Hansem Sohn
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Hansem Sohn has a total of 79 co-authors. Here are the top co-authors by number of shared publications:
- Mehrdad Jazayeri
Massachusetts Institute of Technology
8 shared publications - Jaeseung Jeong
Department of Brain and Cognitive Sciences, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, 34141, Republic of Korea
3 shared publications - Nicolas Meirhaeghe
Harvard-MIT Division of Health Sciences & Technology
3 shared publications - Andrew R. Sedler
Center for Machine Learning, Georgia Institute of Technology, Atlanta, GA, USA
2 shared publications Bradley S. Peterson
Department of Psychiatry College of Physicians and Surgeons Columbia University New York USA
2 shared publications- Chethan Pandarinath
Wallace H. Coulter Department of Biomedical Engineering, Emory University and Georgia Institute of Technology, Atlanta, USA
2 shared publications
Latest Publications
- Open Access
Physical sensorimotor contingencies facilitate learning in monkeys
Summary: Animal studies in neuroscience routinely use virtual tasks that deviate significantly from the animals’ natural behavioral repertoire. This choice is often necessary for conducting careful and controlled experimentation. However, this approach raises a fundamental question: do virtual tasks expose neural computations and mechanisms associated with naturalistic tasks? In this study, we explore how naturalistic settings influence learning timescale and generalization behavior in rhesus monkeys. By training animals in a maze-solving task and comparing their behavior across two environments, we observe significant differences. Specifically, we found that a realistic setup using physical objects enabled learning that was orders of magnitude faster than the learning timescale observed in a comparable computerized virtual environment. Furthermore, we noted limited transfer of performance between the physical and virtual settings. These findings underscore the need for understanding how task environments shape learning and generalization, with potential implications for the design of future neuroscience and AI experiments.
Open AccessPhysical sensorimotor contingencies facilitate learning in monkeys
Summary: Animal studies in neuroscience routinely use virtual tasks that deviate significantly from the animals’ natural behavioral repertoire. This choice is often necessary for conducting careful and controlled experimentation. However, this approach raises a fundamental question: do virtual tasks expose neural computations and mechanisms associated with naturalistic tasks? In this study, we explore how naturalistic settings influence learning timescale and generalization behavior in rhesus monkeys. By training animals in a maze-solving task and comparing their behavior across two environments, we observe significant differences. Specifically, we found that a realistic setup using physical objects enabled learning that was orders of magnitude faster than the learning timescale observed in a comparable computerized virtual environment. Furthermore, we noted limited transfer of performance between the physical and virtual settings. These findings underscore the need for understanding how task environments shape learning and generalization, with potential implications for the design of future neuroscience and AI experiments.
- Open Access
Hypothesis-Driven Identification of Neural Algorithms With Dynamical Structure-Preserving Manifolds
Neuroscience aims to uncover the algorithms by which the brain builds and manipulates rich internal representations of objects, agents, and places. Existing modeling frameworks have had limited success in identifying these algorithms. Here, we present Dynamical Structure-Preserving Manifolds (dSPMs), a multilevel framework that programs, without training, recurrent neural networks to enable hypothesis-driven identification of the brain’s algorithms that generate symbolic, structure-preserving representations of the world. We apply dSPM to identify an algorithm for physical scene prediction in the dorsomedial frontal cortex (DMFC) of macaques intercepting a ball on a Pong-like virtual gameboard. dSPM better explained DMFC than alternatives, made predictions confirmed in neural data, and revealed an algorithm of DMFC that we term ‘manifold-based simulation’: perceived initial conditions rapidly collapse neural activity onto a low-dimensional manifold whose geometry encodes a physics-based representation of the scene, supporting far-out predictions without traditional next-step simulation. dSPM supports effective exploration of the computational foundations of biological intelligence.
Open AccessHypothesis-Driven Identification of Neural Algorithms With Dynamical Structure-Preserving Manifolds
Neuroscience aims to uncover the algorithms by which the brain builds and manipulates rich internal representations of objects, agents, and places. Existing modeling frameworks have had limited success in identifying these algorithms. Here, we present Dynamical Structure-Preserving Manifolds (dSPMs), a multilevel framework that programs, without training, recurrent neural networks to enable hypothesis-driven identification of the brain’s algorithms that generate symbolic, structure-preserving representations of the world. We apply dSPM to identify an algorithm for physical scene prediction in the dorsomedial frontal cortex (DMFC) of macaques intercepting a ball on a Pong-like virtual gameboard. dSPM better explained DMFC than alternatives, made predictions confirmed in neural data, and revealed an algorithm of DMFC that we term ‘manifold-based simulation’: perceived initial conditions rapidly collapse neural activity onto a low-dimensional manifold whose geometry encodes a physics-based representation of the scene, supporting far-out predictions without traditional next-step simulation. dSPM supports effective exploration of the computational foundations of biological intelligence.
