Yoo, A. H., Keglovits, H., Collins, A. (2022). Lowered inter-stimulus discriminability hurts incremental contributions to learning.[pdf]
Badre, D., Bhandari, A., Keglovits, H., & Kikumoto, A. (2021). The dimensionality of neural representations for control. Current Opinion in Behavioral Sciences, 38, 20-28.[pdf]
Galeano Weber, E. M., Keglovits, H., Fisher, A., & Bunge, S. A. (2020). Insights into visual working memory precision at the feature-and object-level from a hemispheric encoding manipulation. Quarterly Journal of Experimental Psychology, 73(11), 1949-1968.[pdf]
structure of cognitive control representations
cognitive control is our brain's ability to flexibly guide thought and behavior. the human cognitive control system is especially flexible because any input stimulus can lead to any output behavior based on internally maintained goals, contexts, and rules. no other biological or technological systems yet developed can match the level of flexibility demonstrated in human behavior. and yet, we still do not have a complete mechanistic account of why control sometimes fails, in small ways when we can't multi-task infinitely, and in larger ways in psychiatric disorders like ocd. theories of cognitive control have proposed a "control representation" which is created and maintained during task completion and integrates necessary information to successfully guide behavior. in line with this theory, neural recordings in non-human primates suggest that both the content and structure of prefrontal cortext neural firing patterns correlate with successful and unsuccessful behavior. thus, parallel studies in humans exhibiting more complex, flexible behavior will enable improved characterization of control representations and their limitations.
measuring representational geometry with fmri
representational geometry, or the relationships among the firing patterns of a population of neurons across its inputs, shows important connections with complex human behavior. one key feature of a population's geometry is its dimensionality, defined as the number of unique axes required to fully specify the position of all the firing patterns of a population across all conditions of a given task. representational dimensionality has been shown to correlate with behavior in highly trained monkeys, but has not been systematically studied in humans. a major obstacle to bridging human and animal studies in this domain is that we presently lack techniques to estimate representational dimensionality non-invasively in humans, using methods like fmri and eeg. however, we have designed simulations which build from neural firing to modeling the fmri bold signal to test the strengths and limitations of various fmri designs for recovering dimensionality. factors like neural noise, correlated history effects of the bold signal, and task design all contribute to how accurately dimensionality can be recovered. we are using these simulations to design new estimation techniques and are currently applying them to human studies, which will enhance our understanding of controlled and automatic behavior.