Sheng Li
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
Sheng Li has a total of 97 co-authors. Here are the top co-authors by number of shared publications:
- Zoe Kourtzi
Department of Psychology, University of Cambridge, Cambridge, UK
11 shared publications Ke Jia
Affiliated Mental Health Center and Hangzhou Seventh People’s Hospital, Zhejiang University School of Medicine, Hangzhou, People’s Republic of China
6 shared publicationsWen Wen
School of Psychological and Cognitive Sciences, Peking University
6 shared publications- Mengyuan Gong
Department of Psychology and Behavioral Sciences, Zhejiang University
5 shared publications Zhibang Huang
School of Psychological and Cognitive Sciences, Peking University, Beijing, China
5 shared publicationsHui Huang
School of Psychological and Cognitive Sciences, Peking University, Beijing, China
4 shared publications
Latest Publications
- Open Access
Behaviorally prioritized entity-relation structure captures human visual cortical representations of natural scenes
Understanding natural scenes requires identifying visible entities and representing how those entities are related. Recent studies have shown that artificial neural networks (ANNs), large language models (LLMs), and vision language models (VLMs) can predict visual cortical responses to natural images. However, the neural organization of relational scene meaning remains poorly understood, in part because these models typically encode scene content in global feature spaces that are difficult to decompose into separable entity and relation components. Here, we combined scene-graph annotations, behavioral measurements, and large-scale neural datasets to characterize structured relational representations during natural vision. We used RotatE, a knowledge-graph embedding model, to represent head-relation-tail triplets annotated for images from the 7T Natural Scenes Dataset. Triplet embeddings reliably captured cortical representational structure across the visual hierarchy. Behavioral judgments further revealed systematic differences in triplet accessibility associated with visual, relational, and graph properties. Prioritizing more behaviorally accessible triplets improved neural correspondence and explained unique variance beyond object co-occurrence, ANN image features, and LLM caption embeddings. Decomposing triplet representations into entity and relation components revealed partially dissociable cortical contributions, with lateral parietal cortex showing sensitivity to both. Triplet-based semantic information also remained spatially grounded: visual-field-specific triplet models preferentially predicted voxels with matching retinotopic preferences. Finally, cross-species comparison indicated that triplet-based semantic features were relatively more aligned with human high-level visual cortex than with macaque inferotemporal cortex. Together, these findings provide new insights into the representation of semantic relational information in the human visual cortex during natural scene perception.
Open AccessBehaviorally prioritized entity-relation structure captures human visual cortical representations of natural scenes
Understanding natural scenes requires identifying visible entities and representing how those entities are related. Recent studies have shown that artificial neural networks (ANNs), large language models (LLMs), and vision language models (VLMs) can predict visual cortical responses to natural images. However, the neural organization of relational scene meaning remains poorly understood, in part because these models typically encode scene content in global feature spaces that are difficult to decompose into separable entity and relation components. Here, we combined scene-graph annotations, behavioral measurements, and large-scale neural datasets to characterize structured relational representations during natural vision. We used RotatE, a knowledge-graph embedding model, to represent head-relation-tail triplets annotated for images from the 7T Natural Scenes Dataset. Triplet embeddings reliably captured cortical representational structure across the visual hierarchy. Behavioral judgments further revealed systematic differences in triplet accessibility associated with visual, relational, and graph properties. Prioritizing more behaviorally accessible triplets improved neural correspondence and explained unique variance beyond object co-occurrence, ANN image features, and LLM caption embeddings. Decomposing triplet representations into entity and relation components revealed partially dissociable cortical contributions, with lateral parietal cortex showing sensitivity to both. Triplet-based semantic information also remained spatially grounded: visual-field-specific triplet models preferentially predicted voxels with matching retinotopic preferences. Finally, cross-species comparison indicated that triplet-based semantic features were relatively more aligned with human high-level visual cortex than with macaque inferotemporal cortex. Together, these findings provide new insights into the representation of semantic relational information in the human visual cortex during natural scene perception.
