Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
Review
. 2017 Oct:46:84-89.
doi: 10.1016/j.conb.2017.07.006. Epub 2017 Aug 24.

Characterizing and interpreting the influence of internal variables on sensory activity

Affiliations
Review

Characterizing and interpreting the influence of internal variables on sensory activity

Richard D Lange et al. Curr Opin Neurobiol. 2017 Oct.

Abstract

The concept of a tuning curve has been central for our understanding of how the responses of cortical neurons depend on external stimuli. Here, we describe how the influence of unobserved internal variables on sensory responses, in particular correlated neural variability, can be understood in a similar framework. We suggest that this will lead to deeper insights into the relationship between stimulus, sensory responses, and behavior. We review related recent work and discuss its implication for distinguishing feedforward from feedback influences on sensory responses, and for the information contained in those responses.

PubMed Disclaimer

Figures

Figure 1
Figure 1
A computational-level description of the brain (upper path) may invoke abstract internal variables (I) that govern behavior (B) and are influenced by stimuli (S). On a neurophysiological level, we seek a similar description in terms of the responses (r) of populations of neurons (lower path) [7]. It is often useful to mix levels of abstraction, for example modeling the relationship between attention (an abstract state, or internal variable) and neural responses. Studying these relationships is limited by what quantities may be directly observed; I is not directly observable, and r is only partially observable (indicated in gray). Experiments in which an estimate of I is available from an observable correlate (S or B) lend themselves to conditional approaches, estimating p(r|I). Unsupervised approaches model the relationship between r and I, estimating both the instantaneous value of an internal variable and its influence on responses given only observations of neural responses p(r).
Figure 4
Figure 4
Cartoon of the functional relationships involved in common FB, or ‘top-down’ signals. a: FB through hierarchical structure: from the perspective of a recorded area (green), e.g. area V4 in the visual modality, lower areas within that modality are considered FF, and influences from other modalities like audition are considered FB. FB signals may also arise from intrinsically generated abstract internal variables (yellow), such as motivation, attention, or beliefs. b: FB through time: dependencies on external inputs at different times, sufficiently long in the past such that ‘memory’ about them within the FF pathway can be excluded, may also be considered FB.
Figure 2
Figure 2
a: The population tuning of two neurons to some variable, f (I1), entails responses spread along f when I1 varies according to p(I1) (each dot indicates the population response on a single trial) [31]. The linear approximation to f models this as co-variability along dfdI1dfdI1 (ellipse), which we call the ‘signature’ of I1 in this population. b: Variability in a second, independent internal variable, I2, sums (equation (2)). Given certain assumptions and enough data, unsupervised methods are able to infer I1, I2, and f(I1, I2) from observations of responses alone.
Figure 3
Figure 3
a: Visualization of equation (2) (Box 1), plots in a–b shown with respect to neurons’ preferred motion direction relative to the task. Assuming limited range structure for C0 and two independently-fluctuating internal variables that act selectively on the task-relevant directions 0 and π. b: The sum of covariance matrices in a implies the correlation (rsc) structure shown here. Red and blue rectangles indicate slices shown in the next panel. c: Cohen and Newsome (2008) [32••] measured pairwise correlations as a function of the difference in the pair’s preferred directions in two (‘same pool’, ‘different pool’) conditions. Data replotted as error bars and compared to model (lines). Note that in the absence of I1 and I2, the two conditions would have identical correlation structure.

References

    1. Felleman DJ, Van Essen DC. Distributed hierachical processing in the primate cerebral cortex. Cerebral Cortex. 1991;1:1–47. - PubMed
    1. Kohn A, Coen-Cagli R, Kanitscheider I, Pouget A. Correlations and Neuronal Population Information. Annual Review of Neuroscience. 2016;39:237–256. - PMC - PubMed
    1. Renart A, Machens CK. Variability in neural activity and behavior. Current Opinion in Neurobiology. 2014;25:211–220. - PubMed
    1. Ecker AS, Berens P, Cotton RJ, Subramaniyan M, Denfield GH, Cadwell CR, Smirnakis SM, Bethge M, Tolias AS. State dependence of noise correlations in macaque primary visual cortex. Neuron. 2014;82:235–248. Fit a shared, slow-varying latent source of additive noise on a population during anesthesia. Finds that it explains a large amount of the difference in correlated variability of V1 populations between awake and anesthetized conditions. - PMC - PubMed
    1. Rabinowitz NC, Goris RL, Cohen M, Simoncelli EP. Attention stabilizes the shared gain of V4 populations. eLife. 2015;4 Modeled V4 neural population activity as a combination of known (attentional cue, stimulus) and unknown (global gain, shared latent modulators) factors. A small number of shared latent factors explained much of the residual variance not accounted for by other sources. Task-dependence of latent factors implies interesting computational role for them. - PMC - PubMed

LinkOut - more resources