Information in complex network interactions
Mental processes arise via interactions between neurons and neuronal systems. My core questions require a mechanistic account of these interactions, yet we lack tools to study them. To address this need, we developed a novel class of analytical tools, termed multivariate directed connectivity analyses (MDCA), to investigate the information content of complex, directed interactions in distributed networks. While existing methods consider regions of the brain in isolation or treat interactions as unidimensional (i.e. two regions can only be "more" or "less" connected), MDCA can resolve the specific informational contents of directed node-to-node communication pathways in cognition.
We have applied these methods to fMRI data to show that the manipulation of visual imagery occurs in a fundamentally distributed, cortex-wide network. In an EEG study of action preparation, MDCA could decode the specific actions that participants were preparing, even well before those actions were carried out. We plan to investigate MDCA’s usefulness in studying other complex neural processes such as social cognition and in the development of technologies including brain-computer interfaces.
These tools apply generally to any complex information network, whether in the brain or otherwise. For example, we recently used MDCA to study the 2016 Republican U.S. presidential candidates' activity on the social media service Twitter. We found that their patterns of influence over other Twitter users initially reflected a multivariate measure of their political views, but became more correlated with "electability" indicators such as polling percentages and market predictions as the election cycle became more active.
- Decoding the information content of complex interactions in neural and social networks. Under Review. .
- 2016) Information processing in the mental workspace is fundamentally distributed. Journal of Cognitive Neuroscience 28(2):295-307. PDF (
Visual working memory and the mental workspace
Logie proposed that a key substrate for human cognition is a "mental workspace" that enables representations such as visual imagery to be flexibly constructed and manipulated (Logie, 2003, Psychol Learn Motiv). In two fMRI studies (Schlegel et al., 2013; Schlegel et al., 2015), we found that a fundamentally distributed neural network mediates the manipulation of visual imagery. Specifically, information about both mental representations and manipulations of those representations occurs throughout the cortex, is shared in common between regions, and is mediated via patterns of bidirectional connectivity throughout the network. These findings contradict traditional models of working memory that argue for functional segregation in discrete anatomical modules, suggesting instead that the component processes of visual working memory are fundamentally distributed.
We are now conducting studies evaluating whether fundamentally distributed processing is unique to visual working memory, or if a network-based approach will show that it also occurs in visual perception or other domains such as audition. We also plan to study the role of these dynamic networks in social cognitive processes such as navigating complex social networks. In an additional ongoing collaboration with Tetsuro Matsuzawa at Kyoto University, I am investigating the extent to which mental workspace abilities such as mental rotation are present in chimpanzees, our closest living evolutionary relative.
Learning and plasticity
A defining characteristic of the human brain is its lifelong plasticity (i.e. you can teach an old human new tricks). Yet, neural reorganization induced by the learning of complex cognitive skills remains poorly understood. While longitudinal learning studies are challenging, they can provide fundamental insights into the developmental origins of high level cognition. By identifying neural circuitry that changes with learning and tracks improvements in abilities, we may also gain insights into the neural basis of those abilities generally.
Large scale network interactions in the brain are determined by the structure of white matter pathways, so we hypothesized that neural reorganization in long term learning would be resolvable using diffusion tensor imaging (DTI), an MRI technique sensitive to white matter organization. In an initial longitudinal study of second language learning, we found extensive reorganization of white matter throughout cortical and subcortical regions in a group of language learners relative to controls (Schlegel et al., 2012). These changes occurred both in traditional left hemisphere language areas and in regions never previously implicated in language processing. In a subsequent study we followed visual art students as they studied drawing and painting (Schlegel et al., 2015), and found that training improved their creative thinking via the widespread reorganization of prefrontal white matter. Thus, flexibility in distributed prefrontal neural circuits mediates learning in several high level cognitive processes.