- Open Access
- 10 September 2026
- Published by eLife Sciences Publications, Ltd in eLife preprints
- Vol. 15
Sequential alpha and theta dynamics resolve competition between rival visual stimuli
Attention plays a central role in selecting information for conscious perception. Because natural environments are structured, prior knowledge about relevant and irrelevant features can guide this selection. However, prior knowledge about distractors can enhance their neural representation, raising the question of how the brain prevents such enhancement from interfering with behavior. We addressed this question by combining a binocular rivalry paradigm with frequency-tagged stimuli to independently track target and distractor signals during sustained perceptual competition. We identified a sequential pattern of oscillatory dynamics underlying stable perception. Early parietal alpha activity (8-12 Hz) was associated with the segregation of competing inputs without altering sensory gain. This was followed by frontal theta activity (3-7 Hz), which was linked to the reactive suppression of distractors representations. Importantly, alpha-mediated gating effects are most pronounced under high perceptual uncertainty, where they facilitated the stabilization of weak target signals and promoted their access to conscious awareness. Together, these findings demonstrate that perceptual competition is resolved through a coordinated sequence of proactive gating and reactive suppression.
- 10 September 2026
- Published by eLife Sciences Publications, Ltd in eLife preprints
- Vol. 15
Open AccessSequential alpha and theta dynamics resolve competition between rival visual stimuli
Attention plays a central role in selecting information for conscious perception. Because natural environments are structured, prior knowledge about relevant and irrelevant features can guide this selection. However, prior knowledge about distractors can enhance their neural representation, raising the question of how the brain prevents such enhancement from interfering with behavior. We addressed this question by combining a binocular rivalry paradigm with frequency-tagged stimuli to independently track target and distractor signals during sustained perceptual competition. We identified a sequential pattern of oscillatory dynamics underlying stable perception. Early parietal alpha activity (8-12 Hz) was associated with the segregation of competing inputs without altering sensory gain. This was followed by frontal theta activity (3-7 Hz), which was linked to the reactive suppression of distractors representations. Importantly, alpha-mediated gating effects are most pronounced under high perceptual uncertainty, where they facilitated the stabilization of weak target signals and promoted their access to conscious awareness. Together, these findings demonstrate that perceptual competition is resolved through a coordinated sequence of proactive gating and reactive suppression.
- Open Access
- 25 March 2026
- Published by Public Library of Science (PLoS) in PLOS Biology
- Vol. 24 (3) , e3003541
Three-dimensional scene boundary representations for wall orientation and distance are represented distinctly in the human visual cortex
Human spatial navigation relies on the brain’s ability to visually represent the 3D layout of the environment. To understand how the brain encodes the layout information, it is crucial to identify the key features of environmental layout and how they are processed in the human brain. The vector coding principle, which highlight the role of boundary distance and orientation, provides a theoretical framework supported by physiological evidence from rodents. In this study, we developed a reconstruction approach to quantitatively estimate 3D layout information from natural indoor scene images. This approach enabled analyses of fMRI data from the large-scale Natural Scenes Dataset (NSD) using vector-based models of 3D layout. To validate the NSD-based results and examine task-related dynamics, we further conducted fMRI and MEG experiments with navigation-related and non-navigational tasks. Controlling for low-, mid-, and high-level visual and semantic features of natural indoor scenes, we found a spatiotemporal dissociation between boundary distance and orientation representations in the human brain. Relative distance was encoded in the early visual cortex during early processing in a task-invariant manner, whereas orientation was represented in scene-selective higher visual areas during later processing and was modulated by navigation-related tasks. Importantly, task modulation manifested as enhanced orientation coding in early visual cortex, potentially reflecting top-down feedback and short-term maintenance mechanisms. Together, these findings provide a novel perspective on how the human brain represents navigation-relevant information about the immediate surrounding environment, advancing our understanding of the neural mechanisms that link perception to action in spatial navigation.
