Rapid adaptation to novel tasks is a hallmark of intelligent behaviour. Humans are remarkably proficient at this, due to their capacity to generalize prior knowledge from related past experience. For example, humans can acquire a new language much more easily if it is similar to their mother tongue. At the same time, too much flexibility can also be costly. For example, the same language learner may experience “false friends” leading to overgeneralization and interference between languages. It is currently unknown how the human brain arbitrates this trade-off to form knowledge representations that are both flexible (to permit generalization) and yet compartmentalised (to minimize interference).
This project will combine functional brain imaging, computational modelling, and pattern analysis techniques from machine learning to interrogate the structure of knowledge representations that balance these demands. Experiments will be conducted to examine how human subjects learn to perform novel categorization tasks in successive training and transfer phases. Stimulus domains and task demands can repeat or change across phases to create tension between knowledge sharing and compartmentalisation. Unlike most human research, but similar to regimens in animal research and machine learning, these tasks will be performed with minimal explicit instruction, requiring subjects to learn the novel tasks via trial and error.
Experiments will be completed while subjects undergo functional brain imaging. A method known as representational similarity analysis will be used to examine the structure of neural representations that are formed as subjects acquire knowledge about novel tasks. A great virtue of representational similarity is that it can be compared across different brain imaging modalities, as well as between recorded brain signals and computational models that are trained to solve the same tasks. As such, it provides a rich analytical framework to investigate how the brain encodes knowledge.
Using these tools, the student will be given the opportunity to develop this line of research and address their own questions of interest. The student will receive in-depth training in methods and concepts from psychology, neuroscience, and machine learning, allowing them to develop a wide range of valuable skills.
For more information, please visit: https://www.birmingham.ac.uk/research/activity/mibtp/index.aspx
Application deadline: January 20, 2023