13144 - Postdoctoral Research Associate

University of Edinburgh; City of Edinburgh; On-site; GBP 41,000 - 49,000; Full time

Job summary

A renowned academic institution in Scotland is seeking a postdoctoral researcher for a project on 'Numerical Analysis for Stable AI'. The successful candidate will engage in research related to algorithms in AI, requiring skills in MATLAB, Python, and numerical analysis. This role offers a collaborative environment with opportunities for professional development and is fixed term for 36 months with possible extension.

Benefits

Generous travel allowance

Collaborative research environment

Access to conferences

Qualifications

• PhD in Mathematics or related field is essential.
• Experience in numerical analysis and/or algorithms for AI is required.
• Strong coding skills in MATLAB, Python, or C are necessary.

Responsibilities

• Conduct research on algorithms in AI from a numerical analysis perspective.
• Collaborate with researchers and students on the project.
• Present findings at national and international conferences.

Skills

Research experience

Experience with numerical analysis

Strong coding skills in MATLAB

Strong coding skills in Python

Excellent academic writing skills

Education

PhD in Mathematics or relevant field

Tools

MATLAB

Python

C

Job description

Overview

Grade UE07 £41,064 – £48,822 per annum

CSE / School of Mathematics

Full time: 35 hours per week

Fixed term: 36 months (with possible extension to 48 months)

The Opportunity

We are looking for a talented and ambitious early career researcher to work on the project “Numerical Analysis for Stable AI” funded by a European Research Council Advanced Grant.

Position details

This postdoctoral position will be devoted to research on algorithms in Artificial Intelligence (AI) from the perspective of numerical analysis and computational mathematics. The overarching aim of the project is to develop new mathematical theory and algorithms to identify, quantify and, where possible, mitigate the empirically observed vulnerabilities in current AI systems. The project is led by Professor Des Higham in the School of Mathematics at the University of Edinburgh, and it will exploit links with EPCC (formerly the Edinburgh Parallel Computing Centre), the Generative AI Lab, and the Centre for Technomoral Futures.

The successful candidate will contribute to both theoretical analysis and computation experimentation (in Python, using GPUs), with the precise balance determined by the candidate’s background and interests. The mathematical tools involved will include matrix analysis, optimization, backward error analysis, condition number theory, high-dimensional analysis (concentration of measure) and stochastic computation.

A record of some high-quality mathematical reports and/or publications, as well as coding skills (e.g. in MATLAB, Python, C) are essential. You will also closely collaborate with a second postdoctoral researcher, who will be hired later in the year, a cohort of PhD students, and a range of visiting experts. Communication skills are essential for this project, as the results will be developed in collaboration with AI developers, end-users, and regulators, and with feedback from colleagues in ethics, law, social science and the creative arts.

You will be part of Edinburgh’s world-leading Applied and Computational Mathematics Group, providing a friendly and vibrant research environment. You will also have access to a generous travel allowance which will enable you to present your work at national and international conferences. There is scope for a strong candidate to shape the research direction.

Skills and attributes for success

• Research experience and some publications in at least one of numerical analysis/scientific computing and/or algorithms for AI
• Enthusiasm to investigate and quantify vulnerabilities in current AI systems
• Excellent academic writing and communication skills
• Strong coding skills in at least one of MATLAB, Python, C, and experience with, or willingness to learn, PyTorch and GPUs