UK jobs. Bigger possibilities.

Research Associate in Generative AI and Information Theory for Communications

Hybrid in London·Full-time·Added today (last checked 17 Nov)

Our summary

Salary
£50,733 - £59,484 a year

About £3,330 - £3,750 a month after tax.

Work mode
Hybrid, London
Contract
Full-time
Closing date
Apply by Sunday 3 January 2027

Skills

Pay & benefits

What you'll get

  • The opportunity to continue your career at a world-leading institution and be part of our mission to continue science for humanity.
  • An international collaboration with leading information theorists at Texas A&M University, with regular joint meetings and funded research visits.
  • Access to GPU computing resources and a software-defined radio testbed for over-the-air experiments, and a dedicated conference travel budget.
  • Freedom to develop your own research directions, opportunities to co-supervise PhD students, and support for independent fellowship applications (e.g., RAEng, EPSRC).
  • Grow your career: gain access to Imperial's sector-leading dedicated career support for researchers as well as opportunities for promotion and progression.
  • Sector-leading salary and remuneration package (including 39 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.

The role

End-to-end learning-based constructions are becoming increasingly popular for communication systems, but what do neural networks actually learn when they are trained to communicate: do they rediscover Shannon-theoretic constructions, or something new? The forward process of a diffusion model is a Gaussian channel, so should a diffusion model be viewed as a joint source-channel decoder? When every block of a communication system is learned, is the bit still the right interface?

We are seeking a postdoctoral researcher to work on these and related questions at the interface between information theory and modern machine learning, as part of a joint NSF-EPSRC project between Imperial College London and Texas A&M University. The aim is to use information theory both as a lens to explain what learned communication systems do and as a guide to designing new ones, from analytically tractable Gaussian models to real data. Potential research directions include, but are not limited to:

These questions are timely: generative AI is reshaping how data are compressed and transmitted, and AI-native, semantic and task-oriented communication are being considered for 6G, yet principled answers about what learned systems can and cannot achieve remain scarce. You will be encouraged to develop your own research agenda within this space, from fundamental limits to over-the-air demonstrations on our software-defined radio testbed.

What you'll do

  • Generative models meet information theory: diffusion models viewed through Gaussian channels and estimation, generative priors for communication and compression at extremely low rates, and rate-distortion-perception trade-offs.
  • Opening the black box of learned codes: what neural encoders learn on tractable (e.g., vector Gaussian) sources, whether they recover known constructions such as Shannon-Kotel'nikov mappings or superposition codes, and what a "neural interface" between source and channel coding should look like.
  • Multi-user and networked settings: correlated sources over multiple-access and relay channels, and communication-efficient inference and learning with large models at the network edge.

This full-time, in-person postdoctoral position is based at Imperial College's South Kensington Campus in London, UK, and is funded for up to 24 months, starting in January 2027 or as soon as possible thereafter.

The Research Associate will be supervised by Prof. Deniz Gunduz within the Electrical and Electronic Engineering Department at Imperial College London. Co-supervision will be provided by Prof. Krishna Narayanan and Prof. Chao Tian at Texas A&M University (USA), through regular joint meetings and research visits.

  • To shape and lead your own research agenda within the themes above
  • To undertake original research of international excellence
  • To develop fundamental theory and turn it into practical learned coding and communication schemes, validated in simulation and, where appropriate, on our software-defined radio testbed
  • To write reports for submission to research sponsors
  • To present findings at international conferences and at joint project meetings with Texas A&M University
  • To publish in leading venues in information theory, machine learning and communications (e.g., IEEE Trans. IT, ISIT, IEEE JSAC, NeurIPS, ICML, ICLR)
  • To collaborate closely with researchers at Texas A&M University, and to help mentor PhD students

Education

  • Hold a PhD (or be close to completion*) in electrical engineering, computer science, mathematics, statistics, or a related field.

Experience

  • Practical experience within a research environment and / or a track record of publications at leading venues
  • Strong expertise in one or more of the following, and a willingness to learn the others:
  • Information theory (e.g., rate-distortion theory, joint source-channel coding, multi-user information theory).
  • Coding and communication theory.
  • Design, implementation and training of deep neural networks, including generative models (e.g., diffusion models, transformers).
  • Optimisation methods.
  • Experience with programming in Python and deep learning frameworks (e.g., PyTorch or JAX)
  • Desirable: prior work on learned compression, neural channel coding, joint source-channel coding, or semantic/task-oriented communication; hands-on experience with diffusion or other generative models; experience with software-defined radio or wireless prototyping.

Knowledge

  • Knowledge and research experience in one or more of the following areas: information theory, coding theory, communication theory, learning theory, optimisation, and generative models.
  • Knowledge of research methods and statistical procedures.

Skills and Abilities

  • Ability to conduct a detailed review of recent literature
  • Ability to develop and apply new concepts
  • Creative approach to problem-solving
  • Excellent verbal communication skills and the ability to deal with a wide range of people
  • Excellent written communication skills and the ability to write clearly and succinctly for publication

About Imperial College London

In their own words

Additional information

Please note that job descriptions cannot be exhaustive, and the post-holder may be required to undertake other duties, which are broadly in line with the above key responsibilities.

How to apply: please submit a CV with a full publication list, a research statement (max. 2 pages) describing how your expertise relates to the themes above, and up to three representative papers. Visa sponsorship is available for this role.

For any specific queries regarding the post please contact Prof Deniz Gunduz (d.gunduz@imperial.ac.uk)

*Candidates who have not yet been officially awarded their PhD will be appointed as Research Assistant.

All open roles at Imperial College London
WhatsApp