Faculty Bibliography

Sensory processing is fundamentally shaped by stimulation history. For example, in visual cortex, neural responses are reduced for repeated or sustained stimuli (adaptation). These phenomena are well characterized and effectively modeled by divisive normalization. We asked whether these same computational principles govern somatosensory processing. We used fMRI (6 participants, all female) and intracranial electroencephalography (iEEG, 2 participants, both male) to measure responses to time-varying vibrotactile stimuli in human somatosensory cortex. Stimuli consisted of single- and paired-pulses with durations and interstimulus intervals ranging from 0.05 to 1.2 s. We extracted BOLD time courses to capture neural response amplitudes, and high-frequency iEEG broadband envelopes to capture fast neural dynamics. In both experiments, we observed pronounced sub-additive temporal summation. Responses to longer or repeated stimuli were consistently lower than predicted by linear integration. Computational modeling revealed that divisive normalization models outperformed linear models in cross-validated accuracy across both datasets. These results demonstrate that somatosensory temporal dynamics closely mirror those in the visual system. Our findings suggest that the nervous system employs similar computational principles across modalities to encode sensory information across time.Significance statement How the brain integrates sensory information over time is a fundamental question in neuroscience. While nonlinear temporal integration is well documented in visual cortex, it has not been extensively mapped in the human somatosensory system. By combining fMRI with intracranial EEG in humans, we demonstrate that somatosensory responses to tactile stimulation exhibit subadditive temporal summation. This nonlinearity is accurately captured by a divisive normalization model, matching observations in the visual system. Our results suggest that normalization is a canonical computation shared across different modalities to manage temporal dynamics, providing a unified framework for understanding how the brain encodes dynamic sensory stimuli.

Recent research has used large language models (LLMs) to study the neural basis of naturalistic language processing in the human brain. LLMs have rapidly grown in complexity, leading to improved language processing capabilities. Here, we utilized several families of transformer-based LLMs to investigate the relationship between model size and their ability to capture linguistic information in the human brain. Crucially, a subset of LLMs were trained on a fixed training set, enabling us to dissociate model size from architecture and training set size. We used electrocorticography (ECoG) to measure neural activity in epilepsy patients while they listened to a 30 min naturalistic audio story. We fit electrode-wise encoding models using contextual embeddings extracted from each hidden layer of the LLMs to predict word-level neural signals. In line with prior work, we found that larger LLMs better capture the structure of natural language and better predict neural activity. We also found a logarithmic relationship where the encoding performance peaks in relatively earlier layers as model size increases. We also observed variations in the best-performing layer across different brain regions, corresponding to an organized language processing hierarchy.

Humans are the only species with the ability to systematically combine words to convey an unbounded number of complex meanings. This process is guided by combinatorial processes whose underlying neural mechanisms remain obscured by inherent limitations of noninvasive brain measures and a near-total focus on comprehension paradigms. Here, we address these limitations with high-resolution neurosurgical recordings (electrocorticography) and a controlled sentence production experiment. We uncover distinct cortical networks encoding word-level and higher-order information. These networks exhibited a hybrid spatial organization: broadly distributed across traditional language areas but with focal concentrations of sensitivity to semantic and structural contrasts in canonical language regions. In contrast to previous comprehension-based findings, we find that these networks are largely nonoverlapping. Most notably, higher-order linguistic information showed an unexpected dissociation from local activity magnitude. This result establishes an operational dissociation between activity magnitude and information content, pointing toward a potentially distinct neural coding scheme for higher-order language, with important implications for the neurobiology of language.

