Richard Granger
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Richard Granger has a total of 62 co-authors. Here are the top co-authors by number of shared publications:
- Lilianne Rivka Mujica-Parodi
Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Harvard Medical School, Boston, MA, United States
4 shared publications - Anand Pathak
Dept. of Psychological and Brain Sciences, Dartmouth College, Hanover, NH, USA
3 shared publications Elijah F. W. Bowen
Brain Engineering Laboratory, Department of Psychological and Brain Sciences, Dartmouth, Hanover, NH, USA
3 shared publicationsHaris Organtzidis
Dept. of Biomedical Engineering, Laufer Center for Physical and Quantitative Biology, State University of New York at Stony Brook, Stony Brook, NY, USA
3 shared publications- Helmut H. Strey
State University of New York at Stony Brook
3 shared publications - Antonio Rodriguez
Dartmouth College
2 shared publications
Latest Publications
- Open Access
- 29 October 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 18 (1) , 1-28
Three Tiers of Computation in Transformers and in Brain Architectures
Enabling AI systems to exhibit both fluent natural language and reliable logical reasoning is of great interest, yet the underlying machinery necessary for these capabilities remains poorly understood. Increases in abilities arise from many factors (scaling, architecture, training), yet the core required computational abilities for any particular task still elude understanding. We show that it is the transition between computational tiers in the automata hierarchy from the Theory of Computation, rather than scaling or particular algorithms, that determines a system’s capabilities. We provide a rigorous theoretical treatment and empirical evaluations of both language models and human subjects, culminating in a theoretically grounded pragmatic benchmarking method that can be used to ascertain the computational tier of an artificial or biological system. Specifically, we identify the tiers for the transition from prelinguistic to fluent natural language, and the tiers for the subsequent transition from articulate language use to reliable logical reasoning. The resulting principled analyses provide explanatory accounts of the abilities and shortfalls of a range of language models, and suggest actionable insights into the expansion of their logic and reasoning abilities.
- 29 October 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 18 (1) , 1-28
Open AccessThree Tiers of Computation in Transformers and in Brain Architectures
Enabling AI systems to exhibit both fluent natural language and reliable logical reasoning is of great interest, yet the underlying machinery necessary for these capabilities remains poorly understood. Increases in abilities arise from many factors (scaling, architecture, training), yet the core required computational abilities for any particular task still elude understanding. We show that it is the transition between computational tiers in the automata hierarchy from the Theory of Computation, rather than scaling or particular algorithms, that determines a system’s capabilities. We provide a rigorous theoretical treatment and empirical evaluations of both language models and human subjects, culminating in a theoretically grounded pragmatic benchmarking method that can be used to ascertain the computational tier of an artificial or biological system. Specifically, we identify the tiers for the transition from prelinguistic to fluent natural language, and the tiers for the subsequent transition from articulate language use to reliable logical reasoning. The resulting principled analyses provide explanatory accounts of the abilities and shortfalls of a range of language models, and suggest actionable insights into the expansion of their logic and reasoning abilities.
- Open Access
- 8 October 2026
- Published by American Physical Society (APS) in PRX Intelligence
- Vol. 1 (2) , 023002
Scientific Machine Learning of Chaotic Systems Learns Reduced-Order Equations for Neural Populations
Extracting interpretable mathematical models from complex dynamical systems is difficult, especially for chaotic dynamics observed with noisy experimental data. We present PEM-UDE, a method that combines prediction-error methodology with universal differential equations to discover governing equations from limited, noise-corrupted observations. Prediction-error feedback smooths the chaotic optimization problem; for noise-free data generated within the model class, it preserves the data-consistent zero-loss set, whereas noise and model misspecification introduce a gain-dependent stability-bias trade-off. Preservation of the zero-loss set is not a guarantee of unique structural identifiability. We test the method on two benchmark chaotic systems—the Rössler attractor and a real electrical circuit—and recover the correct functional forms even when one observed dimension contains noise of 5 times the signal magnitude. The method also accepts prior knowledge of the system as an initial functional form, which we use to learn neural circuit equations that account for sparse connectivity, a feature missing from conventional neural mass models. Applied to a population of Izhikevich neurons, PEM-UDE yields a multiscale neural mass model that ties single-neuron parameters to macroscopic network dynamics and predicts a relationship between connection density, dominant oscillation frequency, and synchrony. We test these predictions against three intracranial recording datasets from rat and human cortices. For the neuroscience application, the learned equations are a reduced-order closure for a specified simulated Izhikevich network family; the experimental recordings provide an indirect consistency check of predicted frequency and synchrony trends, not a direct fit of the equations to recordings.
