Biology subjects

Turk-Browne, N. B.

Publications and source records attributed to Turk-Browne, N. B..

At least 19 recordsLinked to original sources

bioRxiv2026

Developmental tuning of functional manifold dimensionality across the human brain

Neural representations vary in their complexity across brain regions and tasks. How this variation emerges over human development remains poorly understood. We estimated intrinsic dimensionality in five naturalistic fMRI datasets (N = 781 unique participants, aged 3 months to 53 years) with T-PHATE -- a nonlinear manifold learning method robust to noisy, autocorrelated signals. In adults, brain regions relevant to a task had higher-dimensional activity than task-irrelevant regions across auditory, visual, and audiovisual stimuli. The modulation of representational complexity by tasks was absent in infants, emerged in early childhood, and strengthened logarithmically through adolescence. It reflected a selective collapse in dimensionality in task-irrelevant regions, relative to a resting-state baseline, rather than an expansion of dimensionality in task-relevant regions. Such compression followed a trajectory from global and nonselective in infants to local and precise by adulthood. These results identify selective compression as a developmental engine of functional specialization: rather than adding complexity where it is needed, the brain dynamically pares it away where it is not.

neuroscience↗

bioRxiv2026

Unsupervised nonmotor learning in the human cerebellum

A core motor function of the cerebellum is error-based learning: it uses prediction errors--mismatches between predicted and actual sensory feedback--to refine actions. One provocative hypothesis is that the cerebellum performs similar computations in cognitive domains. Here, we ask whether error-based learning computations in the cerebellum extend to a passive statistical learning task that requires neither action nor decision-making. Human participants viewed sequences of visual stimuli with varying probabilistic stimulus-stimulus transitions while undergoing fMRI. Subjects reported no explicit knowledge of the transitional probabilities after the scan. Nonetheless, putative cognitive regions of the cerebellum encoded prediction errors during statistical learning. Moreover, error signals in the cerebellum were encoded in a different manner than error signals in the anterior hippocampus, a well-known substrate of statistical learning: cerebellar activity covaried with trial-by-trial prediction errors that evolved slowly over time (echoing cerebellar computations in motor learning), whereas the hippocampus responded in a more fixed manner to the underlying transition structure. Computational modeling formalized this dissociation, with cerebellar activity explained by a delta-rule model that incrementally adjusted predictions based on recent experience, and hippocampal activity characterized as a form of Bayesian updating of a transition distribution held in memory. Our findings demonstrate that the cerebellum encodes prediction errors during passive nonmotor learning, and thus that it likely plays a domain-general role in error-based learning. SignificanceThe cerebellum is known to support motor learning by using prediction errors to refine behavior. Here, we test whether these computations extend to nonmotor learning contexts that require neither action nor decision-making. We report the discovery of prediction error signals in the cerebellum during statistical learning, an unsupervised process that extracts regularities from the sensory environment. These findings support a domain-general role for the cerebellum in error-based learning, offering a possible explanation for why cerebellar damage is implicated in such wide-ranging outcomes, from impaired motor coordination to complex disorders like autism.

neuroscience↗

bioRxiv2026

Enhanced processing of cartoons in infant visual cortex

Developing sensory systems may have heightened sensitivity to exaggerated features that emphasize diagnostic information, as shown by the benefits of infant-directed speech for language acquisition. Here, we examine this sensory exaggeration hypothesis in the visual domain by testing whether cartoons elicit stronger and more consistent neural representations than realistic movies in human infants. We collected fMRI data while 24 awake infants (4-15 months) watched the same 3-minute clip of the opening sequence from the original animated version of The Lion King (1994) and its shot-for-shot remake with photorealistic CGI (2019). A computer vision model confirmed that the movies differed in low-level features while depicting similar high-level content. Consistent with sensory exaggeration, the animated version yielded more reliable neural responses and better decoding of visual features than the CGI version throughout occipital cortex. This effect was not observed in adults nor in higher-order regions in infants and could not be explained by differential head motion or looking time. These results suggest that the developing visual system may be attuned to diagnostic features and that cartoons may (unwittingly) exploit this early neural preference.

