaneous Reproduction Numbers at Fine Spatial Scales, with an Application to Mapping and Nowcasting the Covid-19 Epidemic in British Local Authorities
Yee Whye Teh, Avishkar Bhoopchand, Peter Diggle, Bryn Elesedy, Bobby He, Michael Hutchinson, Ulrich Paquet, Jonathan Read, Nenad Tomasev, Sheheryar Zaidi
Royal Statistical Society’s Covid-19 Task Force: Special Topic Meeting on R/local R/transmission, 2021
PDF   Website

Target–Aware Bayesian Inference: How to Beat Optimal Conventional Estimators
Tom Rainforth*, Adam Goliński*, Frank Wood, Sheheryar Zaidi
*Equal contribution.
JMLR, 2020
URL