Driving and suppressing the human language network using large language models. Nature Human Behaviour (2024).
     Tuckute, G., Sathe, A., Srikant, S., Taliaferro, M., Wang, M., Schrimpf, M., Kay, K., Fedorenko, E.
     Journal link

This study develops predictive models for a large fMRI dataset targeting linguistic representations elicited by individual sentences. As a demonstration of the generalization power of these models, synthetic sentences predicted to either strongly drive or suppress fMRI responses are shown to indeed produce these very effects.

Tasks and their role in visual neuroscience. Neuron (2023).
     Kay, K., Bonnen, K., Denison, R.N., Arcaro, M.J., Barack, D.L.
     Journal link

In this review, we offer a broad perspective on the concept of tasks, review the diverse ways that tasks impact neural activity in the visual system, and outline an approach for the formal modeling of tasks.

Engaging in word recognition elicits highly specific modulations in visual cortex. Current Biology (2023).
     White, A.L., Kay, K., Tang, K.A., Yeatman, J.D.
     Journal link

Task demands can affect visually evoked responses. Here, we find highly specific task effects in visual word form area (VWFA). Using a tightly controlled paradigm, we demonstrate enhancement of responses in VWFA when the subject is directed to perform a lexical decision task on visually presented words. These effects do not occur in nearby cortical regions nor do they occur in a closely matched control task.

Improving the accuracy of single-trial fMRI response estimates using GLMsingle. eLife (2022).
     Prince, J.S., Charest, I., Kurzawski, J.W., Pyles, J.A., Tarr, M., Kay, K.N.
     Journal link

Functional MRI data suffer from limited signal-to-noise ratio. Here we describe an analysis toolbox, termed GLMsingle, that incorporates several techniques to deliver improved estimates of fMRI responses at the single-trial level. Using several existing fMRI datasets as examples, we demonstrate substantial downstream benefits that are of interest to cognitive and systems neuroscientists. GLMsingle is freely available at glmsingle.org.

Mesoscopic in vivo human T2* dataset acquired using quantitative MRI at 7 Tesla. NeuroImage (2022).
     Gulban, O.F., Bollmann, S., Huber, R., Wagstyl, K., Goebel, R., Poser, B.A., Kay, K.*, Ivanov, D.*
     Journal link

Increasing the spatial resolution of neuroimaging is critical for improving our understanding of the relationship between macroscopic and microscopic brain measurements. Here, we present and release high-quality mesoscopic quantitative T2* measurements of the living human brain. We show these measurements exhibit clear structures such as cortical layers and intracortical vessels, indicating the utility of these data for further investigations.

Non-Neural Factors Influencing BOLD Response Magnitudes within Individual Subjects. Journal of Neuroscience (2022).
     Kurzawski, J.W., Gulban, O.F., Jamison, K., Winawer, J.*, Kay, K.*
     Journal link

Valid interpretation of fMRI data requires understanding physical units and what they might signify. Here, we study possible non-neural contributions to variation in percent BOLD signal change (%BOLD) across primary visual cortex. Such variations are very large and found consistently across different datasets.

The risk of bias in data denoising methods: examples from neuroimaging. PLoS ONE (2022).
     Kay, K.
     Journal link

Denoising methods can seemingly improve data quality, but it is important to carefully think about the possibility of bias. Here, we review classic statistical concepts of bias and variance and provide simple simulations (with accompanying code) demonstrating these concepts in examples drawn from neuroimaging. The results provide insight into the nature of denoising.

A massive 7T fMRI dataset to bridge cognitive neuroscience and artificial intelligence. Nature Neuroscience (2022).
     Allen, E.J., St-Yves, G., Wu, Y., Breedlove, J.L., Prince, J.S., Dowdle, L.T., Nau, M., Caron, B., Pestilli, F., Charest, I., Hutchinson, J.B., Naselaris, T.*, & Kay, K.*
     Journal link

In each of 8 carefully screened participants, we sample responses to 9,000–10,000 distinct natural scenes over the course of 30–40 7T fMRI scan sessions. We also develop novel estimation and denoising methods to improve the accuracy of single-trial response estimates. This neuroimaging resource emphasizes extensive sampling of individuals to better understand cognitive function.

