explanation of cathode ray experiment nyu

b Encoding of sensory information according to the sampling hypothesis: the goal of a model neuron is to represent the posterior distribution (orange, middle) of the feature it encodes.

The image (left) is described as the combination of local oriented features weighted by Gaussian coefficients, further multiplied by a global modulator and corrupted by additive Gaussian noise. However, it is unclear whether this framework can also explain the surround modulation of variability 51, 52 and whether this modulation reflects the statistical properties of natural inputs.Ī Representation of the generative process of the Gaussian scale mixture (GSM) model (Methods Eq.

GSMs can explain the modulation of trial-averaged V1 responses by stimuli in the surround of the receptive field (RF 45, 46, 47, 48, 49, 50). 1a) and global features (e.g., image contrast), and that V1 neurons aim to represent the local features while discarding the global features 10, 27, 33, 43, 44. This model assumes that images are composed by local features (e.g., oriented edges Fig. To test this prediction, we consider a successful modeling framework, the Gaussian scale mixture (GSM 41, 42). Here we hypothesize that modulation of uncertainty by visual input should reflect inferences tuned to the statistics of natural images, and thus the properties of response variability should reflect the statistical structure of images. As a result, variability is reduced by stimulus onset 4 and stimulus contrast 3, 40, because of a reduction in uncertainty 10. According to this view, variability of neuronal activity reflects uncertainty about the visual input (i.e., the width of the inferred probability distribution). This theory builds on the broader idea that the brain approximates operations of probabilistic inference 37, 38, and hypothesizes that instantaneous neuronal activity represents samples from a probability distribution 20, 34, 39. In particular, some aspects of variability in primary visual cortex (V1) can be explained by the theory of neural sampling. Specifically, we propose that probabilistic inference tuned to the statistics of natural images can explain the properties of response variability in visual cortex.Īlthough normative models have typically been used to explain trial-averaged responses, they can also be used to explain response variability 20, 34, 35, 36. In this approach, we hypothesize about functional and computational principles of cortical processing, to generate predictions about cortical activity. Here we develop and test a normative model, based on efficient coding 25, 26, 27, 28 and probabilistic inference 10, 29, 30, 31, 32, 33, to explain the properties of response variability in sensory cortex. Parametric descriptive models can quantify how stimuli modulate neuronal variability 7, 11, 23, 24, but they do not address why modulation of variability occurs and what functional role it might play. Understanding the functional role of variability is at the core of the inquiry of neural coding 17, 18, 19, 20, 21, 22. This variability is modulated by several sensory 3, 4, 5, 6, 7, 8, 9, 10, 11 and non-sensory 12, 13, 14, 15, 16 factors, suggesting it may play a functional role rather than simply reflecting noise. In sensory cortex, neuronal activity is typically variable, both in the absence of sensory input and for repeated presentations of a stimulus 1, 2. Our results show that the properties of a basic aspect of cortical responses-their variability-can be explained by a probabilistic representation tuned to naturalistic inputs. Here we combine analysis of image statistics and recordings in macaque V1 to show that probabilistic inference tuned to natural image statistics explains the widely observed dependence between spike count variance and mean, and the modulation of V1 activity and variability by spatial context in images. However, it is unknown whether V1 variability reflects the statistical structure of visual inputs, as would be required for inferences correctly tuned to the statistics of the natural environment. In primary visual cortex (V1), modulation of variability by sensory and non-sensory factors supports this view. The theory of neural sampling proposes instead that variability encodes the uncertainty of perceptual inferences.

This variability is often considered detrimental to neural coding. Neuronal activity in sensory cortex fluctuates over time and across repetitions of the same input.

Explanation of cathode ray experiment nyu