Our work focuses on category learning and concept formation, which draws on our fundamental cognitive functions including attention, learning, and long-term semantic and episodic memory. We want to figure out how humans learn new concepts and how they gets integrated and consolidated into our long-term semantic knowledge.
We are pursuing a line of work examining how the hippocampus, and anterior/inferior temporal lobe learns and organizes information when learning new concepts and how they are consolidated into long-term semantic memory. We are building computational models to explore the cognitive and neural processes involved in concept learning and consolidation.
We are also pursuing projects on cognitive ageing, exploring how our learning and memory systems change in aging. Episodic memory systems (e.g., hippocampus) exhibit significant declines in ageing, whereas semantic conceptual knowledge systems (e.g., anterior temporal lobe) seem to be preserved. There are a lot of open questions here and lots of modelling and brain imaging work to be done in this space!
To test our questions, we use fMRI, behaviour, and cognitive computational modelling (with plans for M/EEG and trancranial stimulation studies now). Right now, we are using and combining ideas from cognitive models (e.g., clustering models like SUSTAIN; see Love et al., 2004) to deep neural networks using state-of-the-art tools in machine learning and artificial intelligence (using deep learning frameworks like PyTorch). To relate brain processes to cognitive processes in our neuroimaging work, we use analysis methods including model-based representational similarity analysis (RSA) and multivariate pattern analysis (MVPA).
Hippocampus, learning abstract spaces, cognitive maps: In the past few years, I stumbled into the idea of how the hippocampal formation might represent abstract spaces and how this relates to concepts and learning. Is there a link between place cells and grid cells found in the rat during navigation, 'concept cells' in the human hippocampus, and how we learn new concepts? In our modelling paper, we showed that the same learning process (a clustering algorithm based on concept learning models) could produce representations to support concepts as well as spatial representations resembling place and grid cells. Check out our new multilevel version here, going from behavior to neural assemblies!