Generative artificial intelligence (AI) has radically changed the way, in which humans interact with technology. From generating aesthetically pleasing paintings via word prompts to generating new protein sequences that aid drug discovery, generative AI tools have already drastically reduced the time, resources, and human involvement devoted to synthesizing many kinds of data. The advertised PhD project will evaluate the visual representations that humans and computer vision systems generate when viewing AI-generated images and develop general methods for the use of AI-generated content in scientific research.
Generative adversarial networks (GANs) are one powerful tool to artificially generate image content. Briefly, a generator network learns to generate sample images, and a discriminator network tries to discriminate between real and artificially generated samples, guiding the generator network to produce samples that are more difficult to discriminate in the next round. This arms race between the generator and the discriminator networks leads to the generation of samples that, ultimately, perfectly resemble the real samples—at least from the discriminator network’s point of view.
Contemporary GANs can generate a nearly infinite number of images of remarkable quality, holding exceptional promise for their application in various realms of human activity. However, questions regarding the suitability of GAN-generated images for human use remain under-explored.
The advertised PhD project will address those questions, by comparing how humans and object-classification networks represent GAN-generated stimuli. For example, the project will assess the suitability of current network-based measures of GAN quality as proxies for how humans represent these images at the neural and behavioural levels. Additionally, the PhD project will assess the utility of GANs, by developing methods that deploy GAN-generated stimuli in applied settings. For example, the project will develop adaptive learning algorithms that aim to speed up the development of human visual expertise, by leveraging the flexibility with which GANs can generate stimuli.
The Successful Candidate
The PhD project will require the use of techniques related to Machine Learning, Computer Vision, and Cognitive Science. Candidates should have a strong background in one of these fields, and ideally (but not necessarily) will have some experience in one of the other two fields.
The successful candidate will benefit from truly interdisciplinary training by a supervisory team composed of experts from human psychology, cognitive neuroscience, and computer science. Students with a cognitive science background will acquire cutting-edge skills related to the training and deployment of generative AI tools, whereas students with a computer science background will acquire expertise in cognitive testing inside and outside the laboratory. The successful candidate will be affiliated with both the School of Psychology and the School of Computer Science and Informatics at Cardiff University. They will be embedded within dynamic research teams comprising several postdocs and PhD students at both schools and will have access to the infrastructure provided by both Schools, as well as the broader University-wide infrastructure such as the world-class computing facilities of Supercomputing Wales.
The Host Organisations
Both the School of Psychology as well as the School of Computer Science and Informatics will host the project.
The School of Psychology is one of the largest and most successful in the UK (http://www.cf.ac.uk/psych/). The School’s excellent standard of research has been recognised in every Research Excellence Framework exercise, where Cardiff is consistently among the top 10 universities for its world-leading research in psychology, psychiatry, and neuroscience. Our world-class facilities offer unique opportunities for complementary and collaborative studies across methodologies to address novel research questions (https://www.cardiff.ac.uk/psychology/research/facilities).
The School of Computer Science and Informatics is a research-led school in one of the UK’s premier universities with a reputation for excellent teaching and for providing real-world relevant courses(https://www.cardiff.ac.uk/computer-science). In the Research Excellent Framework 2021 assessment, 96% of the School’s research was deemed world-leading or internationally excellent, being recognised as addressing real-world societal and business challenges. The School also works with over 100 academic and industrial partners in the UK and abroad, and their cutting-edge research attracts millions of pounds in funding.
For informal inquiry and discussion of the project, please contact Victor Navarro ([Email Address Removed]) or one of the co-supervisors: Christoph Teufel ([Email Address Removed]), Walter Colombo ([Email Address Removed]), or Hantao Liu ([Email Address Removed]).
Home students are UK Nationals and EU students who can satisfy UK residency requirements (students must have been in the UK for >3 years before the start of the course).
As only a limited number of studentships are available across the Open School competition and a very high standard of applications is typically received, the successful applicants are likely to have a very good first degree (a First or Upper Second class BSc Honours or equivalent) and/or be distinguished by having relevant research experience.