Comparing perceptual and memory decision dynamics by means of machine learning Marieke K van Vugt1, Armin Brandt2, Andreas Schulze-Bonhage2 1University of Groningen, Netherlands, The; 2Epilepsy Center, University Medical Center Freiburg, Germany Evidence accumulation is typically modelled with a drift diffusion process, and neuroscientific research is conflicted about what brain areas are involved in this evidence accumulation process. In this work, we aimed to shed light on that question by using machine learning classifiers in combination with intracranially-recorded brain oscillations. Specifically, we looked at how classifier-derived decision evidence changed over time, and what brain regions carried this evidence. For a perceptual decision making task, decision evidence was primarily visible in visual areas and the anterior insula. Conversely, in a task where decisions required delayed match-to-sample memory judgments, we found accumulator-like activity mostly in parietal cortex and sensory-motor areas. We also observed that there was a large difference in the performance of different classifiers, with better performance for support vector machines that for regularized logistic regression. These findings suggest that classifiers are a useful tool for searching for model-relevant data patterns in brain activity, which can eventually be useful for obtaining single-trial estimates of model parameters such as the drift rate. -------------------------------------------------------------------- Marieke van Vugt, PhD Assistant Professor, Cognitive Modelling Group University of Groningen Bernoulliborg, room 326 Nijenborgh 9 9747 AG Groningen The Netherlands phone: +31-6-5195-4984 (cell) +31-50-363-9487 (office) +44-741-847-6518 (abroad) http://www.ai.rug.nl/~mkvanvugt <http://www.ai.rug.nl/~mkvanvugt> twitter: @mvugt m.k.van.vugt@rug.nl <mailto:m.k.van.vugt@rug.nl> / mkvanvugt@gmail.com