A distributed, dynamic, parallel computational model: the role of noise in velocity storage.
ABSTRACT Networks of neurons perform complex calculations using distributed, parallel computation, including dynamic "real-time" calculations required for motion control. The brain must combine sensory signals to estimate the motion of body parts using imperfect information from noisy neurons. Models and experiments suggest that the brain sometimes optimally minimizes the influence of noise, although it remains unclear when and precisely how neurons perform such optimal computations. To investigate, we created a model of velocity storage based on a relatively new technique--"particle filtering"--that is both distributed and parallel. It extends existing observer and Kalman filter models of vestibular processing by simulating the observer model many times in parallel with noise added. During simulation, the variance of the particles defining the estimator state is used to compute the particle filter gain. We applied our model to estimate one-dimensional angular velocity during yaw rotation, which yielded estimates for the velocity storage time constant, afferent noise, and perceptual noise that matched experimental data. We also found that the velocity storage time constant was Bayesian optimal by comparing the estimate of our particle filter with the estimate of the Kalman filter, which is optimal. The particle filter demonstrated a reduced velocity storage time constant when afferent noise increased, which mimics what is known about aminoglycoside ablation of semicircular canal hair cells. This model helps bridge the gap between parallel distributed neural computation and systems-level behavioral responses like the vestibuloocular response and perception.
- SourceAvailable from: André van Schaik[Show abstract] [Hide abstract]
ABSTRACT: Humans and other animals behave as if we perform fast Bayesian inference underlying decisions and movement control given uncertain sense data. Here we show that a biophysically realistic model of the subthreshold membrane potential of a single neuron can exactly compute the numerator in Bayes rule for inferring the Poisson parameter of a sensory spike train. A simple network of spiking neurons can construct and represent the Bayesian posterior density of a parameter of an external cause that affects the Poisson parameter, accurately and in real time.06/2014;
- [Show abstract] [Hide abstract]
ABSTRACT: Head direction (HD) cells have been identified in a number of limbic system structures. These cells encode the animal's perceived directional heading in the horizontal plane and are dependent on an intact vestibular system. Previous studies have reported that the responses of vestibular neuron within the vestibular nuclei are markedly attenuated when an animal makes a volitional head turn compared to passive rotation. This finding presents a conundrum in that if vestibular responses are suppressed during an active head turn how is a vestibular signal propagated forward to drive and update the HD signal? This review identifies and discusses four possible mechanisms that could resolve this problem. These mechanisms are: 1) the ascending vestibular signal is generated by more than just vestibular-only neurons, 2) not all vestibular-only neurons contributing to the HD pathway have firing rates that are attenuated by active head turns, 3) the ascending pathway may be spared from the affects of the attenuation in that the HD system receives information from other vestibular brainstem sites that do not include vestibular-only cells, 4) the ascending signal is affected by the inhibited vestibular signal during an active head turn, but the HD circuit compensates and uses the altered signal to accurately update the current head direction. Future studies will be needed to decipher which of these possibilities is correct.Neuroscience 04/2014; 270. · 3.33 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: The brain makes use of noisy sensory inputs to produce eye, head, or arm motion. In most instances, the brain combines this sensory information with predictions about future events. Here, we propose that Kalman filtering can account for the dynamics of both visually guided and predictive motor behaviors within one simple unifying mechanism. Our model relies on two Kalman filters: (1) one processing visual information about retinal input; and (2) one maintaining a dynamic internal memory of target motion. The outputs of both Kalman filters are then combined in a statistically optimal manner, i.e., weighted with respect to their reliability. The model was tested on data from several smooth pursuit experiments and reproduced all major characteristics of visually guided and predictive smooth pursuit. This contrasts with the common belief that anticipatory pursuit, pursuit maintenance during target blanking, and zero-lag pursuit of sinusoidally moving targets all result from different control systems. This is the first instance of a model integrating all aspects of pursuit dynamics within one coherent and simple model and without switching between different parallel mechanisms. Our model suggests that the brain circuitry generating a pursuit command might be simpler than previously believed and only implement the functional equivalents of two Kalman filters whose outputs are optimally combined. It provides a general framework of how the brain can combine continuous sensory information with a dynamic internal memory and transform it into motor commands.Journal of Neuroscience 10/2013; 33(44):17301-13. · 6.75 Impact Factor