CoChat

Stochastic Dynamics Underlying Cognitive Stability and Flexibility

Open access
Ueltzhöffer, Kai, Armbruster-Genç, Diana J. N., Fiebach, Christian J. · 2015 · PLOS Computational Biology

The DOI matches its Crossref record; 96 of the 98 references it cites were found in a registry or an index.

Abstract

Cognitive stability and flexibility are core functions in the successful pursuit of behavioral goals. While there is evidence for a common frontoparietal network underlying both functions and for a key role of dopamine in the modulation of flexible versus stable behavior, the exact neurocomputational mechanisms underlying those executive functions and their adaptation to environmental demands are still unclear. In this work we study the neurocomputational mechanisms underlying cue based task switching (flexibility) and distractor inhibition (stability) in a paradigm specifically designed to probe both functions. We develop a physiologically plausible, explicit model of neural networks that maintain the currently active task rule in working memory and implement the decision process. We simplify the four-choice decision network to a nonlinear drift-diffusion process that we canonically derive from a generic winner-take-all network model. By fitting our model to the behavioral data of individual subjects, we can reproduce their full behavior in terms of decisions and reaction time distributions in baseline as well as distractor inhibition and switch conditions. Furthermore, we predict the individual hemodynamic response timecourse of the rule-representing network and localize it to a frontoparietal network including the inferior frontal junction area and the intraparietal sulcus, using functional magnetic resonance imaging. This refines the understanding of task-switch-related frontoparietal brain activity as reflecting attractor-like working memory representations of task rules. Finally, we estimate the subject-specific stability of the rule-representing attractor states in terms of the minimal action associated with a transition between different rule states in the phase-space of the fitted models. This stability measure correlates with switching-specific thalamocorticostriatal activation, i.e., with a system associated with flexible working memory updating and dopaminergic modulation of cognitive flexibility. These results show that stochastic dynamical systems can implement the basic computations underlying cognitive stability and flexibility and explain neurobiological bases of individual differences.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1371/journal.pcbi.1004331

