2025 CCN Workshop: Functional alignment of information encoded in neural activity
Organizers: Jim Haxby, Elizabeth DuPre, Ida Gobbini, Martin Lindquist, Feilong Ma, and Tor Wager
Dates: September 3 and September 4
Location: Kemeny Hall, Dartmouth College
Program
Workshop videos
Speakers
Erica Busch, Yale University
Harnessing neural manifolds of behavior and cognition
Abstract: Human cognition involves countless simultaneous operations for processing, representing, and acting upon the world. How do our brains -- systems with finite computational resources -- support such complexity and reflect the richness and flexibility of psychological experience? In this talk, I present several studies leveraging neural manifolds -- the low-dimensional, latent structure of brain activity -- to illuminate the neural bases of behavior and cognition, within and between individuals. Manifold trajectories and representations reveal shared and unique features of brain activity across narrative comprehension, visual perception, affective experience, emotion processing, and rest. Properties of neural manifolds vary across development (from infancy to childhood and adulthood), distinguish between brain states and tasks, and predict youth risk of mental health issues. Finally, I discuss the potential of neural manifold learning for modeling and guiding changes in real-world behaviors.
Elizabeth DuPre, Université de Montréal
Improving individual brain models with functional alignment
Abstract : “Dense sampling” datasets, in which a small number of participants are intensively scanned, are increasingly adopted across cognitive neuroscience. While these datasets have allowed researchers to learn participant-specific models—revealing important, previously overlooked structure within individual representations—the resulting models have also shown limited generalizability, both across datasets as well as across participants. In their current form, these individual models are therefore of limited clinical and translational utility. In this talk, I will review work developing functional alignment as a method to improve the generalization of individual brain models. I will focus on how the optimal transport problem provides a rich theoretical framework both for learning individual alignments as well as for incorporating biological constraints. I will also introduce our ongoing work in extending these applications, highlighting how this emerging research area mirrors broader challenges for learning in data-limited regimes across theoretical neuroscience.
Caterina Gratton, University of Illinois Urbana-Champaign
Alike and not alike: using precision fMRI to uncover individual differences and shared principles of brain network organization
Abstract: Recent years have emphasized that people differ strikingly from one another in their brain network topography, especially in the fine scale details of this organization. These observations suggest a need to shift to approaches that can accurately capture individual people’s brain organization. Precision fMRI — which is based on extensive data collection within individuals, combined with advanced denoising and network definition methods — can be used to address these needs. In this talk, I will review recent work from our group and others that demonstrates how precision fMRI can be used to characterize individual differences in brain network organization, as well as to identify general principles of brain organization that are shared across people.
James V. Haxby, Dartmouth College
Functional alignment: essential elements and desirable applications
Abstract: Hyperalignment aligns shared information that is embedded in idiosyncratic fine-scale cortical functional architecture. I will present a brief description of hyperalignment as a general framework for functional alignment. The essential elements of this framework are a high-dimensional embedding space for shared information – also called a common model space or template – and transformations that resample idiosyncratic individual functional topographies into the dimensions of the common model embedding space. Hyperalignment uses a continuous, whole-brain embedding space that has the same number of dimensions as the vertices in a high-resolution cortical template (ON-AVG), and each dimension is linked to a vertex, allowing visualization of topographies and statistical results in an anatomically meaningful format. The hyperalignment embedding space and transformations can be based on functional connectivity, affording more flexible application that is not dependent on shared stimuli. Hyperalignment was developed originally to solve the problem of between subject decoding (between-subject multivariate pattern classification – MVPC) but has been developed further to provide high-fidelity estimates of individual functional topographies (e.g. retinotopy and visual categories) and a basis for the study of individual differences in the information embedded in fine-grained cortical architecture, which is more reliable and more predictive of cognitive traits than differences in coarse-grained architecture
Ruby Kong, National University of Singapore
Individual-specific brain parcellations for clinical applications
Abstract: Resting-state functional connectivity (RSFC) has emerged as a powerful approach for characterizing large-scale brain networks. Yet, inconsistencies in how these networks are spatially defined and labeled across studies have posed challenges for interpretation and reproducibility in the field. To address this, we have developed the Network Correspondence Toolbox (NCT) to permit researchers to examine and report spatial correspondence between their novel neuroimaging results and multiple widely used functional brain atlases.
While population-level maps provide valuable insight, they may obscure biologically meaningful, individual-specific features. To better capture individual variability in brain organization, we developed a multi-session hierarchical Bayesian model that generates high-quality, individualized functional parcellations. We further examined how these individual-specific parcellations relate to behavior and explored their potential clinical applications in personalized brain stimulation.
