For perceptual inference, human observers do not estimate sensory uncertainty instantaneously from the current sensory signals alone, but by combining past and current sensory inputs consistent with a Bayesian learner.
Deep neural networks can be trained to automatically find mechanistic models which quantitatively agree with experimental data, providing new opportunities for building and visualizing interpretable models of neural dynamics.
Hierarchical modeling of internalizing symptoms and task performance reveals that difficulty adapting probabilistic learning to second-order uncertainty is common to anxiety and depression and holds across rewarding and punishing outcomes.
Machine learning models of coordinated hippocampal ensemble activity during sharp wave ripple activity encode structure that mirrors the place cell map expressed during exploration, and enable a new paradigm for analyzing and understanding this offline activity.
Attenuated anticipatory activity in ventromedial prefrontal cortex is modulated by dopamine D1 receptor density in nucleus accumbens, and accounts for impaired probabilistic reward learning in older adults.