About

Kenneth Norman is a Professor in Computational and Theoretical Neuroscience at Princeton University, affiliated with the Princeton Neuroscience Institute. His research focuses on using computational models to explore how the brain gives rise to learning and memory phenomena. Norman's lab employs neuroimaging studies to decode thoughts as individuals learn and remember, investigating questions such as the learning rules that govern memory modification, the role of sleep in learning, how memories are time-stamped, and methods for intentionally forgetting memories. His work involves developing new machine learning methods for analyzing distributed neural activity patterns, including data mining algorithms to isolate fMRI and EEG signatures of specific thoughts and memories. Norman's research also includes creating real-time neurofeedback techniques that adapt studies based on participants' thoughts. His contributions extend to advancing understanding of mental illness through collaborative research and developing innovative analysis tools for neuroimaging data.

Research topics

• Psychology
• Natural Language Processing
• Artificial Intelligence
• Computer Science
• Neuroscience
• Cognitive psychology
• Linguistics
• Machine Learning
• Mathematics
• Cognitive science
• Speech recognition
• Communication
• Biology
• Econometrics
• Social psychology
• Statistics
• World Wide Web

Selected publications

• Large language models can segment narrative events similarly to humans

  Behavior Research Methods · 2025-01-02 · 17 citations

  articleOpen access
  PublisherOA PDFDOI

• Neural codes track prior events in a narrative and predict subsequent memory for details

  Communications Psychology · 2025-02-16 · 6 citations

  articleOpen accessSenior author

  Throughout our lives, we learn schemas that specify what types of events to expect in particular contexts and the temporal order in which these events usually occur. Here, our first goal was to investigate how such context-dependent temporal structures are represented in the brain during processing of temporally extended events. To accomplish this, we ran a 2-day fMRI study (N = 40) in which we exposed participants to many unique animated videos of weddings composed of sequences of rituals; each sequence originated from one of two fictional cultures (North and South), where rituals were shared across cultures, but the transition structure between these rituals differed across cultures. The results, obtained using representational similarity analysis, revealed that context-dependent temporal structure is represented in multiple ways in parallel, including distinct neural representations for the culture, for particular sequences, and for past and current events within the sequence. Our second goal was to test the hypothesis that neural schema representations scaffold memory for specific details. In keeping with this hypothesis, we found that the strength of the neural representation of the North/South schema for a particular wedding predicted subsequent episodic memory for the details of that wedding.

  PublisherOA PDFDOI

• Eye movements reveal the cognitive dynamics supporting successful memory suppression

  2025-05-09

  preprintOpen access

  Prior work has shown that stopping memory retrieval can lead to lasting forgetting (suppression-induced forgetting; SIF); however, the factors that lead to successful SIF are not fully understood. Here, we used a novel eye tracking paradigm to examine how participants interact with memory cues to support their retrieval goals. In a preregistered sample (N=34), we show that instructions to retrieve were associated with gaze reinstatement (eye gaze towards the location where the target object was studied), while instructions to suppress were associated with gaze repulsion (eye gaze away from the studied location). We also show that lasting SIF is associated with an initial period of gaze reinstatement during the suppression attempt. These results provide novel evidence for how humans intentionally forget, revealing both the top-down strategies that people employ to suppress retrieval and also the retrieval dynamics that lead to lasting forgetting.

  PublisherOA PDFDOI

• Synthesizing images to map neural networks to the human brain

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-11-25

  preprintOpen access

  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.

  PublisherDOI

• Towards large language models with human-like episodic memory

  Trends in Cognitive Sciences · 2025-07-26 · 3 citations

  review
  PublisherDOI

• The Role of Rapid Eye Movement Sleep in Neural Differentiation of Memories in the Hippocampus

  Journal of Cognitive Neuroscience · 2025-07-30 · 1 citations

  articleOpen accessSenior author

  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 cue from the mispredicted item itself, reducing the likelihood of future prediction errors. Prior work by Kim, G., Norman, K. A., and Turk-Browne, N. B. Neural differentiation of incorrectly predicted memories. Journal of Neuroscience, 37, 2022-2031 [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. Participants first learned that some items predict others (e.g., A predicts B) and 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/dentate gyrus (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. REM-related differentiation effects were also greater in bilateral DG when the predicted item was more strongly reactivated during the violation. Overall, these results provide initial evidence linking REM sleep to changes in the hippocampal representations of memories in humans.

  PublisherOA PDFDOI

• Music-evoked reactivation during continuous perception is associated with enhanced subsequent recall of naturalistic events

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-07-06

  preprintOpen accessSenior authorCorresponding

  Music is a potent cue for recalling personal experiences, yet the neural basis of music-evoked memory remains elusive. We address this question by using the full-length film Eternal Sunshine of the Spotless Mind to examine how repeated musical themes reactivate previously encoded events in cortex and shape next-day recall. Participants in an fMRI study viewed either the original film (with repeated musical themes) or a no-music version. By comparing neural activity patterns between these groups, we found that music-evoked reactivation of neural patterns linked to earlier scenes in the default mode network was associated with improved subsequent recall. This relationship was specific to the music condition and persisted when we controlled for a proxy measure of initial encoding strength (spatial intersubject correlation), suggesting that music-evoked reactivation may play a role in making event memories stick that is distinct from what happens at initial encoding.