Cites

96 of 96 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig 1 Conceptualized potential landscape. Attractors can be visualized as basins in a potential landscape. The depth of the basin indicates the stability of the associated attractor, i.e., the amount of external forcing or noise that is needed to change the state of the system (here schematically represented by a red ball). Here we show a hypothetical one-dimensional potential landscape of a quite flexible system (dotted line) where only a small potential barrier separates two neighboring basins of attraction, a very stable but less flexible system (dashed line), and an intermediate system (solid line).
Fig 1 Conceptualized potential landscape. Attractors can be visualized as basins in a potential landscape. The depth of the basin indicates the stability of the associated attractor, i.e., the amount of external forcing or noise that is needed to change the state of the system (here schematically represented by a red ball). Here we show a hypothetical one-dimensional potential landscape of a quite flexible system (dotted line) where only a small potential barrier separates two neighboring basins of attraction, a very stable but less flexible system (dashed line), and an intermediate system (solid line).CC BY · PubMed Central
Fig 2 Schematic illustration of the task. 80% of the trials required a response to a single digit presented above the fixation cross, i.e., to decide whether it is odd or even (baseline task; parity judgment). In 20% of the trials, a second digit appeared below the fixation cross. Subjects had to ignore this bottom digit if it was darker then the upper digit and continue to respond to the upper digit, indicating its parity (distractor condition). If the bottom digit was brighter than the upper digit, subjects had to switch the task rule, now indicating if the target was greater than or less than five (magnitude judgment), and apply that rule to the bottom digit (task switch).
Fig 2 Schematic illustration of the task. 80% of the trials required a response to a single digit presented above the fixation cross, i.e., to decide whether it is odd or even (baseline task; parity judgment). In 20% of the trials, a second digit appeared below the fixation cross. Subjects had to ignore this bottom digit if it was darker then the upper digit and continue to respond to the upper digit, indicating its parity (distractor condition). If the bottom digit was brighter than the upper digit, subjects had to switch the task rule, now indicating if the target was greater than or less than five (magnitude judgment), and apply that rule to the bottom digit (task switch).CC BY · PubMed Central
Fig 3 Model architecture and dynamics. (A) Original architecture of the network: Rules are represented by two selective populations (R1, R2) embedded in a pool of nonselective excitatory (NS) and inhibitory (I) neurons. Decisions are implemented by the same architecture of four decision selective populations. The interplay of strong local excitation and global inhibition creates winner-take-all dynamics in both modules, leading to a stable state of globally low spiking activity and stable states corresponding to the high activity of a single selective population. (B) Schematic of the stochastic dynamics of reduced system: The rule module is reduced two a two dimensional dynamical system with three stable attracting states: A spontaneous low-activity state and two rule-selective states of high activity. The decision module is reduced to a three-dimensional nonlinear drift-diffusion process, starting at the center of a tetrahedron, where each corner represents one of the four possible choice alternatives. A drift towards one of the corners is created by a combination of direct stimulus input, which favors one of the yellow edges, and biasing input from the selective populations of the rule module, which favors either the red or the blue edge of the tetrahedron. (C) Dynamics of the rule module: The rule module is reduced to a two-dimensional system whose phase space is represented by the synaptic gating variables (S1, S2) of the two rule selective populations. The model is set up in a way in which the two rule representing states as well as the spontaneous state are stable. Trajectories (red) started at an arbitrary point explore the phase space, due to the noise, but tend to stay close to one of the attractors, due to the deterministic dynamics. (D) Slice through a potential landscape of the rule module. This cross-section following a path connecting the two rule attractors via the spontaneous state illustrates the potential barrier that has to be overcome for transitioning from one state to another. Theoretically, all two-dimensional paths are possible to transition from the rule 1 to the rule 2 state. However, due to its dynamical properties, the system stays in a valley that connects the two rule attractors via the basin of attraction of the spontaneous state. The cross-sections along those paths resemble closely the one-dimensional potential sketches often found in the literature discussing the possible effects of attractor stability on cognition [18,19]; compare Fig 1. (E) Dynamics of the decision module: The figure shows the evolution of a spherical volume in the three dimensional phase space centered around the origin, which corresponds to the spontaneous state of the decision network. The trajectories rapidly converge to one of the four directions corresponding to the winner-take-all states of one of the decision selective populations. Due to the symmetry of the system, these directions form the corners of a symmetrical tetrahedron. (F) Threshold independence of decision module: Shown are the projections of a single three-dimensional decision process onto the four directions corresponding to the increase of a single selective population and the simultaneous decrease in all others. Due to the nonlinear dynamics, the trajectory rapidly converges into one of these directions and then diverges to infinity within finite time. This yields decision times that are insensitive to the exact value of the threshold, as long as it is placed far enough from the spontaneous state.
Fig 3 Model architecture and dynamics. (A) Original architecture of the network: Rules are represented by two selective populations (R1, R2) embedded in a pool of nonselective excitatory (NS) and inhibitory (I) neurons. Decisions are implemented by the same architecture of four decision selective populations. The interplay of strong local excitation and global inhibition creates winner-take-all dynamics in both modules, leading to a stable state of globally low spiking activity and stable states corresponding to the high activity of a single selective population. (B) Schematic of the stochastic dynamics of reduced system: The rule module is reduced two a two dimensional dynamical system with three stable attracting states: A spontaneous low-activity state and two rule-selective states of high activity. The decision module is reduced to a three-dimensional nonlinear drift-diffusion process, starting at the center of a tetrahedron, where each corner represents one of the four possible choice alternatives. A drift towards one of the corners is created by a combination of direct stimulus input, which favors one of the yellow edges, and biasing input from the selective populations of the rule module, which favors either the red or the blue edge of the tetrahedron. (C) Dynamics of the rule module: The rule module is reduced to a two-dimensional system whose phase