Bayesian computation through cortical latent dynamics

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.

Distributed coding of stimulus magnitude in rodent prefrontal cortex

“…Our study is therefore conceptually different from the fixed interval or discrimination tasks that have been used in most of the above timing studies. Some studies with monkeys used tasks comparable to ours but focused on intervals lasting only hundreds of milliseconds (Jazayeri and Shadlen, 2015;Sohn et al, 2019). The neural activity in primate parietal and frontal cortices observed in these studies is surprisingly similar to what we found in rodent mPFC.…”

“…The neural representations of these probability distributions and the mechanism underlying the probabilistic computations has yet to be determined. An interesting solution was recently proposed based on recordings from monkey frontal cortex by Sohn et al (2019). While the animals measured time intervals in a task very similar to ours, frontal cortex activity followed low-dimensional curved state-space trajectories.…”

“…Recent work showed that when different time intervals are estimated collective neural activity evolves along very similar trajectories, but which final state is reached depends on the duration of the interval. When a time interval is reproduced trajectories of similar length are found for different stimuli but now the speed of progression decreases with duration (Gouvêa et al, 2015;Remington et al, 2018;Sohn et al, 2019;Wang et al, 2018). Since we separated time course-related from stimulus information through the demixed PCA, we could not see such stimulus-dependent effects directly.…”

“…Recent studies in non-human primates found that neural population dynamics in frontal and parietal cortices subserve magnitude estimation behavior (Jazayeri and Shadlen, 2015;Sohn et al, 2019). These studies investigated the estimation of time intervals lasting hundreds of milliseconds.…”

EEG signals index a global signature of arousal embedded in neuronal population recordings

“…microelectrode arrays and two-photon imaging) made it possible to monitor the activity of neural populations simultaneously (Cohen and Maunsell, 2009; Leavitt et al, 2017; Zohary et al, 1994). At the same time, it is becoming increasingly apparent that major insight about the neurobiological basis of cognition can be gained if one goes beyond studying single and pairwise statistics (Harvey et al, 2012; Khanna et al, 2019; Mante et al, 2013; Murray et al, 2017; Remington et al, 2018; Sadtler et al, 2014; Sohn et al, 2018; Wang et al, 2018). Furthermore, it has been suggested that low-dimensional neural activity patterns can be used to index global brain states, which influence performance on cognitive tasks.…”