See also News and Views by Botch, Roberston, & Finn

Holistic face recognition is an emergent phenomenon of spatial processing in face-selective regions. Nature Communications (2021).
     Poltoratski, S., Kay, K., Finzi, D., & Grill-Spector, K.
     Journal link | PDF

Here, we perform systematic fMRI measurements to upright and inverted faces varying in spatial position and find systematic shifts in face-selective regions of human visual cortex. We then link these effects to variations in perceptual performance across the visual field. The results highlight the critical dependence of both neural representation and behavioral performance on the precise positioning of stimuli relative to fixation.

Extensive sampling for complete models of individual brains. Current Opinion in Behavioral Sciences (2021).
     Naselaris, T., Allen, E., & Kay, K.
     Journal link | PDF

An argument for why we should collect extensive amounts of data on a small number of subjects, as opposed to modest amounts of data on many subjects.

Fractional Ridge Regression: a Fast, Interpretable Reparameterization of Ridge Regression. GigaScience (2020).
     Rokem, A. & Kay, K.
     Pubmed link | Journal link | PDF

Ridge regression is a useful statistical technique for solving regression models. Here, we improve its application by showing how to better parameterize (and therefore estimate) its hyperparameter. MATLAB and Python code toolboxes are provided.

A temporal decomposition method for identifying venous effects in task-based fMRI. Nature Methods (2020).
     Kay, K., Jamison, K.W., Zhang, R.-Y., Ugurbil, K.
     Pubmed link | Journal link | PDF

fMRI temporal dynamics contain information regarding underlying vascular sources of the fMRI signal. Here, we develop a new algorithm and associated code toolbox that helps to estimate and therefore isolate the contributions of large macrovessels.

Understanding multivariate brain activity: evaluating the effect of voxelwise noise correlations on population codes in functional magnetic resonance imaging. PLoS Computational Biology (2020).
     Zhang, R.-Y., Wei, X.-X., Kay, K.
     Pubmed link | Journal link | PDF

The consequence of 'noise correlations' has been heavily studied at the level of single neurons, but little explored at the population level accessible via functional neuroimaging. We provide some theoretical considerations for how these effects may manifest in fMRI data.

Appreciating the variety of goals in computational neuroscience. Neurons, Behavior, Data analysis, and Theory (2020).
     Kording, K.P., Blohm, G., Schrater, P., & Kay, K.
     Journal link | PDF

In this opinion piece, we argue that computational modeling is a rich exercise that can be used and applied in many different ways. An informal survey of computational researchers reveals that this is indeed the case. Thus, we should adopt an flexible, open mind when thinking about and evaluating computational work.

Flexible top-down modulation in human ventral temporal cortex. NeuroImage (2020).
     Zhang, R.-Y. & Kay, K.
     Pubmed link | Journal link | PDF

Here, we reanalyze a previously acquired fMRI dataset to provide further evidence of the complex way in which top-down modulations can manifest in visual cortex, when probed with sufficiently demanding and complex perceptual tasks. The results point to the need for general and comprehensive models that can account for a wide range of observations.

An image-computable model for the stimulus selectivity of gamma oscillations. eLife (2019).
     Hermes, D., Petridou, N., Kay, K.*, & Winawer, J.*
     Pubmed link | Journal link | PDF

The presence of gamma oscillations in neural signals have been intriguing in terms of their source and relevant for brain function. We develop an image-computable model that helps formalize and characterize the nature of these effects; the model is based on high levels of orientation variance.