- 25 March 2026
- Published by Public Library of Science (PLoS) in PLOS Biology
- Vol. 24 (3) , e3003541
Open AccessThree-dimensional scene boundary representations for wall orientation and distance are represented distinctly in the human visual cortex
Human spatial navigation relies on the brain’s ability to visually represent the 3D layout of the environment. To understand how the brain encodes the layout information, it is crucial to identify the key features of environmental layout and how they are processed in the human brain. The vector coding principle, which highlight the role of boundary distance and orientation, provides a theoretical framework supported by physiological evidence from rodents. In this study, we developed a reconstruction approach to quantitatively estimate 3D layout information from natural indoor scene images. This approach enabled analyses of fMRI data from the large-scale Natural Scenes Dataset (NSD) using vector-based models of 3D layout. To validate the NSD-based results and examine task-related dynamics, we further conducted fMRI and MEG experiments with navigation-related and non-navigational tasks. Controlling for low-, mid-, and high-level visual and semantic features of natural indoor scenes, we found a spatiotemporal dissociation between boundary distance and orientation representations in the human brain. Relative distance was encoded in the early visual cortex during early processing in a task-invariant manner, whereas orientation was represented in scene-selective higher visual areas during later processing and was modulated by navigation-related tasks. Importantly, task modulation manifested as enhanced orientation coding in early visual cortex, potentially reflecting top-down feedback and short-term maintenance mechanisms. Together, these findings provide a novel perspective on how the human brain represents navigation-relevant information about the immediate surrounding environment, advancing our understanding of the neural mechanisms that link perception to action in spatial navigation.
- Open Access
Vectorial boundary representation of 3D layout in human visual cortex
Human spatial navigation relies on the brain’s ability to visually represent the 3D layout of the environment. To understand how the brain encodes the layout information, it is crucial to identify the key features of environmental layout and how they are processed in the human brain. The vector coding principle, which highlight the role of boundary distance and orientation, provides a theoretical framework supported by physiological evidence from rodents. In this study, we developed a reconstruction approach to quantitatively estimate 3D layout information from natural indoor scene images. This approach enabled analyses of fMRI data from the large-scale Natural Scenes Dataset (NSD) using vector-based models of 3D layout. To validate the NSD-based results and examine task-related dynamics, we further conducted fMRI and MEG experiments with navigation-related and non-navigational tasks. Controlling for low-, mid-, and high-level visual and semantic features of natural indoor scenes, we found a spatiotemporal dissociation between boundary distance and orientation representations in the human brain. Relative distance was encoded in the early visual cortex during early processing in a task-invariant manner, whereas orientation was represented in scene-selective higher visual areas during later processing and was modulated by navigation-related tasks. Importantly, task modulation manifested as feedback-induced enhancement of orientation coding in the early visual cortex. Together, these findings provide a novel perspective on how the human brain represents navigation-relevant information about the immediate surrounding environment, advancing our understanding of the neural mechanisms that link perception to action in spatial navigation.
Open AccessVectorial boundary representation of 3D layout in human visual cortex
Human spatial navigation relies on the brain’s ability to visually represent the 3D layout of the environment. To understand how the brain encodes the layout information, it is crucial to identify the key features of environmental layout and how they are processed in the human brain. The vector coding principle, which highlight the role of boundary distance and orientation, provides a theoretical framework supported by physiological evidence from rodents. In this study, we developed a reconstruction approach to quantitatively estimate 3D layout information from natural indoor scene images. This approach enabled analyses of fMRI data from the large-scale Natural Scenes Dataset (NSD) using vector-based models of 3D layout. To validate the NSD-based results and examine task-related dynamics, we further conducted fMRI and MEG experiments with navigation-related and non-navigational tasks. Controlling for low-, mid-, and high-level visual and semantic features of natural indoor scenes, we found a spatiotemporal dissociation between boundary distance and orientation representations in the human brain. Relative distance was encoded in the early visual cortex during early processing in a task-invariant manner, whereas orientation was represented in scene-selective higher visual areas during later processing and was modulated by navigation-related tasks. Importantly, task modulation manifested as feedback-induced enhancement of orientation coding in the early visual cortex. Together, these findings provide a novel perspective on how the human brain represents navigation-relevant information about the immediate surrounding environment, advancing our understanding of the neural mechanisms that link perception to action in spatial navigation.