neuroscience↗

bioRxiv2026

Dynamic construction of subjective time through statistical learning of event structure

The perception of time is elastic, often deviating from physical intervals depending on how experience is structured. Yet, what determines how subjective time is constructed remains debated. Here, we tested whether perceived ongoing time is actively constructed from learned event representations rather than a dedicated internal clock. Using a novel pause-adjustment task across three statistical learning experiments, we measured temporal distortions during continuous listening to structured versus unstructured syllable streams. The presence of event structure systematically warped time: pauses were perceived as longer between pseudowords and shorter within pseudowords. This bidirectional temporal warping emerged online and remained stable across pause durations. Enriching these events with semantic meaning eliminated boundary-related dilation while preserving within-event compression. Moreover, physiological tracking of event structure, indexed by pupil dynamics, dissociated from the magnitude of temporal warping. These findings show that subjective time is constructed from hierarchical event representations and depends not only on where events are segmented, but also on how they are represented.

neuroscience↗

bioRxiv2026

Multisensory coding of audiovisual movies in the human hippocampus

The hippocampus receives convergent input from multiple sensory systems, yet in humans it has been studied almost exclusively through vision. Here we examine how the hippocampus contributes to sensory processing beyond the visual modality and to the multisensory integration of visual information with these other modalities. Participants were exposed to multiple repetitions of short naturalistic movie clips, each presented in four formats: auditory-only, visual-only, congruent audiovisual, and incongruent audiovisual (audio and video from different movies). Using high-resolution fMRI, we measured univariate activation and multivariate representations across the longitudinal axis and subfields of the human hippocampus. Whereas univariate analyses detected only visual responses across hippocampal subregions, with no activation for auditory stimuli and no benefit of congruent stimuli, multivariate analyses revealed robust representations of both auditory and visual scenes. The posterior hippocampus showed enhanced pattern similarity for congruent stimuli relative to unisensory stimuli, especially in the left hemisphere, demonstrating multisensory facilitation. The anterior hippocampus showed crossmodal decoding between auditory and visual versions of the same clip, driven by the right hemisphere, suggesting a more abstract representation. Finally, whole-brain searchlight analyses revealed parallel effects in cortical regions known to support multisensory integration. These findings advance understanding of auditory and multisensory coding in the human hippocampus, revealing specialization and lateralization along its longitudinal axis.

neuroscience↗

bioRxiv2025

Synthesizing images to map neural networks to the human brain

Computational models can be used to generate hypotheses about the brain. In the visual system, this approach has revealed similarities between how natural images are represented in convolutional neural networks and in cortical regions. However, natural images generate highly correlated representations across the hierarchy of model layers, meaning that each brain region will not correspond selectively to a single layer. Because each model layer performs additional image transformations, gaining a more selective mapping between individual layers and regions could reveal how specific algorithmic stages of processing are instantiated in the brain. To enable this mapping, we developed a generative framework for image synthesis that minimizes the similarity of image representational similarity matrices across model layers, aiming to orthogonalize the distances between the patterns of unit activations evoked by the same set of images. With the patterns orthogonalized, the resulting similarity matrix for each layer provides a fingerprint of the unique computational role of that layer. To test this approach, we synthesized 16 artificial images from the Inception-V1/GoogLeNet model and scanned participants with fMRI while they viewed these images repeatedly in random order. Image-specific patterns of voxel activity were used to compute image-by-image similarity matrices across the whole brain with searchlights. Most layers could be mapped to circumscribed cortical regions, and these mappings overlapped less than the mappings obtained with natural images. Given the prevalence of existing fMRI datasets with natural images, we used the synthesis method as a benchmark to develop an alternative residual method that can achieve comparable performance for natural image datasets. These approaches could be extended to other neural network architectures and stimulus modalities for targeted mappings of model computations to the brain and behavior.

