CoChat

Hierarchical Models in the Brain

Open access
Friston, Karl · 2008 · PLoS Computational Biology

The DOI matches its Crossref record; 94 of the 102 references it cites were found in a registry or an index; 10 of the 81 sources its claims cite were checked for support.

Abstract

This paper describes a general model that subsumes many parametric models for continuous data. The model comprises hidden layers of state-space or dynamic causal models, arranged so that the output of one provides input to another. The ensuing hierarchy furnishes a model for many types of data, of arbitrary complexity. Special cases range from the general linear model for static data to generalised convolution models, with system noise, for nonlinear time-series analysis. Crucially, all of these models can be inverted using exactly the same scheme, namely, dynamic expectation maximization. This means that a single model and optimisation scheme can be used to invert a wide range of models. We present the model and a brief review of its inversion to disclose the relationships among, apparently, diverse generative models of empirical data. We then show that this inversion can be formulated as a simple neural network and may provide a useful metaphor for inference and learning in the brain.

Overview

CoChat Summary

## TL;DR

This paper presents hierarchical dynamic models (HDMs) as a very general framework that contains many familiar static and dynamic statistical models as special cases, all invertible with the same dynamic expectation maximization (DEM) scheme. It argues that the same inversion can be written as a simple neural network and may serve as a metaphor for inference and learning in the brain.

## Key findings

  • HDMs “subsum[e] many parametric models for continuous data,” ranging from the general linear model for static data to generalized convolution models with system noise for nonlinear time-series analysis.
  • The paper claims that these diverse models can all be inverted using the same scheme, dynamic expectation maximization (DEM), under a Laplace/mean-field variational approximation.
  • In the results, the author shows that HDMs encompass standard tools including the general linear model, expectation maximization (EM), restricted maximum likelihood (ReML), covariance component estimation, Gaussian process models, PCA, factor analysis, ICA, sparse coding, deconvolution, and state-space filtering.
  • A worked example is described in which DEM converged in “about sixteen iterations” using embedding order “n = 6” and roughness hyperparameter “γ = 4,” with the text stating that the recovered states, parameters, and hyperparameters closely matched the true values and that the true values lay largely within the 90% confidence intervals.

## Methods

  • The paper develops a hierarchical dynamic model in generalized coordinates of motion, with hidden states and causes organized across levels; the model includes Gaussian priors and Gaussian assumptions about fluctuations, and uses prediction errors at each level.
  • Inversion is based on variational Bayes under mean-field and Laplace approximations, optimizing free energy / variational action with a dynamic expectation maximization scheme that separates D-step (states), E-step (parameters), and M-step (hyperparameters).
  • The paper then maps this inversion scheme onto a neural-network-like message-passing architecture and illustrates the framework with analytic special cases and a simulation example.

## Limitations

  • The paper states that the Laplace/Gaussian approximation “restrict[s] recognition to unimodal probabilistic representations,” though it frames this as a practical simplification.
  • The text makes evident that many claims are theoretical and illustrative rather than empirically validated in this paper; it says the neuronal implementation “calls for an enormous amount of empirical verification and hypothesis testing” (evident in the text).
  • The paper also makes evident that the examples and special-case derivations cover only a “partial list” of possible models, so the treatment is not exhaustive (evident in the text).