- Open Access
Alpha and theta dynamics successively resolve competition between rival visual stimuli
Attention plays a central role in selecting information for conscious perception. Because natural environments are structured, prior knowledge about relevant and irrelevant features can guide this selection. However, prior knowledge about distractors can enhance their neural representation, raising the question of how the brain prevents such enhancement from interfering with behavior. We addressed this question by combining a binocular rivalry paradigm with frequency-tagged stimuli to independently track target and distractor signals during sustained perceptual competition. We identified complementary alpha and theta dynamics that act successively to stable perception. Early parietal alpha activity (8-12 Hz) was associated with the segregation of competing inputs without altering sensory gain. This was followed by frontal theta activity (3-7 Hz), which was linked to the reactive suppression of distractors representations. Importantly, alpha-mediated gating effects were most pronounced under high perceptual uncertainty when sensory evidence were weak, where they facilitated the stabilization of weak target signals and promoted their access to conscious awareness. Together, these findings demonstrate that perceptual competition is resolved through a coordinated sequence of proactive gating and reactive suppression.
Open AccessAlpha and theta dynamics successively resolve competition between rival visual stimuli
Attention plays a central role in selecting information for conscious perception. Because natural environments are structured, prior knowledge about relevant and irrelevant features can guide this selection. However, prior knowledge about distractors can enhance their neural representation, raising the question of how the brain prevents such enhancement from interfering with behavior. We addressed this question by combining a binocular rivalry paradigm with frequency-tagged stimuli to independently track target and distractor signals during sustained perceptual competition. We identified complementary alpha and theta dynamics that act successively to stable perception. Early parietal alpha activity (8-12 Hz) was associated with the segregation of competing inputs without altering sensory gain. This was followed by frontal theta activity (3-7 Hz), which was linked to the reactive suppression of distractors representations. Importantly, alpha-mediated gating effects were most pronounced under high perceptual uncertainty when sensory evidence were weak, where they facilitated the stabilization of weak target signals and promoted their access to conscious awareness. Together, these findings demonstrate that perceptual competition is resolved through a coordinated sequence of proactive gating and reactive suppression.
- Open Access
- 29 April 2025
- Published by Springer Nature in Research Square
Recurrent code for working memory maintenance in the human frontal cortex
Despite accumulating evidence for the involvement of frontal cortex in working memory, it remains debated whether frontal cortex stores stimulus-specific mnemonic information or coordinates mnemonic representations in sensory areas. Neurophysiological non-human primate studies at microcircuit level largely support the former view, while human neuroimaging findings at macroscale level remain inconsistent. Here, we leverage ultra-high-field (7T) functional magnetic resonance imaging (fMRI) to interrogate the nature of working memory representations in human frontal cortex at submillimeter resolution. Our results reveal decodable working memory contents in the superficial layers of superior precentral sulcus (sPCS), indicating recurrent working memory mechanisms. These mnemonic contents are represented in a continuous format, resembling those observed during encoding. Their fidelity depends on the spatial selectivity of individual voxels and the strength of persistent activity, and closely tracks working memory performance. Our findings highlight the recurrent code for working memory in human frontal cortex, reconciling cross-species controversy into a shared mechanistic framework of working memory.
- 29 April 2025
- Published by Springer Nature in Research Square
Open AccessRecurrent code for working memory maintenance in the human frontal cortex
Despite accumulating evidence for the involvement of frontal cortex in working memory, it remains debated whether frontal cortex stores stimulus-specific mnemonic information or coordinates mnemonic representations in sensory areas. Neurophysiological non-human primate studies at microcircuit level largely support the former view, while human neuroimaging findings at macroscale level remain inconsistent. Here, we leverage ultra-high-field (7T) functional magnetic resonance imaging (fMRI) to interrogate the nature of working memory representations in human frontal cortex at submillimeter resolution. Our results reveal decodable working memory contents in the superficial layers of superior precentral sulcus (sPCS), indicating recurrent working memory mechanisms. These mnemonic contents are represented in a continuous format, resembling those observed during encoding. Their fidelity depends on the spatial selectivity of individual voxels and the strength of persistent activity, and closely tracks working memory performance. Our findings highlight the recurrent code for working memory in human frontal cortex, reconciling cross-species controversy into a shared mechanistic framework of working memory.