A critical assessment of data quality and venous effects in sub-millimeter fMRI. NeuroImage (2019).
     Kay, K., Jamison, K.W., Vizioli, L., Zhang, R., Margalit, E., & Ugurbil, K.
     Pubmed link | Journal link | PDF

Sub-millimeter fMRI is a novel technique that pushes the boundaries of what is possible with fMRI. In this detailed assessment, we investigate and characterize the nature of sub-millimeter fMRI data including the substantial (and potentially unwanted) influence of the brain's vasculature on the data, and discuss the practical implications of these issues.

The HCP 7T Retinotopy Dataset: Description and pRF Analysis. Journal of Vision (2018).
     Benson, N., Jamison, K.W., Arcaro, M.J., Vu, A., Glasser, M.F., Coalson, T.S., Van Essen, D., Yacoub, E., Ugurbil, K., Winawer, J.*, & Kay, K.*
     Pubmed link | Journal link | PDF

Here we describe a freely available, very large 7T fMRI retinotopic mapping dataset (n = 181), and document basic analyses of the data that indicate the data's robustness and quality.

GLMdenoise improves multivariate pattern analysis of fMRI data. NeuroImage (2018).
     Charest, I., Kriegeskorte, N., & Kay, K.
     Pubmed link | Journal link | PDF

We evaluate the previously introduced GLMdenoise technique in terms of its consequence for MVPA- and RSA-flavor analyses, and demonstrate sizable improvements in data quality.

Principles for models of neural information processing. NeuroImage (2018).
     Kay, K.N.
     Pubmed link | Journal link | PDF

In this perspective piece, we lay out basic principles for how one might approach building models of neural information processing and why current deep neural network models of the brain may not necessarily satisfy these principles.

Cognitive Computational Neuroscience: A New Conference for an Emerging Discipline. Trends in Cognitive Sciences (2018).
     Naselaris, T., Bassett, D.S., Fletcher, A.K., Kording, K., Kriegeskorte, N., Nienborg, H., Poldrack, R.A., Shohamy, D., & Kay, K.
     Pubmed link | Journal link | PDF

This paper describes the motivation and goals of a new annual conference (CCN), and reports thoughts and observations that arose at the inaugural event.

Compressive Temporal Summation in Human Visual Cortex. Journal of Neuroscience (2017).
     Zhou, J., Benson, N.C., Kay, K., & Winawer, J.
     Pubmed link | Journal link | PDF

Analogous to compressive summation across the visual field, this paper measures and demonstrates compressive summation over time and how this effect systematically varies across different visual areas.

Binocular Rivalry: A Window into Cortical Competition and Suppression. Journal of the Indian Institute of Science (2017).
     Zhang, R., Engel, S.A., & Kay, K.
     Journal link | PDF

A review of the phenomenon of binocular rivalry.

Defining the most probable location of the parahippocampal place area using cortex-based alignment and cross-validation. NeuroImage (2017).
     Weiner, K., Barnett, M.A., Witthoft, N., Golarai, G., Stigliani, A., Kay, K.N., Gomez, J., Natu, V.S., Amunts, K., Zilles, K., Grill-Spector, K.
     Pubmed link | Journal link | PDF

A systematic investigation of identifying the location of PPA and its consistency across distinct individuals.

Bottom-up and top-down computations in word- and face-selective cortex. eLife (2017).
     Kay, K.N. & Yeatman, J.D.
     Pubmed link | Journal link | PDF | PDF (Figure Supplements)

Here, we tackle the nature of representation and the influence of top-down modulation in high-level visual areas (FFA and VWFA). Using systematic manipulations of stimulus type and the task performed by the observer, we identify and build a computational model of bottom-up and top-down effects. The model is simple, principled, and interpretable, and accounts for clear effects observable in the data.

Resolving ambiguities of MVPA using explicit models of representation. Trends in Cognitive Sciences (2015).
     Naselaris, T. & Kay, K.N.
     Pubmed link | Journal link | PDF

We describe a few limitations of traditional MVPA approaches to analyzing functional brain activity, and suggest the value of building explicit models of representation.

