Aaron Zhao
Computer Scientist & AI Researcher
Download CV Professional WorkTree
2022 -- Present
Assistant Professor, Imperial College London
Department of EEE. Leading the
DeepWok Lab, focused on efficient AI, hardware-algorithm co-design, and ML security. The team publishes consistently at top-tier ML venues (NeurIPS, ICML, ICLR) and also at leading security (USENIX) and hardware/systems venues (MLSys, ISCA). Leading a lab of more than 10 PhD and 10 Master students, with a collective PI and Co-I funding of above £10M.
Sep 2025 -- Present
Building the first secure AI agentic flows.
2022 -- Present
Visiting Researcher, University of Cambridge
2021 -- 2022
Junior Research Fellow, St John's College, Cambridge
College research fellowship in computer science.
2017 -- 2022
PhD in Computer Science, University of Cambridge
Supervised by
Prof. Robert Mullins. Thesis: Software and Hardware Co-design for Efficient Neural Networks.
Dec 2021 -- Jun 2022
AI/ML Scholar & Part-time Researcher, Apple
Selected as an Apple Scholar in AI/ML (2020). Worked on hardware-aware neural network optimization, efficient inference for on-device ML, and building security filters.
Jun 2019 -- Jan 2020
Research Intern & Part-time Researcher, Microsoft Research New England
Jun -- Oct 2018
Research Intern, Microsoft Research Redmond
Intern on
Project Brainwave. Investigated novel arithmetic for training transformers and delivered the first transformer trained with MXINT, which later became NVFP on Blackwell GPUs. Filed
8 US patents.
Jun -- Oct 2017
Research Intern, Microsoft Research Cambridge
Intern on
Project Sirius. Helped deliver a prototype of the Top of Rack switch.
2016 -- 2017
MPhil in Advanced Computer Science, University of Cambridge
Computer Laboratory.
2013 -- 2016
BEng in Electrical & Electronic Engineering, Imperial College London
Research
My research sits at the intersection of hardware, algorithms, and security in machine learning. On the efficient AI side, I work on algorithmic designs — quantization, pruning, architecture search — as well as building AI accelerators. Our open-source framework MASE enables cross-stack ML acceleration research, and PLENA is our AI inference accelerator system. On the AI security side, I study adversarial robustness, architectural backdoors, and model collapse in recursively trained systems (published in Nature).
I review and serve as Area Chair for ICLR, ICML, NeurIPS, CVPR, and other venues.
Awards
- Accelerating Foundation Models Research Award (2023)
- Junior Research Fellowship, St John's College, University of Cambridge (2021)
- Apple Scholar in AI/ML (2020)
- EPSRC International Doctoral Studentship
Initiatives
Teaching
I teach two courses at Imperial College London (2023–2025):
The best way to reach me is email: a.zhao@imperial.ac.uk.
I am currently not taking new PhD students. I occasionally host visiting PhD students, but only when the research topic is a very strong match.
For student projects (MPhil, Part II/III, final year projects, summer internships, and similar): I only work with students from Imperial College London or the University of Cambridge — if you are from elsewhere, I am unlikely to reply. Due to capacity limits, day-to-day supervision comes from a PhD student or RA in the DeepWok Lab or the AIxSIM group, with me providing additional support. Please reach out directly to a member of these groups first and secure their agreement to supervise you before contacting me — I will not be able to take you on without a day-to-day supervisor already in place.