- Open Access
- 31 March 2025
- Published by Springer Nature in Nature Human Behaviour
- Vol. 9 (5) , 1023-1040
A neural geometry approach comprehensively explains apparently conflicting models of visual perceptual learning
Visual perceptual learning (VPL), defined as long-term improvement in a visual task, is considered a crucial tool for elucidating underlying visual and brain plasticity. Previous studies have proposed several neural models of VPL, including changes in neural tuning or in noise correlations. Here, to adjudicate different models, we propose that all neural changes at single units can be conceptualized as geometric transformations of population response manifolds in a high-dimensional neural space. Following this neural geometry approach, we identified neural manifold shrinkage due to reduced trial-by-trial population response variability, rather than tuning or correlation changes, as the primary mechanism of VPL. Furthermore, manifold shrinkage successfully explains VPL effects across artificial neural responses in deep neural networks, multivariate blood-oxygenation-level-dependent signals in humans and multiunit activities in monkeys. These converging results suggest that our neural geometry approach comprehensively explains a wide range of empirical results and reconciles previously conflicting models of VPL.
- 31 March 2025
- Published by Springer Nature in Nature Human Behaviour
- Vol. 9 (5) , 1023-1040
Open AccessA neural geometry approach comprehensively explains apparently conflicting models of visual perceptual learning
Visual perceptual learning (VPL), defined as long-term improvement in a visual task, is considered a crucial tool for elucidating underlying visual and brain plasticity. Previous studies have proposed several neural models of VPL, including changes in neural tuning or in noise correlations. Here, to adjudicate different models, we propose that all neural changes at single units can be conceptualized as geometric transformations of population response manifolds in a high-dimensional neural space. Following this neural geometry approach, we identified neural manifold shrinkage due to reduced trial-by-trial population response variability, rather than tuning or correlation changes, as the primary mechanism of VPL. Furthermore, manifold shrinkage successfully explains VPL effects across artificial neural responses in deep neural networks, multivariate blood-oxygenation-level-dependent signals in humans and multiunit activities in monkeys. These converging results suggest that our neural geometry approach comprehensively explains a wide range of empirical results and reconciles previously conflicting models of VPL.
- Open Access
- 2 May 2023
- Published by Association for Research in Vision and Ophthalmology (ARVO) in Journal of Vision
- Vol. 23 (5) , 1
Reactivation-induced memory integration prevents proactive interference in perceptual learning
We acquire perceptual skills through experience to adapt ourselves to the changing environment. Accomplishing an effective skill acquisition is a main purpose of perceptual learning research. Given the often observed learning effect specificity, multiple perceptual learnings with shared parameters could serve to improve the generalization of the learning effect. However, the interference between the overlapping memory traces of different learnings may impede this effort. Here, we trained human participants on an orientation discrimination task. We observed a proactive interference effect that the first training blocked the second training at its untrained location. This was a more pronounced effect than the well-known location specificity in perceptual learning. We introduced a short reactivation of the first training before the second training and successfully eliminated the proactive interference when the second training was inside the reconsolidation time window of the reactivated first training. Interestingly, we found that practicing an irrelevant task at the location of the second training immediately after the reactivation of the first training could also restore the effect of the second training but in a smaller magnitude, even if the second training was conducted outside of the reconsolidation window. We proposed a two-level mechanism of reactivation-induced memory integration to account for these results. The reactivation-based procedure could integrate either the previously trained and untrained locations or the two trainings at these locations, depending on the activated representations during the reconsolidation process. The findings provide us with new insight into the roles of long-term memory mechanisms in perceptual learning.