neuroscience↗

bioRxiv2025

Visual perceptual learning enhances functional connectivity in retinotopic space

Repeated exposure to perceptual tasks improves behavioral performance. Several neural mechanisms have been proposed to account for such perceptual learning. Computational modeling suggests that plasticity in the connectivity between cortical sites may be responsible, by increasing the fidelity with which task-relevant information is transmitted through sensory hierarchies. Here we explore this theory in humans using fMRI, testing the hypothesis that perceptual learning at one location in space will increase functional connectivity between voxels in visual areas that are tuned to that retinotopic location. Participants learned to detect one of two novel visual shape contours embedded in a noisy background in different visual quadrants. At baseline, there was no difference in behavioral sensitivity for the two shapes, nor a difference in functional connectivity between voxels in V1 and V4 responsive to the retinotopic locations of the two shapes. After training, there was a robust and selective improvement in perceptual detection of the trained shape relative to the control shape, along with increased functional connectivity between V1 and V4 voxels coding for the location of the trained shape in retinotopic space. Moreover, this increase in functional connectivity for the trained versus control shape predicted the improvement in behavioral sensitivity across participants. These results are consistent with the proposal that perceptual learning alters network dynamics so as to enhance the processing of behaviorally relevant information.

neuroscience↗

bioRxiv2025

Stress drives the hippocampus to prioritize statistical prediction over episodic encoding

The hippocampus plays a critical role in encoding individual experiences into episodic memory, while also detecting patterns shared across these experiences that allow us to anticipate future events via statistical learning. Acute stress is known to impact the hippocampus but may have opposite effects on these competing functions. That is, although stress impairs episodic encoding, it may instead enhance statistical learning. Using functional magnetic resonance imaging, we tested how stress influences the hippocampal pathways and subfields that support episodic encoding and statistical learning while participants performed a task that engaged both processes. Across several analyses, stress biased hippocampal processing in favour of statistical learning, suppressing pattern separation between events in the dentate gyrus and enhancing the representation of statistically predictive features. Furthermore, stress drove the hippocampal monosynaptic pathway to support statistical learning rather than episodic encoding during predictive situations. Together, these data suggest that acute stress elicits targeted, adaptive changes in hippocampal pathways which may facilitate predicting and responding to upcoming events.

neuroscience↗

bioRxiv2025

Accelerated learning of a noninvasive human brain-computer interface via manifold geometry

Brain-computer interfaces (BCIs) promise to restore and enhance a wide range of human capabilities. However, a barrier to the adoption of BCIs is how long it can take users to learn to control them. We hypothesized that human BCI learning could be accelerated by leveraging the naturally occurring geometric structure of brain activity, or its intrinsic manifold, extracted using a data-diffusion process. We trained participants on a noninvasive BCI that allowed them to gain real-time control of an avatar in a virtual reality game by modulating functional magnetic resonance imaging (fMRI) activity in brain regions that support spatial navigation. We then perturbed the mapping between fMRI activity patterns and the movement of the avatar to test our manifold hypothesis. When the new mapping respected the intrinsic manifold, participants succeeded in regaining control of the BCI by aligning their brain activity within the manifold. When the new mapping sdid not respect the intrinsic manifold, participants could not learn to control the avatar again. These findings show that the manifold geometry of brain activity constrains human learning of a complex cognitive task in higher-order brain regions. Manifold geometry may be a critical ingredient for unlocking the potential of future human neurotechnologies.

neuroscience↗

bioRxiv2025

Representation of visual sequences in the tuning and topology of neuronal activity in the human hippocampus