space is represented by the synaptic gating variables (S1, S2) of the two rule selective populations. The model is set up in a way in which the two rule representing states as well as the spontaneous state are stable. Trajectories (red) started at an arbitrary point explore the phase space, due to the noise, but tend to stay close to one of the attractors, due to the deterministic dynamics. (D) Slice through a potential landscape of the rule module. This cross-section following a path connecting the two rule attractors via the spontaneous state illustrates the potential barrier that has to be overcome for transitioning from one state to another. Theoretically, all two-dimensional paths are possible to transition from the rule 1 to the rule 2 state. However, due to its dynamical properties, the system stays in a valley that connects the two rule attractors via the basin of attraction of the spontaneous state. The cross-sections along those paths resemble closely the one-dimensional potential sketches often found in the literature discussing the possible effects of attractor stability on cognition [18,19]; compare Fig 1. (E) Dynamics of the decision module: The figure shows the evolution of a spherical volume in the three dimensional phase space centered around the origin, which corresponds to the spontaneous state of the decision network. The trajectories rapidly converge to one of the four directions corresponding to the winner-take-all states of one of the decision selective populations. Due to the symmetry of the system, these directions form the corners of a symmetrical tetrahedron. (F) Threshold independence of decision module: Shown are the projections of a single three-dimensional decision process onto the four directions corresponding to the increase of a single selective population and the simultaneous decrease in all others. Due to the nonlinear dynamics, the trajectory rapidly converges into one of these directions and then diverges to infinity within finite time. This yields decision times that are insensitive to the exact value of the threshold, as long as it is placed far enough from the spontaneous state.CC BY · PubMed Central
Fig 4 Reproduction of behavioral data. (A/B) Decision probabilities assuming the presentation of a stimulus with odd parity in the baseline condition and a stimulus with odd parity above and a stimulus with magnitude greater than 5 below the fixation cross for the distractor and switch condition for two representative subjects (Subjects 26 and 9). Empirical data for each subject is drawn in black, simulated data generated from the fitted models are drawn in orange. Despite the overall very good quality of the fit, one can note the comparatively better fit in the baseline condition. This is due to the higher contribution of the associated multinomial factor to the Bayesian posterior distribution, which strongly depends on the number of experimental trials. (C/D) Cumulative distribution function of reaction times for correct baseline, distractor, and switch trials for the same subjects (empirical data: black; simulated/fitted data: orange). Note again the relatively low resolution of the empirical cumulative density functions in the distractor and switch conditions, due to the relatively low number of trials. (E) Group statistics describing the general goodness of fit across all subjects. Cumulative distribution function of p-values obtained by individual exact multinomial tests and two-sample Kolmogorov-Smirnov-tests of the Null hypothesis that the behavioral data was drawn from distributions whose parameters were derived from the corresponding fitted model (blue). For comparison, a homogeneous distribution on [0,1] is shown in red. The uncorrected significance threshold (p = .05) is shown as solid green line. Note that only 3% of the performed tests fell under the significance threshold. Almost half of the tests reached p = 1.
Fig 4 Reproduction of behavioral data. (A/B) Decision probabilities assuming the presentation of a stimulus with odd parity in the baseline condition and a stimulus with odd parity above and a stimulus with magnitude greater than 5 below the fixation cross for the distractor and switch condition for two representative subjects (Subjects 26 and 9). Empirical data for each subject is drawn in black, simulated data generated from the fitted models are drawn in orange. Despite the overall very good quality of the fit, one can note the comparatively better fit in the baseline condition. This is due to the higher contribution of the associated multinomial factor to the Bayesian posterior distribution, which strongly depends on the number of experimental trials. (C/D) Cumulative distribution function of reaction times for correct baseline, distractor, and switch trials for the same subjects (empirical data: black; simulated/fitted data: orange). Note again the relatively low resolution of the empirical cumulative density functions in the distractor and switch conditions, due to the relatively low number of trials. (E) Group statistics describing the general goodness of fit across all subjects. Cumulative distribution function of p-values obtained by individual exact multinomial tests and two-sample Kolmogorov-Smirnov-tests of the Null hypothesis that the behavioral data was drawn from distributions whose parameters were derived from the corresponding fitted model (blue). For comparison, a homogeneous distribution on [0,1] is shown in red. The uncorrected significance threshold (p = .05) is shown as solid green line. Note that only 3% of the performed tests fell under the significance threshold. Almost half of the tests reached p = 1.CC BY · PubMed Central
Fig 5 Generation of fMRI BOLD timecourses from fitted models. (A) Simulated energy consumption of the rule module. Simulated trials were sorted by condition and decision taken, and the average integral of the sum of the spiking rates r 1+r 2 over the course of a trial was calculated. This estimate of the energy consumption was normalized relative to the average spiking rate integral of a correct baseline trial. (B) Depending on the condition and decision recorded in the individual participants’ behavioral logs, the corresponding energy estimates were placed on a timeline. The resulting time course was z-scored. (C) The normalized timecourse estimating the neural energy consumption was then convolved with a canonical hemodynamic response function (as derived from the SPM8 software package), resulting in (D) a predictor timeseries representing the activation of the rule module over the course of the experiment, individually for each participant. (E) This regressor was entered into a multi-univariate GLM of the functional MRI data, together with regressors for each individual ambiguous trial and with six motion regressors derived from the coregistration of the functional MRI volumes (see Methods section, “Generation and Localization of fMRI Timeseries from Fitted Models”).