Visual representations are dominated by intrinsic fluctuations correlated between areas. NeuroImage (2015).
     Henriksson, L., Khaligh-Razavi, S., Kay, K., & Kriegeskorte, N.
     Pubmed link | Journal link | PDF

We show the very large impact of 'noise correlations' on representational similarity analysis of fMRI data, and demonstrate a simple analysis that can remove these effects.

Attention reduces spatial uncertainty in human ventral temporal cortex. Current Biology (2015).
     Kay, K.N., Weiner, K.S., & Grill-Spector, K.
     Pubmed link | Journal link | PDF (Main text + Supplemental)

We demonstrate selectivity for visual space in face-selective visual regions, estimate population receptive fields (pRFs) at the voxelwise level, and show how pRFs are affected by the task performed by the observer. We conclude that the representation of space is a critical component of the ventral stream and that a full understanding of visual representations must take into account the cognitive state of the observer.

Computational modeling of responses in human visual cortex. In: Brain Mapping: An Encyclopedic Reference, edited by P. Thompson & K. Friston (2015).
     Wandell, B.A., Winawer, J., & Kay, K.N.
     Journal link | PDF

Here we review and provide a perspective on using computational modeling to understand basic aspects of representation in human visual cortex.

GLMdenoise: a fast, automated technique for denoising task-based fMRI data. Frontiers in Neuroscience (2013).
     Kay, K.N., Rokem, A., Winawer, J., Dougherty, R.F. & Wandell, B.A.
     Pubmed link | Journal link | PDF

We present a simple, principled denoising algorithm for task-based fMRI based on combining data-driven noise derivation and cross-validation. The algorithm is demonstrated on a variety of datasets and is packaged into a MATLAB code implementation.

Asynchronous broadband signals are the principal source of the BOLD response in human visual cortex. Current Biology (2013).
     Winawer, J., Kay, K.N., Foster, B.L., Rauschecker, A.M., Parvizi, J., & Wandell, B.A.
     Pubmed link | Journal link | PDF (Main text) | PDF (Supplemental Information)

Using a controlled visual paradigm, we identify two distinct aspects of electrocorticographical (ECoG) signals and show that 'broadband' asynchronous responses most closely match what is observed in fMRI BOLD signals.

A two-stage cascade model of BOLD responses in human visual cortex. PLoS Computational Biology (2013).
     Kay, K.N., Winawer, J., Rokem, A., Mezer, A., & Wandell, B.A.
     Pubmed link | Journal link | PDF

Using carefully controlled stimuli, we develop a two-stage hierarchical cascade model of linear and nonlinear computations that underlie visually evoked responses in early and intermediate cortex. The model indicates the prominent role of second-order contrast computations in extrastriate visual regions.

Compressive spatial summation in human visual cortex. Journal of Neurophysiology (2013).
     Kay, K.N., Winawer, J., Mezer, A., & Wandell, B.A.
     Pubmed link | Journal link | PDF (Main text) | PDF (Supporting Information)

We demonstrate the existence of sub-additive spatial summation in fMRI BOLD responses in many visual areas, and propose a simple nonlinearity into a population receptive field (pRF) model to accommodate this observation. The nonlinearity is shown to vary systematically across visual areas and can be interpreted as a means by which visual responses gain tolerance for changes in the position and size of objects. 

Understanding visual representation by developing receptive-field models. In: Visual Population Codes: Towards a Common Multivariate Framework for Cell Recording and Functional Imaging, edited by N. Kriegeskorte & G. Kreiman (2011).
     Kay, K.N.
     Book link | PDF

Here we provide a perspective on why understanding visual representation is challenging and how explicit image-computable models can help tackle these challenges.