Time is a critical component of memory and yet how the hippocampus incorporates temporal information with the sensory contents of memories remains unclear. We hypothesized that the hippocampus can learn arbitrary sequences through rapid changes in the tuning and population geometry of individual neurons to mirror the sequence structure. We recorded single-unit activity from 134 neurons in the hippocampus of 17 patients with epilepsy while they viewed sequences of visual scenes that were presented repeatedly in the same order by looping from the end to start. As the sequence repeated, hippocampal neurons that were initially responsive to one scene in the sequence began responding to the other scenes proportional to their temporal proximity. The population vectors of spiking activity across recorded hippocampal neurons for each scene came to form a high-dimensional ring topology that encoded the circular structure of the sequence in a manner that preserved the serial order of the scenes. These effects were not observed when sequences were scrambled upon each repetition to destroy temporal structure, nor in brain regions outside of the hippocampus for both structured and random sequences. These findings suggest that temporal structure governs the representation of sensory stimuli in human hippocampal neurons.

neuroscience↗

bioRxiv2025

Data retention in awake infant fMRI: Lessons from more than 750 scanning sessions

Functional magnetic resonance imaging (fMRI) in awake infants has the potential to reveal how the early developing brain gives rise to cognition and behavior. However, awake infant fMRI poses significant methodological challenges that have hampered wider adoption. The present work takes stock after the collection of a substantial amount of awake infant fMRI data across multiple studies from two labs at different institutions. These data were leveraged to glean insights on participant recruitment, experimental design, and data acquisition that could be useful to consider for future studies. Across 766 fMRI sessions with awake infants aged 1-36 months, the authors explored the factors that influenced how much usable data were obtained per session. The age of an infant predicted whether they would successfully enter the scanner (younger more likely) and, if they did enter, the number of minutes of functional data collected (linear, younger more) and retained after preprocessing with lab-specific protocols or harmonized motion exclusion thresholds (quadratic, 12-24 months more than younger and older). The amount of functional data retained was also influenced by assigned sex (female more), experimental paradigm (movies better than blocks and events), and stimulus content (social better than abstract). There were many differences in the research approach between labs making head-to-head comparisons difficult, but Yale was more likely to get infants into the scanner, MIT collected more data from infants who entered, and the amount of data retained after preprocessing did not differ statistically between labs (9 minutes). In addition, the authors assessed the value of attempting to collect multiple experiments per session, an approach that yielded more than one usable experiment averaging across all sessions. Although any given scan is unpredictable, these findings support the feasibility of awake infant fMRI and suggest practices to optimize future research.

neuroscience↗

bioRxiv2024

The role of REM sleep in neural differentiation of memories in the hippocampus

When faced with a familiar situation, we can use memory to make predictions about what will happen next. If such predictions turn out to be erroneous, the brain can adapt by differentiating the representations of the cues that generated the prediction from the mispredicted item itself, reducing the likelihood of future prediction errors. Prior work by Kim et al. (2017) found that violating a sequential association in a statistical learning paradigm triggered differentiation of the neural representations of the associated items in the hippocampus. Here, we used fMRI to test the preregistered hypothesis that this hippocampal differentiation occurs only when violations are followed by rapid eye movement (REM) sleep. In the morning, participants first learned that some items predict others (e.g., A predicts B) then encountered a violation in which a predicted item (B) failed to appear when expected after its associated item (A); the predicted item later appeared on its own after an unrelated item. Participants were then randomly assigned to one of three conditions: remain awake, take a nap containing non-REM sleep only, or take a nap with both non-REM and REM sleep. While the predicted results were not observed in the preregistered left CA2/3/DG ROI, we did observe evidence for our hypothesis in closely related hippocampal ROIs, uncorrected for multiple comparisons: In right CA2/3/DG, differentiation in the group with REM sleep was greater than in the groups without REM sleep (wake and non-REM nap); this differentiation was item-specific and concentrated in right DG. Differentiation effects were also greater in bilateral DG when the predicted item was more strongly reactivated during the violation. Overall, the results presented here provide initial evidence linking REM sleep to changes in the hippocampal representations of memories in humans.

