Neural implementation of Bayesian inference in a sensorimotor behavior.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Human and animal actions integrate prior experience with current sensory evidence.
• This integration often follows Bayesian principles, optimizing decisions under uncertainty.
• The neural mechanisms underlying this Bayesian-like computation, especially prior adaptation, remain incompletely understood.

Purpose of the Study:

• To investigate the neural implementation of Bayesian-like behavior in the frontal eye fields (FEFSEM).
• To examine how prior information is encoded and adapted within FEFSEM during smooth pursuit eye movements.
• To determine if FEFSEM activity can account for both prior adaptation and the generation of Bayesian-like pursuit behavior.

Main Methods:

• Recorded single-neuron activity in the FEFSEM of monkeys performing smooth-pursuit eye movements.
• Manipulated prior expectations of target speed across different experimental contexts.
• Analyzed neural responses in relation to behavioral adaptation and Bayesian computational models.

Main Results:

• Preparatory neural activity in FEFSEM encoded and adapted to changing prior speed information before target motion onset.
• FEFSEM neuronal output during pursuit initiation reflected a maximum a posteriori estimate of target speed.
• This estimate was a weighted combination of the adapted prior and incoming sensory evidence.
• FEFSEM responses were sufficient to drive adaptive prior updating and Bayesian-like pursuit.

Conclusions:

• FEFSEM provides a neural substrate for Bayesian-like computations, including the adaptation of priors.
• The region integrates prior knowledge and sensory data to generate optimal behavioral outputs, such as smooth pursuit.
• These findings elucidate the neural basis of decision-making under uncertainty and the role of FEFSEM in sensorimotor control.