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1371/journal.pcbi.1000211

Cites

94 of 94 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Conditional dependencies of dynamic (right) and hierarchical (left) models, shown as directed Bayesian graphs. The nodes of these graphs correspond to quantities in the model and the responses they generate. The arrows or edges indicate conditional dependencies between these quantities. The form of the models is provided, both in terms of their state-space equations (above) and in terms of the prior and conditional probabilities (below). The hierarchal structure of these models induces empirical priors; dynamical priors are mediated by the equations of generalised motion and structural priors by the hierarchical form, under which states in higher levels provide constraints on the level below.
Figure 1 Conditional dependencies of dynamic (right) and hierarchical (left) models, shown as directed Bayesian graphs. The nodes of these graphs correspond to quantities in the model and the responses they generate. The arrows or edges indicate conditional dependencies between these quantities. The form of the models is provided, both in terms of their state-space equations (above) and in terms of the prior and conditional probabilities (below). The hierarchal structure of these models induces empirical priors; dynamical priors are mediated by the equations of generalised motion and structural priors by the hierarchical form, under which states in higher levels provide constraints on the level below.CC BY · PubMed Central
Figure 2 Image representations of the precision matrices encoding temporal dependencies among the generalised motion of random fluctuations. The precision in generalised coordinates (left) and over discrete samples in time (right) are shown for a roughness of γ = 4 and seventeen observations (with an order of n = 16). This corresponds to an autocorrelation function whose width is half a time bin. With this degree of temporal correlation only a few (i.e., five or six) discrete local observations are specified with any precision.
Figure 2 Image representations of the precision matrices encoding temporal dependencies among the generalised motion of random fluctuations. The precision in generalised coordinates (left) and over discrete samples in time (right) are shown for a roughness of γ = 4 and seventeen observations (with an order of n = 16). This corresponds to an autocorrelation function whose width is half a time bin. With this degree of temporal correlation only a few (i.e., five or six) discrete local observations are specified with any precision.CC BY · PubMed Central
Figure 3 Example of estimation under a mixed-effects or hierarchical linear model. The inversion was cross-validated with expectation maximization (EM), where the M-step corresponds to restricted maximum likelihood (ReML). This example used a simple two-level model that embodies empirical shrinkage priors on the first-level parameters. These models are also known as parametric empirical Bayes (PEB) models (left). Causes were sampled from the unit normal density to generate a response, which was used to recover the causes, given the parameters. Slight differences in the hyperparameter estimates (upper right), due to a different hyperparameterisation, have little effect on the conditional means of the unknown causes (lower right), which are almost indistinguishable.
Figure 3 Example of estimation under a mixed-effects or hierarchical linear model. The inversion was cross-validated with expectation maximization (EM), where the M-step corresponds to restricted maximum likelihood (ReML). This example used a simple two-level model that embodies empirical shrinkage priors on the first-level parameters. These models are also known as parametric empirical Bayes (PEB) models (left). Causes were sampled from the unit normal density to generate a response, which was used to recover the causes, given the parameters. Slight differences in the hyperparameter estimates (upper right), due to a different hyperparameterisation, have little effect on the conditional means of the unknown causes (lower right), which are almost indistinguishable.CC BY · PubMed Central
Figure 4 Example of Factor Analysis using a hierarchical model, in which the causes have deterministic and stochastic components. Parameters and causes were sampled from the unit normal density to generate a response, which was then used for their estimation. The aim was to recover the causes without knowing the parameters, which is effected with reasonable accuracy (upper). The conditional estimates of the causes and parameters are shown in lower panels, along with the increase in free-energy or log-evidence, with the number of DEM iterations (lower left). Note that there is an arbitrary affine mapping between the conditional means of the causes and their true values, which we estimated, post hoc to show the correspondence in the upper panel.
Figure 4 Example of Factor Analysis using a hierarchical model, in which the causes have deterministic and stochastic components. Parameters and causes were sampled from the unit normal density to generate a response, which was then used for their estimation. The aim was to recover the causes without knowing the parameters, which is effected with reasonable accuracy (upper). The conditional estimates of the causes and parameters are shown in lower panels, along with the increase in free-energy or log-evidence, with the number of DEM iterations (lower left). Note that there is an arbitrary affine mapping between the conditional means of the causes and their true values, which we estimated, post hoc to show the correspondence in the upper panel.CC BY · PubMed Central
Figure 5 This schematic shows the linear convolution model used in the subsequent figure in terms of a directed Bayesian graph. In this model, a simple Gaussian ‘bump’ function acts as a cause to perturb two coupled hidden states. Their dynamics are then projected to four response variables, whose time-courses are cartooned on the left. This figure also summarises the architecture of the implicit inversion scheme (right), in which precision-weighted prediction errors drive the conditional modes to optimise variational action. Critically, the prediction errors propagate their effects up the hierarchy (c.f., Bayesian belief propagation or message passing), whereas the predictions are passed down the hierarchy. This sort of scheme can be implemented easily in neural networks (see last section and [5] for a neurobiological treatment). This generative model uses a single cause v (1), two dynamic states and four outputs y 1,…,y 4. The lines denote the dependencies of the variables on each other, summarised by the equations (in this example both the equations were simple linear mappings). This is effectively a linear convolution model, mapping one cause to four outputs, which form the inputs to the recognition model (solid arrow). The inputs to the four data or sensory channels are also shown as an image in the insert.
Figure 5 This schematic shows the linear convolution model used in the subsequent figure in terms of a directed Bayesian graph. In this model, a simple Gaussian ‘bump’ function acts as a cause to perturb two coupled hidden states. Their dynamics are then projected to four response variables, whose time-courses are cartooned on the left. This figure also summarises the architecture of the implicit inversion scheme (right), in which precision-weighted prediction errors drive the conditional modes to optimise variational action. Critically, the prediction errors propagate their effects up the hierarchy (c.f., Bayesian belief propagation or message passing), whereas the predictions are passed down the hierarchy. This sort of scheme can be implemented easily in neural networks (see last section and [5] for a neurobiological treatment). This generative model uses a single cause v (1), two dynamic states and four outputs y 1,…,y 4. The lines denote the dependencies of the variables on each other, summarised by the equations (in this example both the equations were simple linear mappings). This is effectively a linear convolution model, mapping one cause to four outputs, which form the inputs to the recognition model (solid arrow). The inputs to the four data or sensory channels are also shown as an image in the insert.CC BY · PubMed Central