- Open Access
- 5 October 2026
- Published by Springer Nature in Communications Biology
Tuned inhibitory control of neuronal firing thresholds explains predictive sensorimotor behavior
- 5 October 2026
- Published by Springer Nature in Communications Biology
Open AccessTuned inhibitory control of neuronal firing thresholds explains predictive sensorimotor behavior
- Open Access
A number simplex in the human medial temporal lobe
Humans handle numbers nimbly, suggesting a richer neural manifold structure than the prevalent mental number line model. In populations of medial temporal lobe (MTL) neurons in humans performing two simple tasks (dot counting and arithmetic), we find robust neural coding of numerosity that results in high dimensional, simplex-shaped manifolds. This shape affords more flexibility than a linear manifold due to its high shattering dimensionality and expressibility. Dot arrays and Arabic numerals evoked distinct simplicial population codes, yet they were linked by a linearly transferable latent structure within the same task. We find similar simplicial geometry of number representations in large language models (LLMs). Moreover, subjects internally computed arithmetic results were decodable during the calculation period, with decoding accuracy correlating with individual mathematical capacity. Finally, linear transformations of simplicial operand representations modeled the brains conversion of operands into decodable results, suggesting that the brains arithmetic procedures have some resemblance to the attention architecture of LLMs. Together, these findings establish a high dimensional representational foundation for numerical cognition in the brain.
Open AccessA number simplex in the human medial temporal lobe
Humans handle numbers nimbly, suggesting a richer neural manifold structure than the prevalent mental number line model. In populations of medial temporal lobe (MTL) neurons in humans performing two simple tasks (dot counting and arithmetic), we find robust neural coding of numerosity that results in high dimensional, simplex-shaped manifolds. This shape affords more flexibility than a linear manifold due to its high shattering dimensionality and expressibility. Dot arrays and Arabic numerals evoked distinct simplicial population codes, yet they were linked by a linearly transferable latent structure within the same task. We find similar simplicial geometry of number representations in large language models (LLMs). Moreover, subjects internally computed arithmetic results were decodable during the calculation period, with decoding accuracy correlating with individual mathematical capacity. Finally, linear transformations of simplicial operand representations modeled the brains conversion of operands into decodable results, suggesting that the brains arithmetic procedures have some resemblance to the attention architecture of LLMs. Together, these findings establish a high dimensional representational foundation for numerical cognition in the brain.
- Open Access
Concept Generalization in Humans and Large Language Models: Insights from the Number Game
We compare human and large language model (LLM) generalization in the number game, a concept inference task. Using a Bayesian model as an analytical framework, we examined the inductive biases and inference strategies of humans and LLMs. The Bayesian model captured human behavior better than LLMs in that humans flexibly infer rule-based and similarity-based concepts, whereas LLMs rely more on mathematical rules. Humans also demonstrated a few-shot generalization, even from a single example, while LLMs required more samples to generalize. These contrasts highlight the fundamental differences in how humans and LLMs infer and generalize mathematical concepts.
Open AccessConcept Generalization in Humans and Large Language Models: Insights from the Number Game
We compare human and large language model (LLM) generalization in the number game, a concept inference task. Using a Bayesian model as an analytical framework, we examined the inductive biases and inference strategies of humans and LLMs. The Bayesian model captured human behavior better than LLMs in that humans flexibly infer rule-based and similarity-based concepts, whereas LLMs rely more on mathematical rules. Humans also demonstrated a few-shot generalization, even from a single example, while LLMs required more samples to generalize. These contrasts highlight the fundamental differences in how humans and LLMs infer and generalize mathematical concepts.
- Open Access
- 23 April 2025
- Published by Elsevier in Progress in Neurobiology
- Vol. 250, 102760
Whole-brain effective connectivity of the sensorimotor system using 7 T fMRI with electrical microstimulation in non-human primates
- 23 April 2025
- Published by Elsevier in Progress in Neurobiology
- Vol. 250, 102760
Open AccessWhole-brain effective connectivity of the sensorimotor system using 7 T fMRI with electrical microstimulation in non-human primates
- Open Access
- 2 January 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-16
Dynamic tracking of objects in the macaque dorsomedial frontal cortex
A central tenet of cognitive neuroscience is that humans build an internal model of the external world and use mental simulation of the model to perform physical inferences. Decades of human experiments have shown that behaviors in many physical reasoning tasks are consistent with predictions from the mental simulation theory. However, evidence for the defining feature of mental simulation – that neural population dynamics reflect simulations of physical states in the environment – is limited. We test the mental simulation hypothesis by combining a naturalistic ball-interception task, large-scale electrophysiology in non-human primates, and recurrent neural network modeling. We find that neurons in the monkeys’ dorsomedial frontal cortex (DMFC) represent task-relevant information about the ball position in a multiplexed fashion. At a population level, the activity pattern in DMFC comprises a low-dimensional neural embedding that tracks the ball both when it is visible and invisible, serving as a neural substrate for mental simulation. A systematic comparison of different classes of task-optimized RNN models with the DMFC data provides further evidence supporting the mental simulation hypothesis. Our findings provide evidence that neural dynamics in the frontal cortex are consistent with internal simulation of external states in the environment.