neuroscience↗

bioRxiv2023

Inducing representational change in the hippocampus through real-time neurofeedback

AO_SCPLOWBSTRACTC_SCPLOWWhen you perceive or remember one thing, other related things come to mind. This competition has consequences for how these items are later perceived, attended, or remembered. Such behavioral consequences result from changes in how much the neural representations of the items overlap, especially in the hippocampus. These changes can reflect increased (integration) or decreased (differentiation) overlap; previous studies have posited that the amount of coactivation between competing representations in cortex determines which will occur: high coactivation leads to hippocampal integration, medium coactivation leads to differentiation, and low coactivation is inert. However, those studies used indirect proxies for coactivation, by manipulating stimulus similarity or task demands. Here we induce coactivation of competing memories in visual cortex more directly using closed-loop neurofeedback from real-time fMRI. While viewing one object, participants were rewarded for implicitly activating the representation of another object as strongly as possible. Across multiple real-time fMRI training sessions, they succeeded in using the neurofeedback to induce coactivation. Compared with untrained objects, this coactivation led to behavioral and neural integration: The trained objects became harder for participants to discriminate in a categorical perception task and harder to decode from patterns of fMRI activity in the hippocampus.

neuroscience↗

bioRxiv2023

Movies reveal the fine-grained organization of infant visual cortex

Studying infant minds with movies is a promising way to increase engagement relative to traditional tasks. However, the spatial specificity and functional significance of movie-evoked activity in infants remains unclear. Here we investigated what movies can reveal about the organization of the infant visual system. We collected fMRI data from 15 awake infants and toddlers aged 5-23 months who attentively watched a movie. The activity evoked by the movie reflected the functional profile of visual areas. Namely, homotopic areas from the two hemispheres responded similarly to the movie, whereas distinct areas responded dissim-ilarly, especially across dorsal and ventral visual cortex. Moreover, visual maps that typi-cally require time-intensive and complicated retinotopic mapping could be predicted, albeit imprecisely, from movie-evoked activity in both data-driven analyses (i.e., independent com-ponent analysis) at the individual level and by using functional alignment into a common low-dimensional embedding to generalize across participants. These results suggest that the infant visual system is already structured to process dynamic, naturalistic information and that fine-grained cortical organization can be discovered from movie data.

neuroscience↗

bioRxiv2023

Hippocampal Codes for Real-World Human Navigation

As animals move through the world, the hippocampus represents their location, direction, and speed. Parallel studies in humans have been mostly limited to virtual navigation because of physical constraints of suitable neural recording technologies. However, there are known differences between real-world and virtual navigation, leaving open the question of how the hippocampus supports real-world navigation in humans. Here we report evidence from ambulatory patients with chronic brain implants that the location, direction, and speed of humans walking along a linear track are represented in local field potentials from the hippocampus. We further show in a subset of patients who were tested twice after long delays that these representations can be reliable over time. These findings provide the first demonstration of multiple, stable neural codes for real-world navigation in the human hippocampus.

neuroscience↗

bioRxiv2023

Statistical learning in epilepsy: Behavioral, anatomical, and causal mechanisms in the human brain

