Estimates of cognitive processes in decision making from neural signals by an interpretable neural network model
Qinhua Jenny Sun, Khuong Vo, Michael Nunez, Joachim Vandekerckhove, Ramesh Srinivasan, University of California, Irvine, United States
Posters 1 Poster
Pacific Ballroom H-O
Thu, 25 Aug, 19:30 - 21:30 Pacific Time (UTC -7)
Perceptual decision making is a ubiquitous cognitive phenomenon in everyday life. The drift diffusion model (DDM) is a popular model of sequential evidence accumulation that parameterizes a decision maker's speed, caution, and bias as their drift rate, boundary separation and evidencestarting poin}, respectively. Fitting the DDM requires modeling the joint distribution of choice and response time (RT) with a Wiener first-passage time (WFPT) distribution. Previous research has demonstrated that these parameters vary from trial to trial, but choice RT for a given trial, by itself, does not provide enough constraint to estimate single-trial model parameters. We have developed Decision SincNet, a novel neurocognitive model that predicts trial-level drift and boundary parameters of evidence accumulation on single trials by using EEG signals as a source of information to constrain the cognitive parameters that give rise to RT. The model is constructed with interpretable layers that can automatically identify the neural correlates of drift and boundary in time and frequency domains. Critically, we further improve the scientific interpretability of the model by introducing an attention block, allowing us to rank the learned frequency bands by importance. At each frequency band, activation maps are obtained to visualize changes in predictive EEG patterns over time. Current results and architecture of the Decision SincNet model leverage our understanding of the brain dynamics that are critical to modeling the evidence accumulation during decision making.