- 2 January 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-16
Open AccessDynamic tracking of objects in the macaque dorsomedial frontal cortex
A central tenet of cognitive neuroscience is that humans build an internal model of the external world and use mental simulation of the model to perform physical inferences. Decades of human experiments have shown that behaviors in many physical reasoning tasks are consistent with predictions from the mental simulation theory. However, evidence for the defining feature of mental simulation – that neural population dynamics reflect simulations of physical states in the environment – is limited. We test the mental simulation hypothesis by combining a naturalistic ball-interception task, large-scale electrophysiology in non-human primates, and recurrent neural network modeling. We find that neurons in the monkeys’ dorsomedial frontal cortex (DMFC) represent task-relevant information about the ball position in a multiplexed fashion. At a population level, the activity pattern in DMFC comprises a low-dimensional neural embedding that tracks the ball both when it is visible and invisible, serving as a neural substrate for mental simulation. A systematic comparison of different classes of task-optimized RNN models with the DMFC data provides further evidence supporting the mental simulation hypothesis. Our findings provide evidence that neural dynamics in the frontal cortex are consistent with internal simulation of external states in the environment.
- 28 November 2022
- Published by Springer Nature in Nature Methods
- Vol. 19 (12) , 1572-1577
A large-scale neural network training framework for generalized estimation of single-trial population dynamics
Achieving state-of-the-art performance with deep neural population dynamics models requires extensive hyperparameter tuning for each dataset. AutoLFADS is a model-tuning framework that automatically produces high-performing autoencoding models on data from a variety of brain areas and tasks, without behavioral or task information. We demonstrate its broad applicability on several rhesus macaque datasets: from motor cortex during free-paced reaching, somatosensory cortex during reaching with perturbations, and dorsomedial frontal cortex during a cognitive timing task.
- 28 November 2022
- Published by Springer Nature in Nature Methods
- Vol. 19 (12) , 1572-1577
A large-scale neural network training framework for generalized estimation of single-trial population dynamics
Achieving state-of-the-art performance with deep neural population dynamics models requires extensive hyperparameter tuning for each dataset. AutoLFADS is a model-tuning framework that automatically produces high-performing autoencoding models on data from a variety of brain areas and tasks, without behavioral or task information. We demonstrate its broad applicability on several rhesus macaque datasets: from motor cortex during free-paced reaching, somatosensory cortex during reaching with perturbations, and dorsomedial frontal cortex during a cognitive timing task.
- Open Access
Dynamic tracking of objects in the macaque dorsomedial frontal cortex
Cognitive neuroscience posits that humans perform physical inferences using mental simulations. Here, we test this hypothesis by analyzing neural activity in the monkeys’ frontal cortex in a ball interception task. We find a low-dimensional neural embedding of the ball position that tracks the ball both when it is visible and invisible. This embedding may serve as a neural substrate for mental simulation.
Open AccessDynamic tracking of objects in the macaque dorsomedial frontal cortex
Cognitive neuroscience posits that humans perform physical inferences using mental simulations. Here, we test this hypothesis by analyzing neural activity in the monkeys’ frontal cortex in a ball interception task. We find a low-dimensional neural embedding of the ball position that tracks the ball both when it is visible and invisible. This embedding may serve as a neural substrate for mental simulation.
- Open Access
Parametric control of flexible timing through low-dimensional neural manifolds
Biological brains possess an unparalleled ability to adapt behavioral responses to changing stimuli and environments. How neural processes enable this capacity is a fundamental open question. Previous works have identified two candidate mechanisms: a low-dimensional organization of neural activity and a modulation by contextual inputs. We hypothesized that combining the two might facilitate generalization and adaptation in complex tasks. We tested this hypothesis in the framework of flexible timing tasks where dynamics play a key role. Examining trained recurrent neural networks, we found that confining the dynamics to a low-dimensional subspace allowed tonic inputs to parametrically control the overall input-output transform, enabling generalization to novel inputs and adaptation to changing conditions. Reverse-engineering and theoretical analyses demonstrated that this parametric control relies on a mechanism where tonic inputs modulate the dynamics along non-linear manifolds in activity space while preserving their geometry. Comparisons with data from behaving monkeys confirmed the behavioral and neural signatures of this mechanism.
Open AccessParametric control of flexible timing through low-dimensional neural manifolds
Biological brains possess an unparalleled ability to adapt behavioral responses to changing stimuli and environments. How neural processes enable this capacity is a fundamental open question. Previous works have identified two candidate mechanisms: a low-dimensional organization of neural activity and a modulation by contextual inputs. We hypothesized that combining the two might facilitate generalization and adaptation in complex tasks. We tested this hypothesis in the framework of flexible timing tasks where dynamics play a key role. Examining trained recurrent neural networks, we found that confining the dynamics to a low-dimensional subspace allowed tonic inputs to parametrically control the overall input-output transform, enabling generalization to novel inputs and adaptation to changing conditions. Reverse-engineering and theoretical analyses demonstrated that this parametric control relies on a mechanism where tonic inputs modulate the dynamics along non-linear manifolds in activity space while preserving their geometry. Comparisons with data from behaving monkeys confirmed the behavioral and neural signatures of this mechanism.