Statistical learning, the fundamental cognitive ability of humans to extract regularities across experiences over time, engages the medial temporal lobe in the healthy brain. This leads to the hypothesis that statistical learning may be impaired in epilepsy patients, and that this impairment could contribute to their varied memory deficits. In turn, epilepsy patients provide a platform to advance basic understanding of statistical learning by helping to evaluate the necessity of medial temporal lobe circuitry through disease and causal perturbations. We implemented behavioral testing, volumetric analysis of the medial temporal lobe substructures, and direct electrical brain stimulation to examine statistical learning across a cohort of 61 epilepsy patients and 28 healthy controls. Behavioral performance in a statistical learning task was negatively associated with seizure frequency, irrespective of where seizures originated in the brain. The volume of hippocampal subfields CA1 and CA2/3 correlated with statistical learning performance, suggesting a more specific role of the hippocampus. Indeed, transient direct electrical stimulation of the hippocampus disrupted statistical learning. Furthermore, the relationship between statistical learning and seizure frequency was selective: behavioral performance in an episodic memory task was impacted by structural lesions in the medial temporal lobe and by antiseizure medications, but not by seizure frequency. Overall, these results suggest that statistical learning may be hippocampally dependent and that this task could serve as a clinically useful behavioral assay of seizure frequency distinct from existing neuropsychological tests. Simple and short statistical learning tasks may thus provide patient-centered endpoints for evaluating the efficacy of novel treatments in epilepsy.

neuroscience↗

bioRxiv2023

Differentiation and Integration of Competing Memories: A Neural Network Model

What determines when neural representations of memories move together (integrate) or apart (differentiate)? Classic supervised learning models posit that, when two stimuli predict similar outcomes, their representations should integrate. However, these models have recently been challenged by studies showing that pairing two stimuli with a shared associate can sometimes cause differentiation, depending on the parameters of the study and the brain region being examined. Here, we provide a purely unsupervised neural network model that can explain these and other related findings. The model can exhibit integration or differentiation depending on the amount of activity allowed to spread to competitors -- inactive memories are not modified, connections to moderately active competitors are weakened (leading to differentiation), and connections to highly active competitors are strengthened (leading to integration). The model also makes several novel predictions -- most importantly, that when differentiation occurs as a result of this unsupervised learning mechanism, it will be rapid and asymmetric, and it will give rise to anticorrelated representations in the region of the brain that is the source of the differentiation. Overall, these modeling results provide a computational explanation for a diverse set of seemingly contradictory empirical findings in the memory literature, as well as new insights into the dynamics at play during learning.

neuroscience↗

bioRxiv2023

Functional networks in the infant brain during sleep and wake states

Functional brain networks are assessed differently early in development than at maturity: infants are almost universally scanned during sleep, whereas adults are typically scanned awake while resting or performing tasks. Observed differences between infant and adult functional networks may thus reflect these differing states of consciousness rather than or in addition to developmental changes. We explore this question by comparing functional networks in fMRI scans acquired from infants during natural sleep and awake movie-watching. As a reference, we also acquired fMRI scans in adults during awake rest and awake movie-watching. Whole-brain functional connectivity was more similar within-state (sleep-sleep, wake-wake) than across-state (sleep-wake) in infants, demonstrating that movies elicit a different network configuration than typical sleep acquisitions. Indeed, a classifier trained on patterns of functional connectivity during infant sleep versus wake robustly decoded the state of additional infants and even generalized to decode rest versus movie in adults; interestingly, a classifier trained on rest versus movie in adults did not generalize nearly as well to sleep versus wake in infants. Moreover, the overall level of similarity between infant and adult functional connectivity was modulated by adult state (stronger for movie than rest) but not infant state (equivalent for sleep and wake). Nevertheless, the network connections that drove similarity between infants and adults, particularly in frontoparietal network, were modulated by infant state. In sum, infant functional connectivity can differ between sleep and wake states, highlighting the potential value of awake data for studying the early development of functional brain networks. Significance statementFunctional networks in the infant brain provide a foundation for early cognitive abilities and act as a marker of brain maturation and developmental disorders. What we know about these networks comes from fMRI data acquired during sleep, given the challenges of awake infant fMRI. This contrasts with the dominant approach in older populations of assessing networks during awake rest or tasks. These differing levels of consciousness cloud the interpretation of developmental changes. Here we show that whole-brain functional connectivity differs between sleeping and awake infants, and that the similarity of these infant states to adults loads on dissociable network connections. This research suggests that a full understanding of early functional brain networks will benefit from complementary insights in awake infants.

neuroscience↗