Fig 5 Generation of fMRI BOLD timecourses from fitted models. (A) Simulated energy consumption of the rule module. Simulated trials were sorted by condition and decision taken, and the average integral of the sum of the spiking rates r 1+r 2 over the course of a trial was calculated. This estimate of the energy consumption was normalized relative to the average spiking rate integral of a correct baseline trial. (B) Depending on the condition and decision recorded in the individual participants’ behavioral logs, the corresponding energy estimates were placed on a timeline. The resulting time course was z-scored. (C) The normalized timecourse estimating the neural energy consumption was then convolved with a canonical hemodynamic response function (as derived from the SPM8 software package), resulting in (D) a predictor timeseries representing the activation of the rule module over the course of the experiment, individually for each participant. (E) This regressor was entered into a multi-univariate GLM of the functional MRI data, together with regressors for each individual ambiguous trial and with six motion regressors derived from the coregistration of the functional MRI volumes (see Methods section, “Generation and Localization of fMRI Timeseries from Fitted Models”).CC BY · PubMed Central
Fig 6 Localization of the rule module. Using the predicted BOLD time courses for the modeled rule module, we were able to localize it to the left inferior frontal junction (IFJ) and the left intraparietal sulcus (IPS), both known to be crucially involved in task switching and distractor inhibition. Weaker activity is observed in the superior frontal gyrus (SFG). The single-voxel threshold of p = 10–7 corresponds to p = 0.05 Bonferroni corrected for the total number of voxels.
Fig 6 Localization of the rule module. Using the predicted BOLD time courses for the modeled rule module, we were able to localize it to the left inferior frontal junction (IFJ) and the left intraparietal sulcus (IPS), both known to be crucially involved in task switching and distractor inhibition. Weaker activity is observed in the superior frontal gyrus (SFG). The single-voxel threshold of p = 10–7 corresponds to p = 0.05 Bonferroni corrected for the total number of voxels.CC BY · PubMed Central
Fig 7 Reconstructed potential landscapes. (A/B) Color map of the reconstructed potential U = -lnP ss on the phase space spanned by the synaptic gating variables (S1, S2) of the rule selective populations for two representative subjects (subjects 26 and 9). The green line indicates the transition from the rule 1 to the rule 2 attractor that minimizes the path-integral action. The red line corresponds to a transition in the opposite direction. (C/D) Surface plot of the reconstructed potential landscapes. Note the deeper and steeper basins of attraction for subject 9. (E/F) Plot of the potential along the transition from the rule 1 to the rule 2 attractor that minimizes the path-integral action (green) and back (red). The individually fitted noise parameters (σ rule) for each subject were scaled by a factor of 10 for easier visualization.
Fig 7 Reconstructed potential landscapes. (A/B) Color map of the reconstructed potential U = -lnP ss on the phase space spanned by the synaptic gating variables (S1, S2) of the rule selective populations for two representative subjects (subjects 26 and 9). The green line indicates the transition from the rule 1 to the rule 2 attractor that minimizes the path-integral action. The red line corresponds to a transition in the opposite direction. (C/D) Surface plot of the reconstructed potential landscapes. Note the deeper and steeper basins of attraction for subject 9. (E/F) Plot of the potential along the transition from the rule 1 to the rule 2 attractor that minimizes the path-integral action (green) and back (red). The individually fitted noise parameters (σ rule) for each subject were scaled by a factor of 10 for easier visualization.CC BY · PubMed Central
Fig 8 Relationship between physiological parameters and dynamic potential landscape of the rule module. (A) Potential landscape of the rule module on the phase space spanned by the synaptic gating variables (S1, S2) of the two rule-selective populations with standard parameters s NMDA = 1.0,s GABA = 1.0,σ rule = 0.1. (B-D) Changes in the potential landscape of the rule module relative to that standard parameters, i.e., depending on individual increases of (B) s NMDA = 1.005, (C) s GABA = 1.005, and (D) σ rule = 0.15. (E) Potential along the paths corresponding to the minimal action transition from the spontaneous state to a high-activity state for the parameter combinations from A-D. The minimal actions required for the transitions are given in the legend. (F) Potential along the paths corresponding to the minimal action transition from a high-activity state to the spontaneous state for the parameter combinations from A-D. The minimal actions required for the transitions are given in the legend. (G) Potential along the paths corresponding to the minimal action transition between the two high-activity states for the parameter combinations from A-D. The minimal actions required for the transitions are given in the legend.
Fig 8 Relationship between physiological parameters and dynamic potential landscape of the rule module. (A) Potential landscape of the rule module on the phase space spanned by the synaptic gating variables (S1, S2) of the two rule-selective populations with standard parameters s NMDA = 1.0,s GABA = 1.0,σ rule = 0.1. (B-D) Changes in the potential landscape of the rule module relative to that standard parameters, i.e., depending on individual increases of (B) s NMDA = 1.005, (C) s GABA = 1.005, and (D) σ rule = 0.15. (E) Potential along the paths corresponding to the minimal action transition from the spontaneous state to a high-activity state for the parameter combinations from A-D. The minimal actions required for the transitions are given in the legend. (F) Potential along the paths corresponding to the minimal action transition from a high-activity state to the spontaneous state for the parameter combinations from A-D. The minimal actions required for the transitions are given in the legend. (G) Potential along the paths corresponding to the minimal action transition between the two high-activity states for the parameter combinations from A-D. The minimal actions required for the transitions are given in the legend.CC BY · PubMed Central
Fig 9 Correlation of the differential recruitment of brain regions in the switch vs. distractor contrast with the individual attractor stability of high activity rule states. (A) Middle frontal gyrus. p = 0.05 cluster threshold, corrected for whole-brain multiple comparisons. (B) Thalamostriatal network. p = 0.05 cluster threshold, corrected for multiple comparisons within an anatomical mask comprising bilateral thalamus and caudate nucleus (red overlay). (C) Eigenvariate of the middle frontal gyrus cluster in the flexibility versus stability condition versus individual minimum action associated with a transition between the rule attractors. (D) Eigenvariate of the thalamostriatal network in the flexibility versus stability condition versus individual minimum action associated with a transition between the rule attractors.
Fig 9 Correlation of the differential recruitment of brain regions in the switch vs. distractor contrast with the individual attractor stability of high activity rule states. (A) Middle frontal gyrus. p = 0.05 cluster threshold, corrected for whole-brain multiple comparisons. (B) Thalamostriatal network. p = 0.05 cluster threshold, corrected for multiple comparisons within an anatomical mask comprising bilateral thalamus and caudate nucleus (red overlay). (C) Eigenvariate of the middle frontal gyrus cluster in the flexibility versus stability condition versus individual minimum action associated with a transition between the rule attractors. (D) Eigenvariate of the thalamostriatal network in the flexibility versus stability condition versus individual minimum action associated with a transition between the rule attractors.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

