Machine learning utilised for understanding cognitive behaviour

Cognitive science has been gradually shifting from the traditional hypothesis-based group-comparison approach toward utilising data analytics and machine learning tools. These tools are used for image processing, marker searching, disease diagnosis, and for predicting individual differences within large-scale multimodal neuroimaging data. These data-driven approaches are highly promising for learning about the intricate latent relationships between brain and behaviour. 

Despite this, data-learned relationships are usually opaque with large feature spaces, producing difficulties in interpreting the results in terms of understanding how the brain works and what changes occur when something goes wrong. A HKBU team in conjunction with collaborators at Humboldt-Universität zu Berlin and the University of Oldenburg are aiming to integrate and synergise capacity for the long-term development of big-data brain studies through the project “Interpretable machine learning aided understanding complex brain structural and functional interactions underlying the spectra of individual differences in cognitive behaviour.”

The team has so far identified core and extended reading-related brain networks whose features can predict the reading ability of individuals (NeuroImage, 2020), and used human brain complexity profiles to predict general cognitive ability using statistical modelling (Cerebral Cortex Communications, 2020). Furthermore, an artificial recurrent neural network model to train timing tasks and study how timing and non-timing information are coded by the emergent sequential network dynamics and feedforward structure has been developed (PNAS, 2020), and several statistical machine learning and linear regression models have been built to study the brain-behaviour relationship. 

In future, the aforementioned models will be used to study children and adults with varying levels of autistic traits; this combined with clinical assessment results, social cognition task performance, and brain signal data in the model analyses will examine the question of the brain-behaviour relationship relating to autism spectrum disorders. 

The Project Investigator, Prof Changsong Zhou, Associate Head of Physics, is joined by Computer Science’s Prof Yiu Ming Cheung and Dr Ann M. Lui of the Department of Education Studies at HKBU. The team is complemented by German collaborators Prof Werner Sommer of Humboldt-Universität zu Berlin and Prof Andrea Hildebrandt of the University of Oldenburg.