Biology subjects

Cox, C. R.

Publications and source records attributed to Cox, C. R..

7 recordsLinked to original sources

bioRxiv2026

Multidimensional semantic representations emerge from multi-frequency representational similarity learning

Current theories of semantic representation are agnostic about how fine-grained neurophysiological activity corresponds to representation of multidimensional semantic information. By applying an innovative multivariate technique (representational similarity learning; RSL), we adjudicated three hypotheses: (1) that multidimensional semantic structure is represented within a single frequency range (e.g., gamma/high gamma); (2) that each semantic dimension is independently represented within a different frequency range; and (3) that multidimensional semantic information is "transfrequency" (at least some information emerges only when multiple frequencies are considered together). RSL was applied to time-frequency power and phase data extracted from electrocorticography (ECoG) grid electrodes on the surface of human ventral anterior temporal lobe (vATL). We found significant decoding of graded, multidimensional semantic information from a wide range of frequencies (4 - 200 Hz), but not from individual frequency bands, providing clear evidence that multidimensional semantic information is coded in a transfrequency fashion.

neuroscience↗

bioRxiv2025

Adjudicating competing theories of semantic representation with 7T-fMRI and multivariate decoding

The anterior temporal lobes (ATLs) are known to support semantic cognition, but theories about their precise representational coding vary. We collected 7T-fMRI data with a novel acquisition sequence designed to improve signal quality in the ATLs, then employed a pioneering analytical approach (comparative multivariate decoding) to adjudicate between theories. Specifically, we applied multiple decoding methods, each making different assumptions about the content, nature or location of representations within the ATLs, then used the pattern of results across methods to adjudicate competing hypotheses. The results suggest that the ATLs represent domain-general semantic information via a multidimensional inchoate vector-space code that is anatomically clustered within and across individuals, and that posterior temporal and occipitotemporal regions utilise a similar domain-general, inchoate vector-space code. More generally, the comparative multivariate analytical framework utilised here has the potential to reveal how the brain represents, not just semantic knowledge, but any kind of information.

neuroscience↗

bioRxiv2025

All spectral frequencies of neural activity reveal semantic representation in the human anterior ventral temporal cortex

Intracranial electrophysiology offers a unique insight into the nature of information representation in the brain - it can be used to disentangle information encoded in local neuronal activity (gamma and high gamma frequencies) from information encoded via long-range interactions (lower frequencies). We used regularised logistic regression to decode animacy from time-frequency power and phase extracted from electrocorticography (ECoG) grid electrode data recorded on the surface of human vATL. Power in gamma (30 - 60 Hz) and high gamma (60 - 200 Hz) produced reliable decoding, indicating that semantic information is indeed expressed by local populations in vATL. However, power from a wide range of frequencies (4 - 200 Hz) produced significantly higher decoding accuracy and also exhibited the same rapidly-changing dynamic code previously observed when decoding voltage. These findings support the theory that semantic information is encoded by a local vATL "hub" that interacts with distributed cortical "spokes".

neuroscience↗

bioRxiv2022

Representational similarity learning reveals a graded multi-dimensional semantic space in the human anterior temporal cortex

Neurocognitive models of semantic memory have proposed that the ventral anterior temporal lobes (vATLs) encode a graded and multidimensional semantic space--yet neuroimaging studies seeking brain regions that encode semantic structure rarely identify these areas. In simulations we show that this discrepancy may arise from a crucial mismatch between theory and analysis approach. Utilizing an analysis recently formulated to investigate graded multidimensional representations, representational similarity learning (RSL), we decoded semantic structure from ECoG data collected from the vATL cortical surface while participants named line drawings of common items. The results reveal a graded, multidimensional semantic space encoded in neural activity across the vATL, which evolves over time and simultaneously expresses both broad and finer-grained semantic structure amongst animate and inanimate concepts. The work resolves the apparent discrepancy within the semantic cognition literature and, more importantly, suggests a new approach to discovering representational structure in neural data more generally. Conflict of interestThe Department of Epilepsy, Movement Disorders, and Physiology, Kyoto University Graduate School of Medicine conducts Industry-Academia Collaboration Courses, supported by Eisai Co., Ltd, Nihon Kohden Corporation, Otsuka Pharmaceutical Co., and UCB Japan Co., Ltd.

neuroscience↗

bioRxiv2021

Bipartite functional fractionation within the neural system for social cognition supports the psychological continuity of self vs. other

Research of social neuroscience establishes that regions in the brains default network (DN) and semantic network (SN) are engaged by socio-cognitive tasks. Research of the human connectome shows that DN and SN regions are both situated at the high-order end of cortical gradient but differ in their positions on this gradient. In the present study, we integrated these two bodies of research, used the psychological continuity of self vs. other as a test-case, and used fMRI to investigate whether these networks would encode social concepts differently. We found a robust dissociation between the DN and SN - while both networks contained sufficient information for decoding broad-stroke distinction of social categories, the DN carried more generalisable information for cross-classifying across social distance and emotive valence than did the SN. We also found that the overarching distinction of self vs. other was a principal divider of the representational space while social distance was an auxiliary factor (subdivision, nested within the principal dimension), and this representational landscape was more manifest in the DN than in the SN. Taken together, our findings demonstrate how insights from connectome research can benefit social neuroscience, and have implications for clarifying the two networks differential contributions to social cognition.

neuroscience↗

bioRxiv2018

Connecting natural and artificial neural networks in functional brain imaging using structured sparsity

Artificial neural network models have long proven useful for understanding healthy, disordered, and developing cognition, but this work has often proceeded with little connection to functional brain imaging. We consider how analysis of functional brain imaging data is best approached if the representational assumptions embodied by neural networks are valid. Using a simple model to generate synthetic data, we show that four contemporary methods each have critical and complementary blind-spots for detecting distributed signal. The pattern suggests a new approach based on structured sparsity that, in simulation, retains the strengths of each method while avoiding its weaknesses. When applied to functional magnetic resonance imaging data the new approach reveals extensive distributed signal missed by the other methods, suggesting radically different conclusions about how brains encode cognitive information in the well-studied domain of visual face perception.

neuroscience↗

bioRxiv2018

Improving Whole-Brain Neural Decoding of fMRI with Domain Adaptation

In neural decoding, there has been a growing interest in machine learning on whole-brain functional magnetic resonance imaging (fMRI). However, the size discrepancy between the feature space and the training set poses serious challenges. Simply increasing the number of training examples is infeasible and costly. In this paper, we proposed a domain adaptation framework for whole-brain fMRI (DawfMRI) to improve whole-brain neural decoding on target data leveraging pre-existing source data. DawfMRI consists of three steps: 1) feature extraction from whole-brain fMRI, 2) source and target feature adaptation, and 3) source and target classifier adaptation. We evaluated its eight possible variations, including two non-adaptation and six adaptation algorithms, using a collection of seven task-based fMRI datasets (129 unique subjects and 11 cognitive tasks in total) from the OpenNeuro project. The results demonstrated that appropriate source domain can help improve neural decoding accuracy for challenging classification tasks. The best-case improvement is 8.94% (from 78.64% to 87.58%). Moreover, we discovered a plausible relationship between psychological similarity and adaptation effectiveness. Finally, visualizing and interpreting voxel weights showed that the adaptation can provide additional insights into neural decoding.

neuroscience↗