Figure 6 The predictions and conditional densities on the states and parameters of the linear convolution model of the previous figure. Each row corresponds to a level, with causes on the left and hidden states on the right. In this case, the model has just two levels. The first (upper left) panel shows the predicted response and the error on this response (their sum corresponds to the observed data). For the hidden states (upper right) and causes (lower left) the conditional mode is depicted by a coloured line and the 90% conditional confidence intervals by the grey area. These are sometimes referred to as “tubes”. Finally, the grey lines depict the true values used to generate the response. Here, we estimated the hyperparameters, parameters and the states. This is an example of triple estimation, where we are trying to infer the states of the system as well as the parameters governing its causal architecture. The hyperparameters correspond to the precision of random fluctuations in the response and the hidden states. The free parameters correspond to a single parameter from the state equation and one from the observer equation that govern the dynamics of the hidden states and response, respectively. It can be seen that the true value of the causal state lies within the 90% confidence interval and that we could infer with substantial confidence that the cause was non-zero, when it occurs. Similarly, the true parameter values lie within fairly tight confidence intervals (red bars in the lower right).
Figure 6 The predictions and conditional densities on the states and parameters of the linear convolution model of the previous figure. Each row corresponds to a level, with causes on the left and hidden states on the right. In this case, the model has just two levels. The first (upper left) panel shows the predicted response and the error on this response (their sum corresponds to the observed data). For the hidden states (upper right) and causes (lower left) the conditional mode is depicted by a coloured line and the 90% conditional confidence intervals by the grey area. These are sometimes referred to as “tubes”. Finally, the grey lines depict the true values used to generate the response. Here, we estimated the hyperparameters, parameters and the states. This is an example of triple estimation, where we are trying to infer the states of the system as well as the parameters governing its causal architecture. The hyperparameters correspond to the precision of random fluctuations in the response and the hidden states. The free parameters correspond to a single parameter from the state equation and one from the observer equation that govern the dynamics of the hidden states and response, respectively. It can be seen that the true value of the causal state lies within the 90% confidence interval and that we could infer with substantial confidence that the cause was non-zero, when it occurs. Similarly, the true parameter values lie within fairly tight confidence intervals (red bars in the lower right).CC BY · PubMed Central
Figure 7 Ontology of models starting with a simple general linear model with two levels (the PCA model). This ontology is one of many that could be constructed and is based on the fact that hierarchical dynamic models have several attributes that can be combined to create an infinite number of models; some of which are shown in the figure. These attributes include; (i) the number of levels or depth; (ii) for each level, linear or nonlinear output functions; (iii) with or without random fluctuations; (iii) static or dynamic (iv), for dynamic levels, linear or nonlinear equations of motion; (v) with or without state noise and, finally, (vi) with or without generalised coordinates.
Figure 7 Ontology of models starting with a simple general linear model with two levels (the PCA model). This ontology is one of many that could be constructed and is based on the fact that hierarchical dynamic models have several attributes that can be combined to create an infinite number of models; some of which are shown in the figure. These attributes include; (i) the number of levels or depth; (ii) for each level, linear or nonlinear output functions; (iii) with or without random fluctuations; (iii) static or dynamic (iv), for dynamic levels, linear or nonlinear equations of motion; (v) with or without state noise and, finally, (vi) with or without generalised coordinates.CC BY · PubMed Central
Figure 8 Schematic detailing the neuronal architectures that encode an ensemble density on the states and parameters of one level in a hierarchical model. This schematic shows the speculative cells of origin of forward driving connections that convey prediction error from a lower area to a higher area and the backward connections that are used to construct predictions. These predictions try to explain away input from lower areas by suppressing prediction error. In this scheme, the sources of forward connections are the superficial pyramidal cell population and the sources of backward connections are the deep pyramidal cell population. The differential equations relate to the optimisation scheme detailed in the main text and their constituent terms are placed alongside the corresponding connections. The state-units and their efferents are in black and the error-units in red, with causes on the left and hidden states on the right. For simplicity, we have assumed the output of each level is a function of, and only of, the hidden states. This induces a hierarchy over levels and, within each level, a hierarchical relationship between states, where hidden states predict causes.
Figure 8 Schematic detailing the neuronal architectures that encode an ensemble density on the states and parameters of one level in a hierarchical model. This schematic shows the speculative cells of origin of forward driving connections that convey prediction error from a lower area to a higher area and the backward connections that are used to construct predictions. These predictions try to explain away input from lower areas by suppressing prediction error. In this scheme, the sources of forward connections are the superficial pyramidal cell population and the sources of backward connections are the deep pyramidal cell population. The differential equations relate to the optimisation scheme detailed in the main text and their constituent terms are placed alongside the corresponding connections. The state-units and their efferents are in black and the error-units in red, with causes on the left and hidden states on the right. For simplicity, we have assumed the output of each level is a function of, and only of, the hidden states. This induces a hierarchy over levels and, within each level, a hierarchical relationship between states, where hidden states predict causes.CC BY · PubMed Central
Figure 9 Schematic detailing the neuronal architectures that encode an ensemble density on the states and parameters of hierarchical models. This schematic shows how the neuronal populations of the previous figure may be deployed hierarchically within three cortical areas (or macro-columns). Within each area the cells are shown in relation to the laminar structure of the cortex that includes supra-granular (SG) granular (L4) and infra-granular (IG) layers.
Figure 9 Schematic detailing the neuronal architectures that encode an ensemble density on the states and parameters of hierarchical models. This schematic shows how the neuronal populations of the previous figure may be deployed hierarchically within three cortical areas (or macro-columns). Within each area the cells are shown in relation to the laminar structure of the cortex that includes supra-granular (SG) granular (L4) and infra-granular (IG) layers.CC BY · PubMed Central
Figure 10 The ensemble density and its mean-field partition. q(ϑ) is the ensemble density and is encoded in terms of the sufficient statistics of its marginals. These statistics or variational parameters (e.g., mean or expectation) change to extremise free-energy to render the ensemble density an approximate conditional density on the causes of sensory input. The mean-field partition corresponds to a factorization over the sets comprising the partition. Here, we have used three sets (neural activity, modulation and connectivity). Critically, the optimisation of the parameters of any one set depends on the parameters of the other sets. In this figure, we have focused on means or expectations µi of the marginal densities, q(ϑi) = N(ϑi: µi,Ci).
Figure 10 The ensemble density and its mean-field partition. q(ϑ) is the ensemble density and is encoded in terms of the sufficient statistics of its marginals. These statistics or variational parameters (e.g., mean or expectation) change to extremise free-energy to render the ensemble density an approximate conditional density on the causes of sensory input. The mean-field partition corresponds to a factorization over the sets comprising the partition. Here, we have used three sets (neural activity, modulation and connectivity). Critically, the optimisation of the parameters of any one set depends on the parameters of the other sets. In this figure, we have focused on means or expectations µi of the marginal densities, q(ϑi) = N(ϑi: µi,Ci).CC BY · PubMed Central