- 8 October 2026
- Published by American Physical Society (APS) in PRX Intelligence
- Vol. 1 (2) , 023002
Open AccessScientific Machine Learning of Chaotic Systems Learns Reduced-Order Equations for Neural Populations
Extracting interpretable mathematical models from complex dynamical systems is difficult, especially for chaotic dynamics observed with noisy experimental data. We present PEM-UDE, a method that combines prediction-error methodology with universal differential equations to discover governing equations from limited, noise-corrupted observations. Prediction-error feedback smooths the chaotic optimization problem; for noise-free data generated within the model class, it preserves the data-consistent zero-loss set, whereas noise and model misspecification introduce a gain-dependent stability-bias trade-off. Preservation of the zero-loss set is not a guarantee of unique structural identifiability. We test the method on two benchmark chaotic systems—the Rössler attractor and a real electrical circuit—and recover the correct functional forms even when one observed dimension contains noise of 5 times the signal magnitude. The method also accepts prior knowledge of the system as an initial functional form, which we use to learn neural circuit equations that account for sparse connectivity, a feature missing from conventional neural mass models. Applied to a population of Izhikevich neurons, PEM-UDE yields a multiscale neural mass model that ties single-neuron parameters to macroscopic network dynamics and predicts a relationship between connection density, dominant oscillation frequency, and synchrony. We test these predictions against three intracranial recording datasets from rat and human cortices. For the neuroscience application, the learned equations are a reduced-order closure for a specified simulated Izhikevich network family; the experimental recordings provide an indirect consistency check of predicted frequency and synchrony trends, not a direct fit of the equations to recordings.
- Open Access
- 31 August 2026
- Published by MDPI AG in Brain Sciences
- Vol. 16 (9) , 930
Clustering and Principal Component Analysis as Distinct Computational Regimes of a Biologically Motivated Neural Circuit
Background: Characteristic circuitry in superficial layers of neocortex has been repeatedly shown to carry out lateral inhibition: excitatory–inhibitory interactions within a specific anatomical design. Methods: We provide simulation and formal analyses of the emergent operations of this circuitry. Results: Derived directly from anatomical circuit layout and physiological activation patterns, we show that this circuit carries out two distinct effective procedures on its inputs: categorization, and component analysis; moreover, we show that each procedure’s emergence is dependent on a single biological parameter: the relative strength of local feedback inhibitory cells. We characterize the detailed nature of both the biological activity and the emergent statistical operations and evaluate them in the context of extensive related literature in statistics, machine learning, and computational neuroscience. Conclusions: Very notably, the two emergent operations (clustering and component analysis) have not previously been shown to contain deep mathematical connections to each other, let alone to each be derivable from a single overarching algorithmic precursor that has clustering and component analysis as two special cases. The identification of that deep formal mathematical connection, and its arrival directly from a detailed biological circuit, represents a rare instance of novel mathematical relations arising from biological analyses.
- 31 August 2026
- Published by MDPI AG in Brain Sciences
- Vol. 16 (9) , 930
Open AccessClustering and Principal Component Analysis as Distinct Computational Regimes of a Biologically Motivated Neural Circuit
Background: Characteristic circuitry in superficial layers of neocortex has been repeatedly shown to carry out lateral inhibition: excitatory–inhibitory interactions within a specific anatomical design. Methods: We provide simulation and formal analyses of the emergent operations of this circuitry. Results: Derived directly from anatomical circuit layout and physiological activation patterns, we show that this circuit carries out two distinct effective procedures on its inputs: categorization, and component analysis; moreover, we show that each procedure’s emergence is dependent on a single biological parameter: the relative strength of local feedback inhibitory cells. We characterize the detailed nature of both the biological activity and the emergent statistical operations and evaluate them in the context of extensive related literature in statistics, machine learning, and computational neuroscience. Conclusions: Very notably, the two emergent operations (clustering and component analysis) have not previously been shown to contain deep mathematical connections to each other, let alone to each be derivable from a single overarching algorithmic precursor that has clustering and component analysis as two special cases. The identification of that deep formal mathematical connection, and its arrival directly from a detailed biological circuit, represents a rare instance of novel mathematical relations arising from biological analyses.