- 2 May 2023
- Published by Association for Research in Vision and Ophthalmology (ARVO) in Journal of Vision
- Vol. 23 (5) , 1
Open AccessReactivation-induced memory integration prevents proactive interference in perceptual learning
We acquire perceptual skills through experience to adapt ourselves to the changing environment. Accomplishing an effective skill acquisition is a main purpose of perceptual learning research. Given the often observed learning effect specificity, multiple perceptual learnings with shared parameters could serve to improve the generalization of the learning effect. However, the interference between the overlapping memory traces of different learnings may impede this effort. Here, we trained human participants on an orientation discrimination task. We observed a proactive interference effect that the first training blocked the second training at its untrained location. This was a more pronounced effect than the well-known location specificity in perceptual learning. We introduced a short reactivation of the first training before the second training and successfully eliminated the proactive interference when the second training was inside the reconsolidation time window of the reactivated first training. Interestingly, we found that practicing an irrelevant task at the location of the second training immediately after the reactivation of the first training could also restore the effect of the second training but in a smaller magnitude, even if the second training was conducted outside of the reconsolidation window. We proposed a two-level mechanism of reactivation-induced memory integration to account for these results. The reactivation-based procedure could integrate either the previously trained and untrained locations or the two trainings at these locations, depending on the activated representations during the reconsolidation process. The findings provide us with new insight into the roles of long-term memory mechanisms in perceptual learning.
- 29 September 2022
- Published by Springer Nature in Attention, Perception, & Psychophysics
- Vol. 85 (3) , 814-824
Learned low priority of attention after training to suppress color singleton distractor
Allocating attention to significant events, such as a salient object, is effortless. Our brain is effective on this type of processing because doing so is
- 29 September 2022
- Published by Springer Nature in Attention, Perception, & Psychophysics
- Vol. 85 (3) , 814-824
Learned low priority of attention after training to suppress color singleton distractor
Allocating attention to significant events, such as a salient object, is effortless. Our brain is effective on this type of processing because doing so is
- Open Access
Reactivation-induced memory integration prevents proactive interference in perceptual learning
We acquire perceptual skills through experience to adapt ourself to the changing environment. Accomplishing an effective skill acquisition is a main purpose of perceptual learning research. Given the often observed learning effect specificity, multiple perceptual learnings with shared parameters could serve to improve the generalization of the learning effect. However, the interference between the overlapping memory traces of different learnings may impede this effort. Here, we trained human participants on an orientation discrimination task. We observed a proactive interference effect that the first training blocked the second training at its untrained location. This was a more pronounced effect than the well-known location specificity in perceptual learning. We introduced a short reactivation of the first training before the second training and successfully eliminated the proactive interference when the second training was inside the reconsolidation time window of the reactivated first training. Interestingly, we found that practicing an irrelevant task at the location of the second training immediately after the reactivation of the first training could also restore the effect of the second training but in a smaller magnitude, even if the second training was conducted outside of the reconsolidation window. We proposed a two-level mechanism of reactivation-induced memory integration to account for these results. The reactivation-based procedure could integrate either the previously trained and untrained locations or the two trainings at these locations, depending on the activated representations during the reconsolidation process. The findings provide us with new insight into the roles of long-term memory mechanisms in perceptual learning.
Open AccessReactivation-induced memory integration prevents proactive interference in perceptual learning
We acquire perceptual skills through experience to adapt ourself to the changing environment. Accomplishing an effective skill acquisition is a main purpose of perceptual learning research. Given the often observed learning effect specificity, multiple perceptual learnings with shared parameters could serve to improve the generalization of the learning effect. However, the interference between the overlapping memory traces of different learnings may impede this effort. Here, we trained human participants on an orientation discrimination task. We observed a proactive interference effect that the first training blocked the second training at its untrained location. This was a more pronounced effect than the well-known location specificity in perceptual learning. We introduced a short reactivation of the first training before the second training and successfully eliminated the proactive interference when the second training was inside the reconsolidation time window of the reactivated first training. Interestingly, we found that practicing an irrelevant task at the location of the second training immediately after the reactivation of the first training could also restore the effect of the second training but in a smaller magnitude, even if the second training was conducted outside of the reconsolidation window. We proposed a two-level mechanism of reactivation-induced memory integration to account for these results. The reactivation-based procedure could integrate either the previously trained and untrained locations or the two trainings at these locations, depending on the activated representations during the reconsolidation process. The findings provide us with new insight into the roles of long-term memory mechanisms in perceptual learning.