Probing population codes and circuit dynamics of probabilistic learning
Nuttida Rungratsameetaweemana, The Salk Institute for Biological Studies, United States; Shruti Kumar, Javier Garcia, US Combat Capabilities Development Command Army Research Laboratory, United States
Posters 3 Poster
Pacific Ballroom H-O
Sat, 27 Aug, 19:30 - 21:30 Pacific Time (UTC -7)
Adaptive learning in a complex, dynamic sensory environment is a cornerstone of natural intelligence. However, the neural mechanisms that underlie successful adaption to changes in sensory statistics is not well understood. We hypothesized that the brain actively evaluates incoming stimuli over time and constructs an expectation about the sensory statistics such that stimuli with high likelihood ('expected') are represented and processed differently from stimuli with low likelihood ('unexpected'). To tackle this question at the level of population and circuit codes, we collected human whole-brain electroencephalography (EEG) and compared it against a publicly available rodent primary visual cortex 2-photon calcium imaging data recorded as humans and mice underwent probabilistic changes in the sensory statistics over time. Specifically, humans actively monitored and reported the orientation of a coherent iso-oriented bars that changed from trial to trial in a probabilistic manner whereas mice performed a passive naturalistic orientation viewing task. Behaviorally, we found that expectation acquired through probabilistic learning in humans biased choice responses in favor of the expected stimuli. At the level of population dynamics, we found that unexpected sensory signals heighten late-stage information processes such as attentional engagement and executive control. By analyzing the rodent data, we found that this population-level computation is accompanied by increased response intensity in the primary sensory circuits when the learned expectation is violated. Together, these results suggested that violation of expectation modulates both population-level and circuit-level computations to facilitate seamless behavioral adaption in a dynamic sensory environment.