Neural implementation of Bayesian inference in a sensorimotor behavior

Darlington, Timothy R; Beck, Jeffrey M.; Lisberger, Stephen G.

doi:10.1038/s41593-018-0233-y

Cited by 90 publications

(138 citation statements)

References 51 publications

Supporting

Mentioning

123

Contrasting

Order By: Relevance

“…Others have shown that task-related firing rates of single neurons before the presentation of sensory information may be modulated by prior expectations [83][84][85] , and firing rates after the presentation of sensory information may reflect Bayesian estimate of behaviorally-relevant variables 30,[86][87][88] . There have also been attempts to apply reliability-weighted linear updating schemescommonly used in cue combination studies [89][90][91] -to explain how single-neuron firing rates might combine sensory evidence with prior expectations 92,93 . However, the fact that single neurons encode various components of Bayesian models has not led to an overarching framework for understanding how networks of neurons perform Bayesian computations.…”

Section: Discussionmentioning

confidence: 99%

“…For example, natural statistics are thought to shape tuning properties and/or spontaneous activity of neurons through adjustments of synaptic connections in early sensory areas [6][7][8][9][10] . Single-unit responses in higher-level cortices also encode recent sensory events 11 , motor responses [12][13][14] , and reward probabilities [15][16][17][18] . However, an understanding of how experience-dependent neural representations enable Bayesian computations is lacking.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Bayesian computation through cortical latent dynamics

Sohn

Narain

Meirhaeghe

et al. 2018

Preprint

View full text Add to dashboard Cite

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.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Bayesian computation through cortical latent dynamics

Sohn

Narain

Meirhaeghe

et al. 2018

Preprint

View full text Add to dashboard Cite

show abstract

“…More recently an effort has been made to provide a more general theoretical framework, which is based on Bayesian inference and the reliability-based cue combination. Although the mapping of this theoretical framework onto neuronal functions remains to be elucidated, it has the clear advantage of generality, as for instance, it would encompass all forms of smooth pursuit behavior, including prediction-based and visually-guided tracking [42][43][44][45].…”

Section: Anticipatory Smooth Pursuit Eye Movements (Aspem)mentioning

confidence: 99%

Humans adapt their anticipatory eye movements to the volatility of visual motion properties

Pasturel

Montagnini

Perrinet

2019

Preprint

View full text Add to dashboard Cite

AbstractAnimal behavior constantly adapts to changes, for example when the statistical properties of the environment change unexpectedly. For an agent that interacts with this volatile setting, it is important to react accurately and as quickly as possible. It has already been shown that when a random sequence of motion ramps of a visual target is biased to one direction (e.g. right or left), human observers adapt their eye movements to accurately anticipate the target’s expected direction. Here, we prove that this ability extends to a volatile environment where the probability bias could change at random switching times. In addition, we also recorded the explicit prediction of the next outcome as reported by observers using a rating scale. Both results were compared to the estimates of a probabilistic agent that is optimal in relation to the assumed generative model. Compared to the classical leaky integrator model, we found a better match between our probabilistic agent and the behavioral responses, both for the anticipatory eye movements and the explicit task. Furthermore, by controlling the level of preference between exploitation and exploration in the model, we were able to fit for each individual’s experimental dataset the most likely level of volatility and analyze inter-individual variability across participants. These results prove that in such an unstable environment, human observers can still represent an internal belief about the environmental contingencies, and use this representation both for sensory-motor control and for explicit judgments. This work offers an innovative approach to more generically test the diversity of human cognitive abilities in uncertain and dynamic environments.Author summaryUnderstanding how humans adapt to changing environments to make judgments or plan motor responses based on time-varying sensory information is crucial for psychology, neuroscience and artificial intelligence. Current theories for how we deal with the environment’s uncertainty, that is, in response to the introduction of some randomness change, mostly rely on the behavior at equilibrium, long after after a change. Here, we show that in the more ecological case where the context switches at random times all along the experiment, an adaptation to this volatility can be performed online. In particular, we show in two behavioral experiments that humans can adapt to such volatility at the early sensorimotor level, through their anticipatory eye movements, but also at a higher cognitive level, through explicit ratings. Our results suggest that humans (and future artificial systems) can use much richer adaptive strategies than previously assumed.

