Elad Ganmor

Sparse low-order interaction network underlies a highly correlated and learnable neural population code

Information is carried in the brain by the joint activity patterns of large groups of neurons. Understanding the structure and function of population neural codes is challenging because of the exponential number of possible activity patterns and dependencies among neurons. We report here that for groups of ~100 retinal neurons responding to natural stimuli, pairwise-based models, which were highly accurate for small networks, are no longer sufficient. We show that because of the sparse nature of the neural code, the higher-order interactions can be easily learned using a novel model and that a very sparse low-order interaction network underlies the code of large populations of neurons. Additionally, we show that the interaction network is organized in a hierarchical and modular manner, which hints at scalability. Our results suggest that learnability may be a key feature of the neural code.

Shift in the Balance between Excitation and Inhibition during Sensory Adaptation of S1 Neurons

Sustained stimulation of sensory organs results in adaptation of the neuronal response along the sensory pathway. Whether or not cortical adaptation affects equally excitation and inhibition is poorly understood. We examined this question using patch recordings of neurons in the barrel cortex of anesthetized rats while repetitively stimulating the principal whisker. We found that inhibition adapts more than excitation, causing the balance between them to shift toward excitation. A comparison of the latency of thalamic firing and evoked excitation and inhibition in the cortex strongly suggests that adaptation of inhibition results mostly from depression of inhibitory synapses rather than adaptation in the firing of inhibitory cells. The differential adaptation of the evoked conductances that shifts the balance toward excitation may act as a gain mechanism which enhances the subthreshold response during sustained stimulation, despite a large reduction in excitation.

The Architecture of Functional Interaction Networks in the Retina

Sensory information is represented in the brain by the joint activity of large groups of neurons. Recent studies have shown that, although the number of possible activity patterns and underlying interactions is exponentially large, pairwise-based models give a surprisingly accurate description of neural population activity patterns. We explored the architecture of maximum entropy models of the functional interaction networks underlying the response of large populations of retinal ganglion cells, in adult tiger salamander retina, responding to natural and artificial stimuli. We found that we can further simplify these pairwise models by neglecting weak interaction terms or by relying on a small set of interaction strengths. Comparing network interactions under different visual stimuli, we show the existence of local network motifs in the interaction map of the retina. Our results demonstrate that the underlying interaction map of the retina is sparse and dominated by local overlapping interaction modules.

Intensity-Dependent Adaptation of Cortical and Thalamic Neurons Is Controlled by Brainstem Circuits of the Sensory Pathway

Current views of sensory adaptation in the rat somatosensory system suggest that it results mainly from short-term synaptic depression. Experimental and theoretical studies predict that increasing the intensity of sensory stimulation, followed by an increase in firing probability at early sensory stages, is expected to attenuate the response at later stages disproportionately more than weaker stimuli, due to greater depletion of synaptic resources and the relatively slow recovery process. This may lead to coding ambiguity of stimulus intensity during adaptation. In contrast, we found that increasing the intensity of repetitive whisker stimulation entails less adaptation in cortical neurons. In a series of recordings, from the trigeminal ganglion to the thalamus, we pinpointed the source of the unexpected pattern of adaptation to the brainstem trigeminal complex. We suggest that low-level sensory processing counterbalances later effects of short-term synaptic depression by increasing the throughput of high-intensity sensory inputs.

Near-optimal integration of orientation information across saccades

We perceive a stable environment despite the fact that visual information is essentially acquired in a sequence of snapshots separated by saccadic eye movements. The resolution of these snapshots varies-high in the fovea and lower in the periphery-and thus the formation of a stable percept presumably relies on the fusion of information acquired at different resolutions. To test if, and to what extent, foveal and peripheral information are integrated, we examined human orientation-discrimination performance across saccadic eye movements. We found that humans perform best when an oriented target is visible both before (peripherally) and after a saccade (foveally), suggesting that humans integrate the two views. Integration relied on eye movements, as we found no evidence of integration when the target was artificially moved during stationary viewing. Perturbation analysis revealed that humans combine the two views using a weighted sum, with weights assigned based on the relative precision of foveal and peripheral representations, as predicted by ideal observer models. However, our subjects displayed a systematic overweighting of the fovea, relative to the ideal observer, indicating that human integration across saccades is slightly suboptimal.