Verification

Claims the paper makes

9 Valid · 2 Not valid · 48 Not checked · 185 Own results and interpretations

122 more claims · 1 with their cited sources checked

Create a free account to see all 244 claims and to verify the 3 references still open.

See all claims — free account
  • Probably not a substantial claim. 65
  • It may state two separate claims as one. 65
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

18 more claims about this paper

Create a free account to see every claim other papers make about this one.

See all claims — free account

Citations

Cites

Research works printed but not linked: 4

  1. DJC MacKay, 1995, Free-energy minimisation algorithm for decoding and cryptoanalysis., Electron Lett, 31, 445
  2. T Ozaki, 1992, A bridge between nonlinear time-series models and nonlinear stochastic dynamical systems: A local linearization approach., Stat Sin, 2, 113
  3. ME Tipping, 2001, Sparse Bayesian learning and the Relevance Vector Machine., J Mach Learn Res, 1, 211
  4. CE Rasmussen, 1996, Evaluation of Gaussian Processes and Other Methods for Nonlinear Regression [PhD thesis]. Toronto, Canada: Department of Computer Science, University of Toronto.

Cited by

See all 973

Metadata

Authors
Friston, K.
Affiliations
Wellcome Centre for Human Neuroimaging (GB)
Corresponding author
Friston, Karl
Venue
PLoS Computational Biology
Published
2008-11-07
Volume
4
Issue
11
Pages
e1000211
ISSN
  • 1553-734X (ISSN-L)
  • 1553-7358
OpenAlex
W1977628053
Cited by
952
References
102
Citations by year
2026: 53 · 2025: 71 · 2024: 58 · 2023: 79 · 2022: 66 · 2021: 84 · 2020: 106 · 2019: 60 · 2018: 57 · 2017: 55 · 2016: 52 · 2015: 46 · 2014: 38 · 2013: 35 · 2012: 32
FWCI
31.65
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • PLoS Computational Biology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • UCL Discovery (University College London) (submittedVersion)
Funders
Wellcome Trust (088130)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Bayesian Modeling and Causal Inference
  • Physical Sciences › Computer Science › Artificial Intelligence › Gaussian Processes and Bayesian Inference
  • Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Applications
Keywords
hierarchical models, state-space model, dynamic causal modeling, model inversion, nonlinear time series analysis, general linear model, neural network inversion, brain learning
MeSH terms
Algorithms, Animals, Brain, Humans, Mental Processes, Models, Neurological, Nerve Net, Probability, Linear Models, Neural Networks, Computer, Nonlinear Dynamics

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.