The Voytek Lab studies the role neural oscillations and neural noise play in cognition, aging, and disease. We approach this from two complementary angles: from the experimental side, we conduct human behavioral experiments while recording invasive (ECoG) and non-invasive (EEG) electrophysiological data. We combine this with large scale data-mining, data science approaches, and machine-learning techniques to discover neural features—oscillatory and otherwise—that are informative of cognitive states.
From the computational side, we build models of neural circuitry to probe the computational capabilities of neural populations to investigate how oscillations impact computation and communication. Additionally, we model how these circuits contribute to the electrical brain signals we record. We validate these models by comparing to a wide range of electrophysiological data, including single-unit recordings and intracranial field potentials from humans and animals.
Ultimately, our research goal is to bridge these two approaches and construct an understanding of cognition built on the first principles of neurophysiology. Rather than asking, “What brain regions correlate with working memory or attentional load?” we ask, “Given what we know about the computational properties of neurons and neural systems, how can neural systems interact to give rise to cognitive phenomena we equate with ‘attention’ and ‘working memory’, and what are the behavioral and cognitive limitations and consequences of these biological constraints?”
In collaboration with his wife, Jessica Bolger Voytek, Voytek built and published brainSCANr, an algorithmic approach to aggregating information from more than 2 million peer-reviewed neuroscience articles. This project was his first attempt at automated science: using machine-learning tools to algorithmically generate novel hypotheses. Our philosophy with regards to the role of data-driven approaches to neuroscience is that large scale data analytics can complement and guide in-lab experimental research, but should not replace it.