- Open Access
Mechanistic simulation identifies predictive dose-dependent biomarkers of propofol anesthesia
Understanding how receptor-level pharmacological modulation reorganizes large-scale brain circuits remains a central challenge in neuropharmacology. We introduce a multiscale mechanistic model with explicit core–matrix thalamocortical architecture, driven solely by GABA-A modulation without parameter fitting to any anesthesia data, to examine how propofol reorganizes brainwide activity from individual receptors to systems-level circuits. The model exhibits anesthetic effects spanning individual synaptic conductances to widespread changes in spiking, field potentials, and coherence. Without training on any task-specific data, our simulation of sensory processing in a standard auditory oddball paradigm matches independent macaque datasets. The same simulation, unmodified, also reproduces changes to functional connectivity in anesthetized humans, exhibiting selective attenuation of matrix thalamocortical loops relative to core loops. Most importantly, the simulation identified a dose-dependent biomarker of propofol concentration — elevated residual inter-stimulus cortical activity — that was subsequently confirmed in empirical macaque data where it had previously gone unnoticed. This simulation-first discovery, arising from mechanistic circuit dynamics rather than statistical comparison of clinical populations, illustrates a generative framework for translating receptor-level modulation into circuit-scale biomarkers with potential applications across predictive neuropharmacology.
Open AccessMechanistic simulation identifies predictive dose-dependent biomarkers of propofol anesthesia
Understanding how receptor-level pharmacological modulation reorganizes large-scale brain circuits remains a central challenge in neuropharmacology. We introduce a multiscale mechanistic model with explicit core–matrix thalamocortical architecture, driven solely by GABA-A modulation without parameter fitting to any anesthesia data, to examine how propofol reorganizes brainwide activity from individual receptors to systems-level circuits. The model exhibits anesthetic effects spanning individual synaptic conductances to widespread changes in spiking, field potentials, and coherence. Without training on any task-specific data, our simulation of sensory processing in a standard auditory oddball paradigm matches independent macaque datasets. The same simulation, unmodified, also reproduces changes to functional connectivity in anesthetized humans, exhibiting selective attenuation of matrix thalamocortical loops relative to core loops. Most importantly, the simulation identified a dose-dependent biomarker of propofol concentration — elevated residual inter-stimulus cortical activity — that was subsequently confirmed in empirical macaque data where it had previously gone unnoticed. This simulation-first discovery, arising from mechanistic circuit dynamics rather than statistical comparison of clinical populations, illustrates a generative framework for translating receptor-level modulation into circuit-scale biomarkers with potential applications across predictive neuropharmacology.
- Open Access
A logical re-conception of neural networks: Hamiltonian bitwise part-whole architecture
We introduce a simple initial working system in which relations (such as part-whole) are directly represented via an architecture with operating and learning rules fundamentally distinct from standard artificial neural network methods. Arbitrary data are straightforwardly encoded as graphs whose edges correspond to codes from a small fixed primitive set of elemental pairwise relations, such that simple relational encoding is not an add-on, but occurs intrinsically within the most basic components of the system. A novel graph-Hamiltonian operator calculates energies among these encodings, with ground states denoting simultaneous satisfaction of all relation constraints among graph vertices. The method solely uses radically low-precision arithmetic; computational cost is correspondingly low, and scales linearly with the number of edges in the data. The resulting unconventional architecture can process standard ANN examples, but also produces representations that exhibit characteristics of symbolic computation. Specifically, the method identifies simple logical relational structures in these data (part-of; next-to), building hierarchical representations that enable abductive inferential steps generating relational position-based encodings, rather than solely statistical representations. Notably, an equivalent set of ANN operations are derived, identifying a special case of embedded vector encodings that may constitute a useful approach to current work in higher-level semantic representation. The very simple current state of the implemented system invites additional tools and improvements.
