Decision-Making Computations in Neuronal Networks

Recent advances in neuroscience provide a new perspective on how categorical decisions are computed in the brain and how the choice circuits performing such computations are organized. This knowledge was instrumental in developing the mechanistic model of the choice circuit presented in this dissertation. The proposed model was based on the recurrent attractor network of binary neurons. The model produced biologically realistic network dynamics and behaviorally realistic predictions. The model predictions were consistent with the results of the dots motion discrimination experiments which were used to validate other decision-making models in the past. The model was applied in the investigation of the impact of alpha oscillations on the decision process, speed accuracy tradeoff (SAT), and Weber’s Law. The model predictions were qualitatively consistent with experimental studies and at the same time challenged some of the established views on the neuronal mechanisms of decision-making. In particular, our analysis revealed that the SAT mechanism based on the gain modulation hypothesis, the leading view on how SAT is implemented in the brain, was inconsistent with experimental results. The study on Weber’s Law showed a mild violation of the logarithmic relation between stimulus level and discrimination across the entire range of stimulus intensities and a convex relation between stimulus intensities and Weber fractions. The proposed model also produced new testable predictions. In particular, it showed the existence of the optimal level of the amplitude of alpha oscillations that enhanced the discriminatory power of the decision system above the baseline associated with the system performance with no oscillations. It also suggested the hypothesis that the violation of Weber’s Law in the low intensity range could be linked to SAT effects. These predictions could be tested in the future, lead to new discoveries, and support or falsify the proposed modeling approach.