Encoding and decoding in fMRI. NeuroImage (2011).
     Naselaris, T., Kay, K.N., Nishimoto, S. & Gallant, J.L.
     Pubmed link | Journal link | PDF

In this review, we provide a framework for understanding how different 'encoding' and 'decoding' fMRI studies are related to one another. The notion of a 'feature space' is introduced, which is one way to help clarify the similarities and differences therein.

Bayesian reconstruction of natural images from human brain activity. Neuron (2009).
     Naselaris, T., Prenger, R.J., Kay, K.N., Oliver, M. & Gallant, J.L.
     Pubmed link | Journal link | PDF (Main text) | PDF (Supplementary Information)

Here we demonstrate a method for decoding the stimulus viewed by an observer based on brain activity. We introduce the idea of putting priors on the stimulus distribution and include a method for incorporating information conveyed in high-level visual areas.

I can see what you see. Nature Neuroscience (2009).
     Kay, K.N. & Gallant, J.L.
     Pubmed link | Journal link | PDF

We comment on Miyawaki et al., Neuron, 2008 and put it into historical context.

Identifying natural images from human brain activity. Nature (2008).
     Kay, K.N., Naselaris, T., Prenger, R.J. & Gallant, J.L.
     Pubmed link | Journal link | PDF (Main text) | PDF (Supplementary Information) | Commentary

Using extensive sampling of many natural images shown to the observer during fMRI scanning, this paper develops an 'encoding model' approach in which an explicit image-computable model (based on Gabor filters) is proposed to characterize how voxels in early visual cortex represent the viewed image. After building this model, it is then demonstrated that the model is accurate enough such that the specific image viewed by an observer can be identified with high accuracy based solely on the brain activity elicited by the image.

Modeling low-frequency fluctuation and hemodynamic response timecourse in event-related fMRI. Human Brain Mapping (2008).
     Kay, K.N., David, S.V., Prenger, R.J., Hansen, K.A. & Gallant, J.L.
     Pubmed link | Journal link | PDF

Here we evaluate and demonstrate methods for estimating timecourses and signal drifts in event-related fMRI, with an emphasis on precision at the single voxel level. Cross-validation is shown as a means for controlling overfitting.

Topographic organization in and near human visual area V4. The Journal of Neuroscience (2007).
     Hansen, K.A., Kay, K.N. & Gallant, J.L.
     Pubmed link | Journal link | PDF

Novel retinotopic mapping stimuli and techniques are used to map early and intermediate human visual cortex using fMRI. Based on the resulting maps, a proposal for V4 in human visual cortex is made.

Better models of human high-level visual cortex emerge from natural language supervision with a large and diverse datasets. Nature Machine Intelligence (2023).
     Wang, A.Y., Kay, K., Naselaris, T., Tarr, M.J., Wehbe, L.
     Journal link

Here, we demonstrate improvements in modeling fMRI activity based on neural network models that are trained on language-based descriptions of visual inputs. The results suggest a close connection between the processing of visual inputs in the brain and the extraction of linguistic information from visual inputs.

Re-expression of CA1 and entorhinal activity patterns preserves temporal context memory at long timescales. Nature Communications (2023).
     Zou, F., Guo, W., Allen, E.J., Wu, Y., Charest, I., Naselaris, T., Kay, K., Kuhl, B.A., Hutchinson, J.B.*, DuBrow, S.*
     Journal link

We exploit the Natural Scenes Dataset which provides a unique look at neural activity as humans store and recall memories across long timescales. Evidence is shown for the encoding of temporal context in the hippocampus and entorhinal cortex.

Brain-optimized deep neural network models of human visual areas learn non-hierarchical representations. Nature Communications (2023).
     St-Yves, G., Allen, E.J., Wu, Y., Kay, K.*, Naselaris, T.*
     Journal link

Deep neural network models have been entertained as possible models of biological visual systems. Here, we examine the common presumption that the hierarchical structure of such models reflects an important facet of computation in brains. Using the Natural Scenes Dataset, we demonstrate brain-optimized models that predict brain activity well but do not reflect straightforward hierarchical relationships. This suggests deeper and more precise characterizations are necessary to understand the connection between brains and models.

