Opinion
Discovering the Computational Relevance of Brain Network Organization

https://doi.org/10.1016/j.tics.2019.10.005Get rights and content

Highlights

  • Recent results suggest that localized functions in brain areas are specified primarily by their distributed global connectivity patterns.

  • Recent approaches go beyond stimulus/task brain mapping (and encoding/decoding) to characterize the role of brain connectivity in neural information processing.

  • We introduce network coding models as a framework encompassing neural network models optimized for task performance and those optimized for biological realism.

  • Biological and task performance constraints are complementary for aiding the search for accurate models of empirical brain function.

  • The activity flow algorithm is a core computational mechanism underlying network coding models, linking brain activity and connectivity in a biologically plausible mechanistic framework.

Understanding neurocognitive computations will require not just localizing cognitive information distributed throughout the brain but also determining how that information got there. We review recent advances in linking empirical and simulated brain network organization with cognitive information processing. Building on these advances, we offer a new framework for understanding the role of connectivity in cognition: network coding (encoding/decoding) models. These models utilize connectivity to specify the transfer of information via neural activity flow processes, successfully predicting the formation of cognitive representations in empirical neural data. The success of these models supports the possibility that localized neural functions mechanistically emerge (are computed) from distributed activity flow processes that are specified primarily by connectivity patterns.

Section snippets

Placing Brain Network Organization within a Computational Framework

A central goal of neuroscience is to understand how neural entities (such as brain regions) interact to compute cognitive functions. Historically, cognitive neuroscientists have tried to understand neural systems by mapping cognitive processes to neural entities. However, a comprehensive functional map of neural entities would still be insufficient to explain how cognition emerges through the collective interaction among these components. What could facilitate such mechanistic inferences? To

From Mapping Localized Functions to Building Network Coding Models

Network coding models facilitate understanding of the function of localized neural entities (such as brain regions) by clarifying how they send and receive information: connectivity. This also has the advantage of clarifying the role of each brain connection in computing cognitive functions. In contrast to network coding models, cognitive neuroscience has primarily mapped tasks and stimuli to activity in neurons and neural populations: function-structure mappings (Figure 1A, Key Figure) [5].

Connectionist Architectures and Cognitive Computations

Decades of neuroscience have focused on function-structure mapping. Why might incorporating brain network connectivity (and, by extension, network coding models) address this strategy’s limitations? Evidence from multiple sources suggests connectivity can provide a mechanistic explanation of how function emerges in neural networks (Box 1). First, decades of ‘connectionist’ work with artificial neural network models have demonstrated the plausibility of distributed connectivity-based processes

Additional Approaches for Mapping Cognitive Function with Connectivity

Given the strong evidence that connectivity is central to neural computation, any methods that link cognitive function with connectivity are likely to provide useful theoretical insight. In this section we focus on efforts that characterize information in distributed networks, as well as efforts that quantify how cognitive information representations quantitatively change between brain areas. In the subsequent section we will focus on network coding models, which provide a mechanistic

Network Coding Models: Computing Cognitive Information in Neural Networks

We and others have made the case that a particularly powerful framework for characterizing the functionality of brain regions is to use encoding and decoding models [12, 13, 14]. However, most uses of encoding and decoding models are designed to characterize information of interest to the experimenter and are inconsistent with how neural entities likely encode and decode task information biophysically [89,90]. An example of this are function-structure mappings that map high-level, human

Concluding Remarks

The recent proliferation of large neural data sets with rich task features has created a wealth of opportunities in functional brain mapping. However, data-driven approaches to mapping task features to neural responses largely disregard the biological mechanisms of neural information processing: distributed cognitive processing through brain network connectivity. Since the early days of connectionism, the functionality of neural entities has long been hypothesized to be embedded in its patterns

Acknowledgments

The authors would like to thank Ruben Sanchez-Romero for assistance with preparing this article. The authors acknowledge the support of the US National Institutes of Health under awards R01 AG055556 and R01 MH109520, and the Behavioral and Neural Sciences Graduate Program at Rutgers, The State University of New Jersey. The content is solely the responsibility of the authors and does not necessarily represent the official views of any of the funding agencies.

Glossary

Activity flow
a fundamental computation (Figure 2B) describing the movement of activity between neural units as a function of their connectivity (e.g., propagating spikes over axons in the brain). This is equivalent to the activation and propagation rules used in connectionist research.
Connectionist/Connectionism
a subfield within cognitive science that focuses on implementing cognitive and mental phenomena computationally through artificial neural networks.
Connectivity mapping
the quantification

References (100)

  • M.W. Cole

    Intrinsic and task-evoked network architectures of the human brain

    Neuron

    (2014)
  • M.W. Cole

    Task activations produce spurious but systematic inflation of task functional connectivity estimates

    NeuroImage

    (2019)
  • S. Saxena et al.