Martin Lindquist, Johns Hopkins Bloomberg School of Public Health
Individualized spatial topography in functional neuroimaging
Abstract: Neuroimaging is poised to take a substantial leap forward in understanding the neurophysiological underpinnings of human behavior, due to a combination of improved analytic techniques and the quality of imaging data. These advances are allowing researchers to develop population-level multivariate models of the functional brain representations underlying behavior, performance, clinical status and prognosis, and other outcomes. Population-based models can identify patterns of brain activity, or ‘signatures’, that can predict behavior and decode mental states in new individuals, producing generalizable knowledge and highly reproducible maps. These signatures can capture behavior with large effect sizes and can be used and tested across research groups. However, the potential of such signatures is limited by neuroanatomical constraints, in particular individual variation in functional brain anatomy. To circumvent this problem, current models are either applied only to individual participants, severely limiting generalizability, or force participants’ data into anatomical reference spaces (atlases) that do not respect individual functional topology and boundaries. Here we seek to overcome this shortcoming by developing new topographical models for inter-subject alignment, which register participants’ functional brain maps to one another. This increases effective spatial resolution, and more importantly allows us to explicitly analyze the spatial topology of functional maps and make inferences on differences in activation location and shape across persons and psychological states. In this talk we discuss several approaches towards functional alignment and highlight promises and pitfalls.
Feilong Ma, University of South Carolina
Individual differences in fine-grained brain functional architecture
Abstract: A key focus in neuroscience is to characterize how brains differ from each other, and how these differences relate to cognition, affect, mental health, and social behaviors. Functional alignment methods, such as hyperalignment, afford a powerful tool to separate the information content encoded in the brain (“what”) from how each brain encodes the information in idiosyncratic topographies (“where”). In other words, hyperalignment factors out topographic idiosyncrasies. It brings fMRI data into alignment at the vertex/voxel level, allowing for the examination of individual differences in fine-grained spatial patterns. Through a series of studies, I will illustrate that individual differences in fine-grained brain functional architecture are (a) highly reliable across independent data, and (b) highly predictive of cognitive abilities. Lastly, I will present the Individualized Neural Tuning (INT) model, which leverages hyperalignment algorithms to characterize individual differences in brain functional architecture and to accurately predict individualized brain responses to new stimuli.
Bogdan Petre, Dartmouth College
Dual processes underlying interindividual differences in brain maps
Abstract: Interindividual variation in functional brain maps reflects at least two separable sources: protomap patterning during development and experience-dependent refinement. Using HCP resting-state and task data, we show that functional topographies (how computations are implemented) are less consistent and more heritable than representational geometry (what computational distinctions a region makes). Convergent representations with idiosyncratic topographies are especially prevalent in cortical areas where architectural features like thickness, myelination, and network hierarchy are most permissive of learning. This motivates a dual-process model where genetic blueprints are adjusted through experience and yield convergent computations with variable implementations.
We test this model with nested alignment comparisons. A reduced model applies diffeomorphic alignment to capture protomap-like differences. An augmented model adds hyperalignment in high-dimensional representational space. Diffeomorphic alignment best predicts unimodal responses, while the combined approach performs best in transmodal cortex, supporting the dual-process account.
To facilitate testing and application, we introduce an open, modular pipeline integrating individualized resting-state networks, MSM, and ANTs. This provides a practical foundation for improved functional localization, representational similarity analysis, and hyperalignment workflows that respect both sources of individual variation.
Vincent Taschereau-Dumouchel, University of Montreal
How visual representations become distressing thoughts
Abstract: We all have access to the same visual world, yet our experiences of reality can diverge in striking ways. Recent advances in naturalistic functional magnetic resonance imaging (fMRI) have enabled researchers to map the semantics of visual and narrative stimuli onto their underlying neural representations. But can these methods also reveal how psychological factors—such as phobia or delusion—shape and distort semantic representations in the brain? In this session, we will present a series of experiments probing the visual information processing hierarchy to identify where shared sensory representations give way to distortions introduced by psychological states. These findings will be connected to emerging conceptualisations that distinguish perceptual from “thought-like” neural representations, and open new opportunities for more naturalistic modeling of thoughts and subjective experience in intensive fMRI.
Alex Williams, NYU
Leveraging decoders to quantify similarity in neural population codes
Abstract: A common approach to characterize the information encoded by a neural population is to build regression models or "decoders" that reconstruct sensory input or behavioral output from neural responses. In this talk, I will explain how decoding analyses can be leveraged to quantify differences in neural population codes across subjects and across deep artificial neural network models. I will contrast this approach with established geometric measures of similarity, including representational similarity analysis (RSA), canonical correlation analysis (CCA), and Procrustes shape distance. Perhaps surprisingly, there are deep yet concise mathematical relationships between these two perspectives, providing a novel basis for unified understanding. This is based on joint work with Sarah E. Harvey and David Lipshutz.