The Carlson Lab's major focus is on developing novel machine learning methodologies to facilitate and advance a diverse set of applications. Specifically, we are building tools to facilitate data-driven science, where information automatically derived from large, complex observations of “big data” are used to facilitate experimental design and hypothesis generation. We have active collaborations with a variety of collaborators where we are developing and using interpretable probabilistic models and deep learning to glean understanding from a variety of signals. The applications of our work are divers. For example, we are involved in learning treatment and diagnostic biomarkers from electrophysiological data collected during an Autism Spectrum Disorder clinical trial. In addition, our methods have previously revealed neural biomarkers of stress susceptibility in an animal model of depression. I additionally collaborate with investigators on a variety of other applied problems, including air quality estimation and computational toxicology.
A major recent methodological focus of the research group is on "little big data," which occurs where we have lots of data samples but they only come from a small number of groups; a real-world example of this is trying to understand the relationship between neural activity and behavior, where we get many repeats per individual but only from a few participants (in clinical studies) or animals. We are building machine learning methods to adapt to this situation by utilizing the "big data" but respecting the fact that there are inherent limitations in the number of individuals. These techniques are drastically improving generalization in many real-world data problems.