Abstract
Statistical regularities in the environment create prior beliefs that we rely on to optimize our behavior when sensory information is uncertain. Bayesian theory formalizes how prior beliefs can be leveraged, and has had a major impact on models of perception 1, sensorimotor function 2,3, and cognition 4. However, it is not known how recurrent interactions among neurons mediate Bayesian integration. Using a time interval reproduction task in monkeys, we found that prior statistics warp the underlying structure of population activity in the frontal cortex allowing the mapping of sensory inputs to motor outputs to be biased in accordance with Bayesian inference. Analysis of neural network models performing the task revealed that this warping was mediated by a low-dimensional curved manifold, and allowed us to further probe the potential causal underpinnings of this computational strategy. These results uncover a simple and general principle whereby prior beliefs exert their influence on behavior by sculpting cortical latent dynamics.
Acknowledgements
H.S. and N.M. are supported by the Center for Sensorimotor Neural Engineering. D.N. was supported by the Rubicon grant (446-14-008) by the Netherlands Scientific Organization and the Marie Sklodowska Curie Reintegration Grant (PredOpt 796577) by the European Union. M.J. is supported by NIH (NINDS-NS078127), the Sloan Foundation, the Klingenstein Foundation, the Simons Foundation, the McKnight Foundation, the Center for Sensorimotor Neural Engineering, and the McGovern Institute.
Author contributions
H.S. and M.J. conceived the in-vivo experiments. H.S. collected the physiology data. D.N. and M.J. conceived the in-silico experiments with recurrent neural networks. D.N. trained and simulated the networks. H.S., N.M. and D.N. analyzed the data. M.J. supervised the project. All authors were involved in writing the manuscript.