Open AccessA logical re-conception of neural networks: Hamiltonian bitwise part-whole architecture
We introduce a simple initial working system in which relations (such as part-whole) are directly represented via an architecture with operating and learning rules fundamentally distinct from standard artificial neural network methods. Arbitrary data are straightforwardly encoded as graphs whose edges correspond to codes from a small fixed primitive set of elemental pairwise relations, such that simple relational encoding is not an add-on, but occurs intrinsically within the most basic components of the system. A novel graph-Hamiltonian operator calculates energies among these encodings, with ground states denoting simultaneous satisfaction of all relation constraints among graph vertices. The method solely uses radically low-precision arithmetic; computational cost is correspondingly low, and scales linearly with the number of edges in the data. The resulting unconventional architecture can process standard ANN examples, but also produces representations that exhibit characteristics of symbolic computation. Specifically, the method identifies simple logical relational structures in these data (part-of; next-to), building hierarchical representations that enable abductive inferential steps generating relational position-based encodings, rather than solely statistical representations. Notably, an equivalent set of ANN operations are derived, identifying a special case of embedded vector encodings that may constitute a useful approach to current work in higher-level semantic representation. The very simple current state of the implemented system invites additional tools and improvements.
- Open Access
- 29 December 2025
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Biomimetic model of corticostriatal micro-assemblies discovers a neural code
Although computational models have deepened our understanding of neuroscience, it is still highly challenging to link actual low-level physiological activity (spiking, field potentials) and biochemistry (transmitters and receptors) directly with high-level cognitive abilities (decision-making, working memory) and associated disorders. Here, we introduce a mechanistically accurate multi-scale model directly generating simulated physiology from which extended neural and cognitive phenomena emerge. The model produces spiking, fields, phase synchronies, and synaptic change, directly generating working memory, decisions, and categorization. These were then validated on extensive experimental macaque data from which the model received no prior training of any kind. Moreover, the simulation uncovered a previously unknown neural code (“incongruent neurons”) that specifically predicts upcoming erroneous behaviors, also subsequently confirmed in empirical data. The biomimetic model thus directly and predictively links decision and reinforcement signals, of computational interest, with spiking and field codes, of neurobiological importance.
- 29 December 2025
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Open AccessBiomimetic model of corticostriatal micro-assemblies discovers a neural code
Although computational models have deepened our understanding of neuroscience, it is still highly challenging to link actual low-level physiological activity (spiking, field potentials) and biochemistry (transmitters and receptors) directly with high-level cognitive abilities (decision-making, working memory) and associated disorders. Here, we introduce a mechanistically accurate multi-scale model directly generating simulated physiology from which extended neural and cognitive phenomena emerge. The model produces spiking, fields, phase synchronies, and synaptic change, directly generating working memory, decisions, and categorization. These were then validated on extensive experimental macaque data from which the model received no prior training of any kind. Moreover, the simulation uncovered a previously unknown neural code (“incongruent neurons”) that specifically predicts upcoming erroneous behaviors, also subsequently confirmed in empirical data. The biomimetic model thus directly and predictively links decision and reinforcement signals, of computational interest, with spiking and field codes, of neurobiological importance.
- Open Access
A Proposal for Extending the Common Model of Cognition to Emotion
Cognition and emotion must be partnered in any complete model of a humanlike mind. This article proposes an extension to the Common Model of Cognition -- a developing consensus concerning what is required in such a mind -- for emotion that includes a linked pair of modules for emotion and metacognitive assessment, plus pervasive connections between these two new modules and the Common Model's existing modules and links.
Open AccessA Proposal for Extending the Common Model of Cognition to Emotion
Cognition and emotion must be partnered in any complete model of a humanlike mind. This article proposes an extension to the Common Model of Cognition -- a developing consensus concerning what is required in such a mind -- for emotion that includes a linked pair of modules for emotion and metacognitive assessment, plus pervasive connections between these two new modules and the Common Model's existing modules and links.
- Open Access
Biomimetic model of corticostriatal micro-assemblies discovers new neural code
Although computational models have deepened our understanding of neuroscience, it is still highly challenging to link actual low-level physiological activity (spiking, field potentials) and biochemistry (transmitters and receptors) with high-level cognitive abilities (decision-making, working memory) nor with corresponding disorders. We introduce an anatomically-organized multi-scale model directly generating simulated physiology from which extended neural and cognitive phenomena emerge. The model produces spiking, fields, phase synchronies, and synaptic change, directly generating working memory, decisions, and categorization, all of which were then validated on extensive experimental macaque data from which the model received zero prior training of any kind. Moreover, the simulation uncovered a previously unknown neural code specifically predicting upcoming erroneous (“incongruous”) behaviors, also subsequently confirmed in empirical data. The biomimetic model thus directly and predictively links novel decision and reinforcement signals, of computational interest, with novel spiking and field codes, of potential behavioral and clinical relevance.
