Cognitively Motivated Concept Representations for Robotics
Much work in concept representation in Artificial Intelligence focuses on crisp high-level descriptions, often using logics. This allows for reasoning which is useful in certain applications. While this captures part of human abilities, there is much more: Humans can imagine instantiations of concepts and simulate them right down to the perceptual level, allowing similarities to be found which are not apparent at the high level (e.g. to facilitate analogical reasoning, like a paper can potentially act as a container). Humans can have intuitions about vague similarities between things, which guide many decisions. Humans can switch perspectives to see a situation in a different way if the need arises; i.e. the concepts applied have loose definitions that can be forced to fit situations. These abilities seem to require concepts which have: multiple alternative representations, a certain fluidity in application, and also a compositional nature, such that constituent parts of a concept have some fragment of meaning in their own right. For example a frying pan can be used to transfer liquid, or to bash someone; a paper can be used to write on or to start a fire, or to wrap some rice to carry it somewhere (this kind of creative use will not be given by a standard ontology). This project will build a more human-like representation of a set of early concepts; it will aim at toddler concepts, and study their development. This project will build a computer system for conceptual reasoning, which builds on cognitive science works that provide alternative views of concepts. While these existing works generally work in toy domains with artificial data, we will do reasoning with real data in computer vision and robotics. Although the project is to be based at Aberdeen, we will collaborate with robotics and vision experts outside Aberdeen.
The successful applicant will have a first or upper second class degree (or equivalent) in Computer Science or Informatics. Other closely related backgrounds can also be considered, like Engineering or Mathematics with computer algorithms experience (e.g. in Matlab). Knowledge of - Important: programming ability (in any language). Also Desirable: some knowledge of any of: machine learning, basic robotics, computer vision, cognitive science.
There is no funding attached to this project, it is for self-funded students only.
Formal applications can be completed online: http://www.abdn.ac.uk/postgraduate/apply. You should apply for PhD in Computing Science, to ensure that your application is passed to the correct College for processing. Please ensure that you quote the project title and supervisor on the application form. Informal academic enquiries should be directed to Dr F Guerin ([email protected]) with a copy of your current CV. All general enquiries should be directed to the Graduate School Admissions Unit ([email protected]).