show abstract

“…Preparation is an important component of voluntary movement. Throughout the motor system, many neurons show modulation of firing rate that occurs during movement preparation that is linked to parameters of the ensuing movement (Dorris et al, 1997; Glimcher and Sparks, 1992; Hanes and Schall, 1996; Munoz and Wurtz, 1995; Tanaka and Fukushima, 1998; Tanji and Evarts, 1976; Weinrich and Wise, 1982; Wurtz and Goldberg, 1972; Churchland et al, 2006a, 2006b; Darlington et al, 2018; Hanes and Schall, 1996; Mahaffy and Krauzlis, 2011; Messier and Kalaska, 2000). Importantly, many neurons that seem to be tied to movement still have preparatory activity, even though there is neither movement nor muscle activity during preparation.…”

Section: Introductionmentioning

confidence: 99%

“…Importantly, the same cells that discharge in relation to the initiation of pursuit also show an impressive ramp increase in firing rate during fixation, at a time when there is neither visual motion nor smooth eye movement (Mahaffy and Krauzlis, 2011; Tanaka and Fukushima, 1998). The preparatory activity grows over the hundreds of milliseconds preceding visual motion and the onset of pursuit, and its amplitude encodes expectation of upcoming target speed and combines with visual-motion input in a reliability-weighted manner to determine pursuit-related responses (Darlington et al, 2018).…”

Section: Introductionmentioning

confidence: 99%

Mechanisms that allow cortical preparatory activity without inappropriate movement

Darlington

Lisberger

2019

Preprint

Self Cite

View full text Add to dashboard Cite

32We reveal a novel mechanism that explains how preparatory activity can evolve in motor-related 33 cortical areas without prematurely inducing movement. The smooth eye movement region of the 34 frontal eye fields (FEFSEM) is a critical node in the neural circuit controlling smooth pursuit eye 35 movement. Preparatory activity evolves in FEFSEM during fixation in parallel with an objective 36 measure of visual-motor gain. We propose that the use of FEFSEM output as a gain signal 37 allows for preparation to progress in the pursuit system without causing movement. We also 38 show that preparatory modulation of firing rate in FEFSEM progresses in a way that predicts 39 movement, providing evidence against the "movement-null" space hypothesis of how 40 preparatory activity can progress without movement. Finally, there is partial reorganization of 41 FEFSEM population activity between preparation and movement. We propose that this 42 reorganization allows for a directionally non-specific component of preparatory visual-motor 43 gain enhancement in the pursuit system. 44 83of an all-or-none inhibitory gate created by the action of omnipause neurons in the brainstem 84 (Cohen and Henn, 1972; Keller, 1974a; Luschei and Fuchs, 1972). Ramps of preparatory activity 85 in the saccadic region of the frontal eye fields fail to trigger eye movements until the collective 86 descending activity from the frontal eye fields and superior colliculus cause the omnipause 87 4 neurons to cease firing, releasing the circuits that generate movement (Dorris et al., 1997; Hanes 88 and Schall, 1996). However, there is no evidence that pursuit system uses the all-or-none gate 89 that controls saccades (Missal and Keller, 2002; Schwartz and Lisberger, 1994). 90Here we propose a third, different, mechanism for movement prevention during motor 91 preparation. The outputs from the FEFSEM seem to control the gain of visual-motor transmission 92 for pursuit (Nuding et al., 2009; Tanaka and Lisberger, 2001). Further, visual-motor gain is 93 dialed up in preparation for an imminent smooth eye movement (Kodaka and Kawano, 2003; 94 Tabata et al., 2006). In the present paper, we show striking parallels between the progression of 95 preparatory activity in FEFSEM and the preparatory enhancement of visual-motor gain in the 96 pursuit system. We argue that the output of FEFSEM dials up visual-motor gain in preparation for 97 a behaviorally-relevant visual motion of a specific direction and speed, and that the use of 98 FEFSEM output as a sensorimotor gain signal allows preparation to proceed without causing 99 movement. We also show that preparatory activity in FEFSEM contradicts the predictions of the 100 dynamical system framework that has arisen from work in arm motor cortex (Churchland et al., 101 2010; Elsayed et al., 2016; Kaufman et al., 2014).102 103 Results 104Here, we present three primary findings to support the conclusion that FEFSEM preparatory 105 activity can evolve without causing smooth pursuit eye movement because its output d...

show abstract

Neural implementation of Bayesian inference in a sensorimotor behavior

Cited by 90 publications

References 51 publications

Bayesian computation through cortical latent dynamics

Bayesian computation through cortical latent dynamics

Humans adapt their anticipatory eye movements to the volatility of visual motion properties

Mechanisms that allow cortical preparatory activity without inappropriate movement

Contact Info

Product

Resources

About