Open AccessBiomimetic model of corticostriatal micro-assemblies discovers new neural code
Although computational models have deepened our understanding of neuroscience, it is still highly challenging to link actual low-level physiological activity (spiking, field potentials) and biochemistry (transmitters and receptors) with high-level cognitive abilities (decision-making, working memory) nor with corresponding disorders. We introduce an anatomically-organized multi-scale model directly generating simulated physiology from which extended neural and cognitive phenomena emerge. The model produces spiking, fields, phase synchronies, and synaptic change, directly generating working memory, decisions, and categorization, all of which were then validated on extensive experimental macaque data from which the model received zero prior training of any kind. Moreover, the simulation uncovered a previously unknown neural code specifically predicting upcoming erroneous (“incongruous”) behaviors, also subsequently confirmed in empirical data. The biomimetic model thus directly and predictively links novel decision and reinforcement signals, of computational interest, with novel spiking and field codes, of potential behavioral and clinical relevance.
- Open Access
- 11 October 2022
- Published by Association for Research in Vision and Ophthalmology (ARVO) in Journal of Vision
- Vol. 22 (11) , 4
Visual stream connectivity predicts assessments of image quality
Despite extensive study of early vision, new and unexpected mechanisms continue to be identified. We introduce a novel formal treatment of the psychophysics of image similarity, derived directly from straightforward connectivity patterns in early visual pathways. The resulting differential geometry formulation is shown to provide accurate and explanatory accounts of human perceptual similarity judgments. The direct formal predictions are then shown to be further improved via simple regression on human behavioral reports, which in turn are used to construct more elaborate hypothesized neural connectivity patterns. It is shown that the predictive approaches introduced here outperform a standard successful published measure of perceived image fidelity; moreover, the approach provides clear explanatory principles of these similarity findings.
- 11 October 2022
- Published by Association for Research in Vision and Ophthalmology (ARVO) in Journal of Vision
- Vol. 22 (11) , 4
Open AccessVisual stream connectivity predicts assessments of image quality
Despite extensive study of early vision, new and unexpected mechanisms continue to be identified. We introduce a novel formal treatment of the psychophysics of image similarity, derived directly from straightforward connectivity patterns in early visual pathways. The resulting differential geometry formulation is shown to provide accurate and explanatory accounts of human perceptual similarity judgments. The direct formal predictions are then shown to be further improved via simple regression on human behavioral reports, which in turn are used to construct more elaborate hypothesized neural connectivity patterns. It is shown that the predictive approaches introduced here outperform a standard successful published measure of perceived image fidelity; moreover, the approach provides clear explanatory principles of these similarity findings.
- Open Access
- 31 August 2022
- Published by Frontiers Media SA in Frontiers in Communication
- Vol. 7
Phonemes in continuous speech are better recognized in context than in isolation
The contribution of context to phoneme perception is a subject of extensive study. In recent years, while the perception of phonemes in and out of context has become characterized as well-understood, new studies have emerged to challenge prevailing wisdom. Findings derived from rigorously controlled stimuli have failed to hold up when tested against continuous or more naturalistic speech, and vowels produced in isolation have been shown to possess different frequencies than vowels in spontaneous speech. In the present study, we examine the effect of context on vowel recognition, via stimuli taken directly from natural continuous speech in an audiobook. All tested vowel sounds, except /EH/, were better recognized with surrounding context than in isolation, affirming the resilience of findings from past studies.
- 31 August 2022
- Published by Frontiers Media SA in Frontiers in Communication
- Vol. 7
Open AccessPhonemes in continuous speech are better recognized in context than in isolation
The contribution of context to phoneme perception is a subject of extensive study. In recent years, while the perception of phonemes in and out of context has become characterized as well-understood, new studies have emerged to challenge prevailing wisdom. Findings derived from rigorously controlled stimuli have failed to hold up when tested against continuous or more naturalistic speech, and vowels produced in isolation have been shown to possess different frequencies than vowels in spontaneous speech. In the present study, we examine the effect of context on vowel recognition, via stimuli taken directly from natural continuous speech in an audiobook. All tested vowel sounds, except /EH/, were better recognized with surrounding context than in isolation, affirming the resilience of findings from past studies.