Neural population coding: combining insights from microscopic and mass signals - PubMed 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
. 2015 Mar;19(3):162-72.
doi: 10.1016/j.tics.2015.01.002. Epub 2015 Feb 7.

Neural population coding: combining insights from microscopic and mass signals

Stefano Panzeri  1 , Jakob H Macke  2 , Joachim Gross  3 , Christoph Kayser  3
Affiliations
Review

Neural population coding: combining insights from microscopic and mass signals

Stefano Panzeri et al. Trends Cogn Sci. 2015 Mar.

Abstract

Behavior relies on the distributed and coordinated activity of neural populations. Population activity can be measured using multi-neuron recordings and neuroimaging. Neural recordings reveal how the heterogeneity, sparseness, timing, and correlation of population activity shape information processing in local networks, whereas neuroimaging shows how long-range coupling and brain states impact on local activity and perception. To obtain an integrated perspective on neural information processing we need to combine knowledge from both levels of investigation. We review recent progress of how neural recordings, neuroimaging, and computational approaches begin to elucidate how interactions between local neural population activity and large-scale dynamics shape the structure and coding capacity of local information representations, make them state-dependent, and control distributed populations that collectively shape behavior.

Keywords: cross-correlation; multiscale processing; neural code; neuroimaging; oscillation; relative timing; sensory processing; state-dependent coding.

PubMed Disclaimer

Figures

Figure 1
Figure 1
The impact of neural heterogeneity on population codes. Panels A–C plot information about natural sounds carried by neurons in primate auditory cortex and are reproduced with permission from . (A) Stimulus information in randomly subsampled populations (dark blue) increases steadily with population size. However, an ‘optimized’ subpopulation built by selecting first the most informative neurons shows a much quicker saturation of information with population size (light blue), demonstrating that a small subset of neurons carries all the information available in the population. (B) The distribution of the information carried by single neurons (dots, with the line denoting a fit to an exponential distribution) shows a high heterogeneity: only a relatively small fraction of all recorded neurons carry substantial amounts of information. (C) Contour plot of stimulus information in optimized populations across variations in the read-out precision used for information decoding (x axis) and in population size (y axis), showing that the information that is lost when decoding responses at coarse temporal precision cannot be recovered by increasing population size. High values of information can only be reached with 5–10 ms precision, whatever the population size. Panels D–F: schematic of the importance of mixed nonlinear selectivity, reproduced with permission from . (D) Responses (spikes/s) of hypothetical neurons to two continuous stimulus parameters (a,b) that characterize two stimulus features. Neurons 1 and 2 are tuned to a single parameter. Neuron 3 is a linear mixed-selectivity neuron whose response is a linear combination of responses to individual parameters. The circles indicate the responses to three sensory stimuli parameterized by three combinations of the two stimulus parameters. (E) Neuron 4 is a nonlinear mixed-selectivity neuron: its response cannot be explained by a linear superposition of responses to the individual parameters. (F) The population response projected onto two sub-spaces created by neurons 1,2,3 and 1,2,4, respectively (the axes indicate the firing rates of the neurons). The left case lacks nonlinear mixed-selectivity neurons, and hence the responses to the three stimuli lie on a line and cannot be discriminated by a linear classifier (an appropriately positioned plane). The right case includes a mixed-selectivity neuron, and thus the population responses to the three stimuli lie on a plane, making the stimulus discrimination possible with a linear classifier.
Figure 2
Figure 2
Phase coding observed from single-trial analysis of mass signals. (A) Encoding of visual features in the phase of EEG signals. (Left) Information about visual stimuli carried by time-frequency EEG data reveals phase-specific coding of stimulus features (eyes and mouth of a face). (Right) Time-frequency representation of feature coding. In both panels, information carried by the EEG signal is color-coded, with warmer colors indicating higher information values. The contralateral eye was encoded by the phase at the 10 Hz signal (reproduced from [94]). (B) Phase encoding of continuous speech in auditory cortex. Theta-band (3–7 Hz) phase in bilateral auditory cortex dynamically encodes temporal variations in the envelope of continuous speech and modulates the amplitude of high-frequency gamma oscillations (reproduced from [92]). (C) Correlation between the performance (quantified as percentage of correctly decoded trials) in decoding which natural sound was presented when using auditory cortical firing rates in non-human primates, and the performance in decoding natural sounds when using theta-band EEG phase/power in humans. The same natural sound stimuli were presented to both species. Theta-band EEG phase captures better the stimulus selectivity of cortical firing rates (reproduced from with permission of Oxford University Press).
Figure 3
Figure 3
Insights about population coding from joint analysis of spiking activity and mass signals. (A) State-dependent neural firing in auditory cortex (data from [101]). The raster displays the response of one auditory cortical neuron to several repeats of a sequence of naturalistic sounds. Each spike is colored with the phase of the 4–8 Hz local field potential (LFP) (see right-hand inset for the color-coding of phase quadrants). This neuron is more active during the blue and green phase periods (as shown by the histogram of spikes at different phase quadrants, right inset), suggesting that LFP phase indexes changes in network excitability. Note that at several points during stimulus presentations there are identical firing-rate peaks that can be discriminated between each other only by the different phase ‘color’ at which they are fired. This means that the phase adds information complementary to that of spikes. (B) Schematic of how patterns of oscillatory phase coupling across multiple brain areas may coordinate anatomically dispersed neuronal ensembles (reproduced, with permission, from [106]). (Top row) How patterns of phase coupling among many different areas may synchronize anatomically dispersed neuronal ensembles. Electrodes at several locations in the monkey brain (left) identify that neurons fire preferentially in the presence of specific patterns of phases across different recording sites (indicated by network diagrams in center and right panels). Neurons that lock to the same or similar patterns of phases show the same time course of firing (center). This makes it possible to identify a pattern of LFP–LFP phase coupling across areas (center) that recruits a cell ensemble A (right) comprising the neurons A1, A2, and A3. (Bottom row) How the differential sensitivity to distinct brain rhythms or coupling patterns permits selective control of multiple coactive ensembles. (Left) Multiple functional ensembles, each spanning several brain areas, overlap in space. (Center) Interference between ensembles is minimized when each ensemble responds to a different frequency (ensembles A and C) or distinct phase-coupling pattern (ensembles A and B). (Right) Frequency and pattern selectivity permits dynamic, independent coordination of multiple coactive ensembles.
Figure I
Figure I
Schematic illustration of informative neural response features in a population code. The figure illustrates the responses of four neurons (A–D) to a set of five different stimuli each repeated twice (the geometric forms in the frames of the top row). Each small vertical line denotes one action potential. The slow wave at the bottom denotes a mass signal (e.g., an LFP) capturing the different phases of ongoing slow fluctuations in network state or excitability.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. Averbeck B.B. Neural correlations, population coding and computation. Nat. Rev. Neurosci. 2006;7:358–366. - PubMed
    1. Stanley G.B. Reading and writing the neural code. Nat. Neurosci. 2013;16:259–263. - PubMed
    1. Hebb D.O. Wiley; 1949. The Organization of Behaviour. A Neuropsychological Theory.
    1. Jacobs A.L. Ruling out and ruling in neural codes. Proc. Natl. Acad. Sci. U.S.A. 2009;106:5936–5941. - PMC - PubMed
    1. Stark E. Diode probes for spatiotemporal optical control of multiple neurons in freely moving animals. J. Neurophysiol. 2012;108:349–363. - PMC - PubMed

Publication types

LinkOut - more resources