Color-biased regions in the ventral visual pathway are food selective. Current Biology (2022).
     Pennock, I.M.L., Racey, C., Allen, E.J., Wu, Y., Naselaris, T., Kay, K.N., Franklin, A., Bosten, J.M.
     Journal link

Color-biased regions have been reported in the ventral visual stream. Here, we use exploit the rich sampling present in the Natural Scenes Dataset to investigate the representational properties of these regions. We find that both color and form independently contribute to increased activity in these regions. In particular, these regions appear to be preferentially activated by images with food content, even for images with low color saturation.

Natural scene sampling reveals reliable coarse-scale orientation tuning in human V1. Nature Communications (2022).
     Roth, Z.N., Kay, K.*, Merriam, E.P.*
     Journal link

Here, we leverage the large-scale Natural Scenes Dataset to estimate voxel-level orientation tuning using model-based methods that compensate for vignetting (or edge) effects. Results demonstrate clear evidence for coarse-scale orientation representational biases in primary visual cortex.

Variability of the Surface Area of the V1, V2, and V3 Maps in a Large Sample of Human Observers. Journal of Neuroscience (2022).
     Benson, N.C., Yoon, J., Forenzo, D., Engel, S.A., Kay, K.N., Winawer, J.
     Journal link

Using the freely available Human Connectome Project 7T Retinotopy Dataset, we trained human "anatomists" to manually define the size and extent of visual areas V1, V2, and V3. The results enable us to demonstrate a very wide range of sizes (up to 3.5x) of these early visual areas across healthy young adults. These results indicate the challenge of comparing and combining measures of brain structure and function across individuals.

Multiple traces and altered signal-to-noise in systems consolidation: Evidence from the 7T fMRI Natural Scenes Dataset. PNAS (2022).
     Vanasse, T.J., Boly, M., Allen, E.J., Wu, Y., Naselaris, T., Kay, K., Cirelli, C., Tononi, G.
     Journal link

To study systems-level memory consolidation, we leverage the longitudinal large-scale Natural Scenes Dataset measuring recognition performance for images over the course of nearly a year. We find that the medial temporal lobe contribution to recognition persists over 200 days, supporting multiple-trace theory.

NeuroGen: activation optimized image synthesis for discovery neuroscience. NeuroImage (2022).
     Gu, Z., Jamison, K.W., Khosla, M., Allen, E.J., Wu, Y., St-Yves, G., Naselaris, T., Kay, K., Sabuncu, M.R., Kuceyeski, A.
     Journal link

Visual representations of naturalistic scenes in the brain are rich and complex. Here, a computational framework is developed that uses optimization to generate synthetic stimuli designed to achieve specific targeted activations in the brain. This technique may prove useful for controlling brain activity and generating hypotheses about visual representations.

Distinct representations of tonotopy and pitch in human auditory cortex. Journal of Neuroscience (2022).
     Allen, E., Mesik, J., Kay, K., Oxenham, A.
     Journal link

Using 7T fMRI and carefully controlled stimuli, we identify auditory cortical regions that reflect pitch per se, as opposed to simpler forms of spectral content.

The retrocalcarine sulcus maps different retinotopic representations in macaques and humans. Brain Structure and Function (2021).
     Arcaro, M.J., Livingstone, M.S., Kay, K.N., & Weiner, K.S.
     Journal link

Careful examination of functional neuroanatomy reveals that specific sulci in occipital cortex are predictive of retinotopic representations within species, but diverge substantially across species. This observation places constraints on theories of structure-function relationships.

Sensory recruitment revisited: Ipsilateral V1 involved in visual working memory. Cerebral Cortex (2021).
     Zhao, Y., Kay, K.N., Tian, Y., & Ku, Y.
     Pubmed link | Journal link

Here, we use fMRI and a visual working memory task to demonstrate evidence for ipsilateral cortex in maintaining visual working memory representations.