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?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Cited by

See all 81

Metadata

Authors
Ueltzhöffer, K., Armbruster-Genç, D. J. N., & Fiebach, C. J.
Affiliations
Goethe University Frankfurt (DE)
Corresponding author
  • Ueltzhöffer, Kai
  • Fiebach, Christian J.
Venue
PLOS Computational Biology
Published
2015-06-12
Volume
11
Issue
6
Pages
e1004331
ISSN
  • 1553-734X (ISSN-L)
  • 1553-7358
OpenAlex
W634858973
Cited by
78
References
98
Citations by year
2026: 6 · 2025: 7 · 2024: 15 · 2023: 10 · 2022: 7 · 2021: 13 · 2020: 8 · 2019: 6 · 2018: 2 · 2017: 3 · 2016: 1
FWCI
1.20
Type
journal article
Language
en
Open access
Locations
  • PLoS Computational Biology (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (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)
  • Figshare (submittedVersion, cc-by)
  • Publication Server of Goethe University Frankfurt am Main (Goethe University Frankfurt) (submittedVersion, cc-by)
Funders
  • Deutsche Forschungsgemeinschaft (FI 848/3-1, DFG FI 848/3-1)
  • Bundesministerium für Bildung und Forschung (01GQ1003A)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural and Behavioral Psychology Studies
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural dynamics and brain function
  • Life Sciences › Neuroscience › Cellular and Molecular Neuroscience › Neurotransmitter Receptor Influence on Behavior
Keywords
cognitive flexibility, cognitive stability, task switching, distractor inhibition, working memory, drift-diffusion model, winner-take-all networks, attractor dynamics, frontoparietal network, intraparietal sulcus, inferior frontal junction, dopamine, individual differences, reaction time, decision-making
MeSH terms
Adult, Brain, Brain Mapping, Cognition, Female, Humans, Magnetic Resonance Imaging, Male, Memory, Short-Term, Models, Neurological, Reaction Time, Task Performance and Analysis, Computational Biology, Young Adult

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.

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.