  PublisherOA PDFDOI

• Fast-timescale hippocampal processes bridge between slowly unfurling neocortical states during memory search

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-02-13 · 1 citations

  preprintOpen accessSenior author

  Prior behavioral work showed that event structure plays a key role in our ability to mentally search through memories of continuous naturalistic experience. We hypothesized that, neurally, this memory search process involves a division of labor between slowly unfurling neocortical states representing event knowledge and fast hippocampal-neocortical communication that supports retrieval of new information at transitions between events. To test this, we tracked slow neural state-patterns in a sample of ten patients undergoing intracranial electroencephalography as they viewed a movie and then searched their memories in a structured naturalistic interview. As patients answered questions ("after X, when does Y happen next?"), state-patterns from movie-viewing were reinstated in neocortex; during memory-search, states unfurled in a forward direction. Moments of state-transition were marked by low-frequency power decreases in cortex and preceded by power decreases in hippocampus that correlated with reinstatement. Connectivity-analysis revealed information-flow from hippocampus to cortex underpinning state-transitions. Together, these results support our hypothesis that fast hippocampal processes bridge between slow neocortical states during memory search.

  PublisherDOI

• Reconciling shared versus context-specific information in a neural network model of latent causes

  Scientific Reports · 2024-07-22 · 7 citations

  articleOpen accessSenior author

  It has been proposed that, when processing a stream of events, humans divide their experiences in terms of inferred latent causes (LCs) to support context-dependent learning. However, when shared structure is present across contexts, it is still unclear how the "splitting" of LCs and learning of shared structure can be simultaneously achieved. Here, we present the Latent Cause Network (LCNet), a neural network model of LC inference. Through learning, it naturally stores structure that is shared across tasks in the network weights. Additionally, it represents context-specific structure using a context module, controlled by a Bayesian nonparametric inference algorithm, which assigns a unique context vector for each inferred LC. Across three simulations, we found that LCNet could (1) extract shared structure across LCs in a function learning task while avoiding catastrophic interference, (2) capture human data on curriculum effects in schema learning, and (3) infer the underlying event structure when processing naturalistic videos of daily events. Overall, these results demonstrate a computationally feasible approach to reconciling shared structure and context-specific structure in a model of LCs that is scalable from laboratory experiment settings to naturalistic settings.

  PublisherOA PDFDOI

• Blocked training facilitates learning of multiple schemas

  Communications Psychology · 2024-04-09 · 12 citations

  articleOpen accessSenior author

  We all possess a mental library of schemas that specify how different types of events unfold. How are these schemas acquired? A key challenge is that learning a new schema can catastrophically interfere with old knowledge. One solution to this dilemma is to use interleaved training to learn a single representation that accommodates all schemas. However, another class of models posits that catastrophic interference can be avoided by splitting off new representations when large prediction errors occur. A key differentiating prediction is that, according to splitting models, catastrophic interference can be prevented even under blocked training curricula. We conducted a series of semi-naturalistic experiments and simulations with Bayesian and neural network models to compare the predictions made by the "splitting" versus "non-splitting" hypotheses of schema learning. We found better performance in blocked compared to interleaved curricula, and explain these results using a Bayesian model that incorporates representational splitting in response to large prediction errors. In a follow-up experiment, we validated the model prediction that inserting blocked training early in learning leads to better learning performance than inserting blocked training later in learning. Our results suggest that different learning environments (i.e., curricula) play an important role in shaping schema composition.

  PublisherOA PDFDOI

Recent grants

• Computational, Neural, and Behavioral Studies of Competition-Dependent Learning

  NIH · $7.2M · 2004–2030

• NRSA Training Grant in Quantitative Neuroscience

  NIH · $7.1M · 2002–2029

• Neural dynamics supporting integration and recall over long timescales during natural continuous input

  NIH · $528k · 2017–2021

• NCS-FO: Collaborative Research: Sleep's role in determining the fate of individual memories

  NSF · $594k · 2015–2019

• Text, Neuroimaging, and Memory: Unified Models of Corpora and Cognition

  NSF · $732k · 2010–2015

Frequent coauthors

• Nicholas B. Turk‐Browne

  Yale University

  99 shared

• Uri Hasson

  Neuroscience Institute

  98 shared

• Jonathan D. Cohen

  Princeton University

  56 shared

• Alex Nguyen

  Concordia University

  34 shared

• Victoria J. H. Ritvo

  Princeton University

  32 shared

• Samuel J. Gershman

  Harvard University

  31 shared

• Sebastian Michelmann

  Princeton University

  28 shared

• Samuel A. Nastase

  Princeton University

  28 shared

Labs

• Princeton Computational Memory LabPI

  Principal Investigator Ken Norman Huo Professor in Computational and Theoretical Neuroscience Professor of Psychology and Neuroscience Ph.D., Harvard University (1999) M.A., Harvard University (199…

Awards & honors

• Psychonomic Society Mid-Career Award