The purpose of the project is to build a secure by design AI inference stack using the latest advancements in AI, security, and formal methods.
Modern AI inference systems combine complex software components, including model runtimes, memory managers, schedulers, accelerator kernels, isolation mechanisms, and distributed communication, which significantly increases their trusted computing base and attack surface, but also make any performance optimisation a challenge.
The post will be based in the Department of Computing at Imperial College London at the South Kensington Campus. The Department of Computing is a leading department of Computer Science in the UK and the world and has consistently been awarded the highest research rating. In the 2021 REF assessment, the Department claimed the top spot for computer science and informatics. In the latest QS World University Rankings, Imperial College London was recognised as the top university in Europe and the top 2 globally. Overall, Imperial College London ranked first in the UK for research outputs, first in the UK for research environment, and first for research impact among Russell Group universities.
This project aims to investigate a different approach: designing an AI inference stack in which critical security and correctness properties can be specified formally and mechanically verified. Doing so will allow us to provide end-to-end isolation guarantees, while also fearlessly implementing aggressive performance optimisations across the entire stack, from the low-level of individual kernels all the way to the request scheduler at the cluster level. The end goal is a fully verified opensource inference stack with competitive performance).
Further information on the research of Dr. Marios Kogias can be found at https://marioskogias.github.io/ .
- Research Associate: A PhD (or be close to completion) in an area pertinent to the subject area, i.e. operating systems, networking, distributed systems, cloud computing.
- Research Assistant: A strong Master’s degree in an area pertinent to the subject area, i.e. Computer Science or a related area.
- A proven track record of publications in top conferences in relevant fields, such as systems (e.g. OSDI, SOSP), PL (e.g. PLDI, POPL), verification (e.g. CAV) or top venues in closely related fields.
- Experience in at least one of the following (ideally more than one): formal verification tools (especially proof assistants), GPU programming, low-level systems engineering in languages such as C/C++/Rust, systems for ML.
- Experience in writing academic papers.
Please see job description for a full list of essential and desirable requirements.
- Candidates who have not yet been officially awarded their PhD will be hired at Research Assistant level: salary scale £45,399- £48,876 per annum.
- The opportunity to continue your career at a world-leading institution and be part of our mission to continue science for humanity.
- Grow your career: gain access to Imperial’s sector-leading dedicated career support for researchers as well as opportunities for promotion and progression.
- As a member of research staff, you have 10 development days to use to develop your skills and explore your career prospects
- Sector-leading salary and remuneration package (including 43 days off a year and generous pension schemes).
- Be part of a diverse, inclusive and collaborative work culture with various staff networks and resources to support your personal and professional wellbeing.
Full time, fixed term contract to start ASAP for 15 months
In addition to completing the online application candidates should attach:
- A full CV, with a list of all publications
- A research statement indicating what you see as interesting research issues relating to the above post and why your expertise is relevant.
For informal enquiries about the position, please email Dr. Marios Kogias ([email protected] ).
For queries regarding the application process, contact Jamie Perrins ([email protected] ).
Attached documents are available under links. Clicking a document link will initialize its download.
