Compilers + ML/AI/LLM: 2 PhD positions

2 PhD studentships are available in the following general areas:
  • using compilers to improve ML/AI performance
  • using ML/AI to lift and translate binary code

    • Transformer Optimization:

    Transformer based large language models are increasingly popular but require significant resources to train and deploy. This project will explore compiler/algorithm co-design to dramatically reduce the cost on real-world hardware. One potential direction is to build on our prior work for neural architecture search [1][2].

    Neural lifting:

    Translating one language to another is a key requirement for the development of new systems. This proposal aims to lift existing source binaries to an intermediate representation (IR) using large language models and then use existing compiler technology to lower the IR. The main challenge here is accuracy and scale. One potential direction is to build on our prior work [3]][4]

    Flexibility

    The exact details are flexible depending on the candidate's interests and background.

    Eligibility

    Candidates from all countries are welcome. Under-represented groups are particularly encouraged to apply.

    Funding

    The studentships will cover all tuition fees and a tax-free stipend at the EPSRC rate 19,237 per annum for 3.5 years There are no courses or prelims required.

    Candidate Profile

    Suitable candidates will have a strong first degree in Computer Science or Mathematics and a strong interest in optimizing compilers, machine learning or verification. The exact topic of the PhD is flexible depending on the candidate's interests. We are looking for the brightest minds to pursue research in a cutting-edge arena. Candidate with Masters degrees or significant industrial experience are also strongly encouraged to apply.

    Research Partner

    The Institute for Computing Systems Architecture (ICSA) is one of seven research institutes in the School of Informatics at the University of Edinburgh. ICSA is primarily concerned with the architecture and engineering of future computing systems. Within its five research groups, ICSA covers topics which include: performance and scalability, innovative algorithms, architectures, compilers, languages, and protocols. Candidates are encouraged to contact Michael O'Boyle to informally discuss projects further. Formal application will be through the School's normal PhD application process.

    References:
    [1] Blockswap: Fisher-guided block substitution for network compression on a budget Jack Turner, Elliot J Crowley, Michael O'Boyle, Amos Storkey, Gavin Gray
    [2] Neural architecture search as program transformation exploration Jack Turner, Elliot J Crowley, Michael FP O'Boyle ASPLOS21
    [3] SLaDe: A Portable Small Language Model Decompiler for Optimized Assembly Jordi Armengol-Estape, Jackson Woodruff, Chris Cummins, Michael FP O'Boyle CGO24
    [4] Forklift: An Extensible Neural Lifter Jordi Armengol-Estape, Rodrigo CO Rocha, Jackson Woodruff, Pasquale Minervini, Michael FP O'Boyle COLM 24