A thesaurus for a neural population code

Information is carried in the brain by the joint spiking patterns of large groups of noisy, unreliable neurons. This noise limits the capacity of the neural code and determines how information can be transmitted and read-out. To accurately decode, the brain must overcome this noise and identify which patterns are semantically similar. We use models of network encoding noise to learn a thesaurus for populations of neurons in the vertebrate retina responding to artificial and natural videos, measuring the similarity between population responses to visual stimuli based on the information they carry. This thesaurus reveals that the code is organized in clusters of synonymous activity patterns that are similar in meaning but may differ considerably in their structure. This organization is highly reminiscent of the design of engineered codes. We suggest that the brain may use this structure and show how it allows accurate decoding of novel stimuli from novel spiking patterns. DOI: http://dx.doi.org/10.7554/eLife.06134.001

How fast can we learn maximum entropy models of neural populations

Most of our knowledge about how the brain encodes information comes from recordings of single neurons. However, computations in the brain are carried out by large groups of neurons. Modelling the joint activity of many interacting elements is computationally hard because of the large number of possible activity patterns and limited experimental data. Recently it was shown in several different neural systems that maximum entropy pairwise models, which rely only on firing rates and pairwise correlations of neurons, are excellent models for the distribution of activity patterns of neural populations, and in particular, their responses to natural stimuli. Using simultaneous recordings of large groups of neurons in the vertebrate retina responding to naturalistic stimuli, we show here that the relevant statistics required for finding the pairwise model can be accurately estimated within seconds. Furthermore, while higher order statistics may, in theory, improve model accuracy, they are, in practice, harmful for times of up to 20 minutes due to sampling noise. Finally, we demonstrate that trading accuracy for entropy may actually improve model performance when data is limited, and suggest an optimization method that automatically adjusts model constraints in order to achieve good performance.

Near-optimal integration of orientation information across saccadic eye movements

Visual acuity rapidly declines with eccentricity. Consequently humans move their eyes several times per second, repeatedly placing a new target region of the scene under the high-resolution scrutiny of the fovea. But what happens to the low resolution information acquired in the periphery just before the fovea lands on its new target? Integrating the pre-saccadic information into the post-saccadic estimate would make sense, as it can only reduce the error of the internal representation. But given the discrepancy in acuity between fovea and periphery, under normal viewing conditions the pre-saccadic view will contribute only marginally to the overall accuracy of the estimate. We used the well-established theoretical and experimental framework of cue integration to examine whether, and to what extent, human observers integrate pre- and post-saccadic orientation information. Untrained observers were asked to saccade on cue to one of two pre-determined peripheral targets (5 deg. eccentricity left or right). An oriented Gabor patch was visible either before the saccade, after the saccade, or both before and after saccade. Subjects were asked to determine whether the Gabor was oriented clockwise or counter-clockwise relative to vertical. In order to maximize the potential gain from saccadic integration, the contrast of the Gabor differed between the foveal and peripheral views (unbeknownst to the observers) and was experimentally set so that, for each subject, discriminability at the foveal and peripheral locations was approximately equal. Analysis of the results indicates that subjects integrate pre- and post-saccadic orientation information near optimally, achieving better orientation discriminability when the target is visible both before and after saccade. We conclude that humans are equipped with the apparatus to integrate low-level visual information across saccadic eye movements. Meeting abstract presented at VSS 2015.

The V1 Population Gains Normalization

In this issue of Neuron, Busse et al. describe the population response to superimposed visual stimuli while Sit et al. examine the spatiotemporal evolution of cortical activation in response to small visual stimuli. Surprisingly, these two studies of V1 report that a single gain control model accounts for their results.