Emerging ethical issues raised by highly portable MRI research in remote and resource-limited international settings. NeuroImage (2021).
     Shen, F.X., Wolf, S.M., Bhavnani, S., Deoni, S., Elison, J.T., Fair, D., Geethanath, S., Garwood, M., Gee, M.S., Kay, K., Lim, K.O., Estrin, G.L., Luciana, M., Peloquin, D., Rommelfanger, K., Schiess, N., Siddiqui, K., Torres, E., & Vaughan, J.T.
     Pubmed link | Journal link | PDF

The prospects of portable MRI scanners open new avenues of possibilities in research, clinical, and everyday applications.

Ultra-high-resolution fMRI of Human Ventral Temporal Cortex Reveals Differential Representation of Categories and Domains. The Journal of Neuroscience (2020).
     Margalit, E., Jamison, K.W., Weiner, K.S., Vizioli, L., Zhang, R.-Y., Kay, K.N.*, & Grill-Spector, K.*
     Pubmed link | Journal link | PDF

We use high-resolution 7T fMRI to probe the nature of stimulus representations in ventral temporal cortex, and find a medial-lateral distinction with respect to how categories and domains are represented.

Predicting neuronal dynamics with a delayed gain control model. PLoS Computational Biology (2019).
     Zhou, J., Benson, N.C., Kay, K., & Winawer, J.
     Pubmed link | Journal link | PDF

The temporal dynamics of neural responses to simple presentations of stimuli show complex behaviors. Here we develop a simple unified computational model that can account for a wide range of these phenomena.

Foreground-background segmentation revealed during natural image viewing. eNeuro (2018).
     Papale, P., Leo, A., Cecchetti, L., Handjaras, G., Kay, K., Pietrini, P., & Ricciardi, E.
     Pubmed link | Journal link | PDF

Using image-computable models applied to perturbed versions of the original input stimuli, this study investigates the nature of foreground and background representation in visual cortex.

A non-invasive, quantitative study of broadband spectral responses in human visual cortex. PLOS One (2018).
     Kupers, E.R., Wang, H.X., Amano, K., Kay, K.N., Heeger, D.J., & Winawer, J.
     Pubmed link | Journal link | PDF

Here we introduce a new EEG denoising method that aids in the measurements of broadband responses in EEG.

The Functional Neuroanatomy of Human Face Perception. Annual Review of Vision Science (2017).
     Grill-Spector, K., Weiner, K.S., Kay, K., & Gomez, J.
     Pubmed link | Journal link | PDF

A review of the anatomical and functional properties of human visual cortex that subserve the perception of faces.

A new modular brain organization of the BOLD signal during natural vision. Cerebral Cortex (2017).
     Kim, D., Kay, K., Shulman, G., & Corbetta, M.
     Pubmed link | Journal link | PDF

Using both resting-state and movie-watching fMRI data, this paper applies connectivity analyses to characterize the network architecture of macroscopic brain areas observed at rest and during naturalistic stimulation.

Encoding and decoding V1 fMRI responses to natural images with sparse nonparametric models. Annals of Applied Statistics (2011).
     Vu, V.Q., Ravikumar, P., Naselaris, T., Kay, K.N., Gallant, J.L. & Yu, B.
     Pubmed link | Journal link | PDF

Using V1 fMRI responses as an example data scenario, we demonstrate statistical techniques that enable the learning of flexible nonlinearities in large-scale regression models.

Nonparametric sparse hierarchical models describe V1 fMRI responses to natural images. In: Advances in Neural Information Processing Systems 21, edited by D. Koller, D. Schuurmans, Y. Bengio, & L. Bottou (2009).
     Ravikumar, P., Vu, V.Q., Yu, B., Naselaris, T., Kay, K.N. & Gallant, J.L.
     Book link | PDF

 A demonstration of sparse additive models to improve encoding-model performance in V1.