BACKGROUND AND OBJECTIVES/OBJECTIVE:Direct electrocortical stimulation (DES) is the gold standard for mapping eloquent cortex, yet existing functional atlases are limited by sampling biases and density-based methods that obscure a region's true functional probability. Consequently, interpatient variability and the functional contributions of nontraditional language areas, such as the middle frontal gyrus, remain poorly characterized, particularly in epilepsy populations where functional reorganization is common. We therefore developed a probabilistic functional atlas of extraoperative DES and applied data-driven methods to characterize the functional organization of language, motor, and sensory cortex. METHODS:This was a retrospective observational study of patients undergoing intracranial monitoring for drug-resistant epilepsy (2008-2023). Electrical stimulation was delivered to intracranial electrodes during language tasks, and language, motor, and sensory findings were recorded. Positive and negative stimulation sites were analyzed in standard patient space and using the Human Connectome Project parcellation atlas. A multilevel statistical framework, including probability mapping, bootstrapped region-of-interest analyses, and kernel density estimation, defined structure-function relationships. Generalized linear mixed-effects models assessed the influence of clinical variables on language disruption. RESULTS:= 0.003) independently predicted a lower probability of language disruption in the temporal lobe. DISCUSSION/CONCLUSIONS:This extraoperative DES atlas provides a comprehensive benchmark for understanding functional cortical organization in epilepsy. We add evidence to the growing literature that language and motor systems are more distributed and variable than classically described. Substantial interpatient variability underscores the necessity of individualized mapping to guide safe neurosurgical planning. Limitations include the retrospective design and sampling bias inherent to electrode placement.

The ability to quickly learn and generalize is one of the brain's most impressive feats and recreating it remains a major challenge for modern artificial intelligence research. One of the most mysterious one-shot learning abilities displayed by humans is one-shot perceptual learning, whereby a single viewing experience drastically alters visual perception in a long-lasting manner. Where in the brain one-shot perceptual learning occurs and what mechanisms support it remain enigmatic. Combining psychophysics, 7 T fMRI, and intracranial recordings, we identify the high-level visual cortex as the most likely neural substrate wherein neural plasticity supports one-shot perceptual learning. We further develop a deep neural network model incorporating top-down feedback into a vision transformer, which recapitulates and predicts human behavior. The prior knowledge learnt by this model is highly similar to the neural code in the human high-level visual cortex. These results reveal the neurocomputational mechanisms underlying one-shot perceptual learning in humans.

Recent research demonstrates that large language models can predict neural activity recorded via electrocorticography during natural language processing. To predict word-by-word neural activity, most prior work evaluates encoding models within individual electrodes and participants, limiting generalizability. Here we analyze electrocorticography data from eight participants listening to the same 30-min podcast. Using a shared response model, we estimate a common information space across participants. This shared space substantially enhances large language model-based encoding performance and enables denoising of individual brain responses by projecting back into participant-specific electrode spaces-yielding a 37% average improvement in encoding accuracy (from r = 0.188 to r = 0.257). The greatest gains occur in brain areas specialized for language comprehension, particularly the superior temporal gyrus and inferior frontal gyrus. Our findings highlight that estimating a shared space allows us to construct encoding models that better generalize across individuals.

Large Language Models (LLMs) offer a framework for understanding language processing in the human brain. Unlike traditional models, LLMs represent words and context through layered numerical embeddings. Here, we demonstrate that LLMs' layer hierarchy aligns with the temporal dynamics of language comprehension in the brain. Using electrocorticography (ECoG) data from participants listening to a 30-minute narrative, we show that deeper LLM layers correspond to later brain activity, particularly in Broca's area and other language-related regions. We extract contextual embeddings from GPT-2 XL and Llama-2 and use linear models to predict neural responses across time. Our results reveal a strong correlation between model depth and the brain's temporal receptive window during comprehension. We also compare LLM-based predictions with symbolic approaches, highlighting the advantages of deep learning models in capturing brain dynamics. We release our aligned neural and linguistic dataset as a public benchmark to test competing theories of language processing.

Sound structures such as phonemes and words have highly variable durations. Therefore, there is a fundamental difference between integrating across absolute time (for example, 100 ms) versus sound structure (for example, phonemes). Auditory and cognitive models have traditionally cast neural integration in terms of time and structure, respectively, but the extent to which cortical computations reflect time or structure remains unknown. Here, to answer this question, we rescaled the duration of all speech structures using time stretching and compression and measured integration windows in the human auditory cortex using a new experimental and computational method applied to spatiotemporally precise intracranial recordings. We observed slightly longer integration windows for stretched speech, but this lengthening was very small (~5%) relative to the change in structure durations, even in non-primary regions strongly implicated in speech-specific processing. These findings demonstrate that time-yoked computations dominate throughout the human auditory cortex, placing important constraints on neurocomputational models of structure processing.