    Towards the neural population doctrine

    Curr. Opin. Neurobiol.

    (2019)
  • R.C. Williamson

    Bridging large-scale neuronal recordings and large-scale network models using dimensionality reduction

    Curr. Opin. Neurobiol.

    (2019)
  • J.M. Shine

    Neuromodulatory influences on integration and segregation in the brain

    Trends Cogn. Sci.

    (2019)
  • R.B. Mars

    Connectivity fingerprints: from areal descriptions to abstract spaces

    Trends Cogn. Sci.

    (2018)
  • G.R. Yang

    How to study the neural mechanisms of multiple tasks

    Curr. Opin. Behav. Sci.

    (2019)
  • J.H. Reynolds et al.

    The normalization model of attention

    Neuron

    (2009)
  • R. Kashyap

    Individual-specific fMRI-subspaces improve functional connectivity prediction of behavior

    NeuroImage

    (2019)
  • F.H. Sinz

    Engineering a less artificial intelligence

    Neuron

    (2019)
  • J.D. Power

    Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion

    NeuroImage

    (2012)
  • A.T. Reid

    Advancing functional connectivity research from association to causation

    Nat. Neurosci.

    (2019)
  • C. Craver

    Explaining the Brain

    (2007)
  • D.S. Bassett et al.

    Network neuroscience

    Nat. Neurosci.

    (2017)
  • R. Henson

    What can functional neuroimaging tell the experimental psychologist?

    Q. J. Exp. Psychol. Sect. A

    (2005)
  • R.A. Poldrack

    Handbook of Functional MRI Data Analysis

    (2011)
  • S.J. Luck

    An Introduction to the Event-Related Potential Technique

    (2014)
  • N. Kriegeskorte

    Circular analysis in systems neuroscience: the dangers of double dipping

    Nat. Neurosci.

    (2009)
  • J. Gallant

    System identification, encoding models, and decoding models, a powerful new approach to fMRI research

  • N. Kriegeskorte et al.

    Cognitive computational neuroscience

    Nat. Neurosci.

    (2018)
  • D.H. Hubel et al.

    Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex

    J. Physiol.

    (1962)
  • M.J. Arcaro

    Widespread correlation patterns of fMRI signal across visual cortex reflect eccentricity organization

    eLife

    (2015)
  • K. Kay et al.

    Bottom-up and top-down computations in high-level visual cortex

    eLife

    (2017)
  • M.W. Cole

    Global connectivity of prefrontal cortex predicts cognitive control and intelligence

    J. Neurosci. Off. J. Soc. Neurosci.

    (2012)
  • E.S. Finn

    Functional connectome fingerprinting: identifying individuals using patterns of brain connectivity

    Nat. Neurosci.

    (2015)
  • S.M. Smith

    A positive-negative mode of population covariation links brain connectivity, demographics and behavior

    Nat. Neurosci.

    (2015)
  • B. Biswal

    Functional connectivity in the motor cortex of resting human brain using echo-planar MRI

    Magn. Reson. Med.

    (1995)
  • B.T. Yeo

    The organization of the human cerebral cortex estimated by intrinsic functional connectivity

    J. Neurophysiol.

    (2011)
  • M.W. Cole

    Multi-task connectivity reveals flexible hubs for adaptive task control

    Nat. Neurosci.

    (2013)
  • F.M. Krienen

    Reconfigurable task-dependent functional coupling modes cluster around a core functional architecture

    Philos. Trans. R. Soc. B Biol. Sci.

    (2014)
  • L. Hearne

    Interactions between default mode and control networks as a function of increasing cognitive reasoning complexity

    Hum. Brain Mapp.

    (2015)
  • L.J. Hearne

    Reconfiguration of brain network architectures between resting-state and complexity-dependent cognitive reasoning

    J. Neurosci.

    (2017)
  • T. Ito

    Task-evoked activity quenches neural correlations and variability in large-scale brain systems

    bioRxiv.

    (2019)
  • C.J. Honey

    Network structure of cerebral cortex shapes functional connectivity on multiple time scales

    Proc. Natl. Acad. Sci. U. S. A.

    (2007)
  • P. Hagmann

    Mapping the structural core of human cerebral cortex

    PLoS Biol.

    (2008)
  • C.I. Bargmann et al.

    From the connectome to brain function

    Nat. Methods

    (2013)
  • D.E. Rumelhart et al.
    (1986)
  • J.D. Cohen

    On the control of automatic processes: a parallel distributed processing account of the Stroop effect

    Psychol. Rev.

    (1990)
  • V. Mante

    Context-dependent computation by recurrent dynamics in prefrontal cortex

    Nature

    (2013)
  • G.R. Yang

    Task representations in neural networks trained to perform many cognitive tasks

    Nat. Neurosci.

    (2019)
  • Cited by (0)

    View full text