About Me

I am a researcher at MIT (CoCoSci) studying how people perceive, learn, and reason effectively given limited cognitive resources. I did my graduate work with Steve Piantadosi at UC Berkeley. My thesis work was aimed at understanding the mechanisms underlying visual numerosity perception. I've used behavioral experiments and computational models to link the visual information people gather from a scene to their ultimate perceptions of quantities. In one project, I showed that the psychophysics of number, including "subitizing" small quantities and Weber's law for larger quantities, reflects optimal inference under a limited informational capacity. See an explainer here. You can read a précis of my thesis here.

My recent projects explore how people learn geometric patterns, reason about physical systems, and actively seek information in ways that respect their cognitive limitations. One direction I am particularly excited about is modeling and experimentally testing how people reason in complex settings without holding many pieces of information in memory simultaneously. In an ongoing project, we developed CRUMB (Constraint Reasoning Under Memory Bounds), a model of how people solve constraint problems like Minesweeper by building small subproblems that fit within a fixed memory capacity. See an interactive explainer here.

Interests

  • Bayesian & information-theoretic modeling
  • Visual perception
  • Numerical cognition
  • Active information-seeking
  • Concept learning

Education

  • Postdoctoral Researcher, 2022-Present
    MIT, Computational Cognitive Science Lab

  • PhD in Psychology, 2016-2022
    UC Berkeley, Computation and Language Lab

  • BS in Cognitive Science, 2012-2016
    Carnegie Mellon University
Curriculum Vitae
Download CV

Project Highlights

CRUMB: How people reason about problems too big to hold in mind
The CRUMB model incrementally building three subproblems out of a constraint graph
The model incrementally builds subproblems out of a larger constraint graph, adding one constraint at a time (blue squares) and only keeping proposals that fit in memory. It abandons the first subproblem without making deductions, then commits to variable assignments (black circles) in the next two.

Human active learning balances informativeness and interpretability
Informativity-interpretability tradeoff illustration
An illustration of the informativity-interpretability tradeoff. Paths show sequences of intermediate representations from uncertainty to certainty. Higher-information queries reach certainty in fewer steps but require more complex intermediate representations.

A serial, foveal accumulator underlies approximate numerical estimation
Example fixation paths
Example fixation paths. Bottom labels show N (dots shown), F (dots foveated), and E (participant estimate).

A unified account of numerosity perception
Probability of numeric responses over time for N=3, 6, 9 with model fits
The probability (y-axis) of numeric responses (x-axis) over presentation times (faceted) for N=3, N=6, and N=9. Bars are shown for the human data and lines are shown for the model predictions.

Spatiotemporal program learning across development and species
Program induction model and participant predictions across timepoints
Left: Illustration of a program induction model predicting how a 2-D sequence unfolds. Right: Predictions at selected timepoints from each population; older children and adults show more structured predictions; younger children and monkeys tend to track the locally linear trend.

Selected Publications

Adaptive, online subproblem construction explains how people reason about complex problems

Cheyette, S.J., Chen, T., Hofer, M., Callaway, F., Bramley, N., & Tenenbaum, J.B.

In preparation

Explainer →

Just-in-Time World Modeling Supports Human Planning and Reasoning

Chen, T., Cheyette, S.J., Allen, K.R., Tenenbaum, J.B., & Smith, K.A.

Under Review

Human active learning trades off informativity and interpretability

Cheyette, S.J., Callaway F.L., Bramley N., Nelson, J., Tenenbaum J.B.

Under Review

PDF →

Spatiotemporal program learning in human adults, children, and monkeys

Mills, T., Coates, N., Silva, A.A., Ji, K., Ferrigno, S., Schulz, L.E., Tenenbaum, J.B., Cheyette, S.J.

Proceedings of the National Academy of Sciences (2026)Forthcoming

PDF →

People use fast and flat simulation to reason about new games

Collins, K.M., Zhang, C., Wong, L., Barba, C.M., Todd, G., Weller, A., Cheyette, S.J., Griffiths, T., & Tenenbaum, J.B.

Nature (2026)

Longitudinal, self-directed gameplay as a window into the development of expertise

Cheyette, S.J., Chen, T., Hofer, M., Callaway, F., Bramley, N., & Tenenbaum, J.B.

Proceedings of the 48th Annual Cognitive Science Conference (2026)

Do humans use push-down stacks when learning or producing center-embedded structures?

Ferrigno, S., Cheyette, S.J., & Carey, S.

Cognitive Science (2025)

PDF →

Decompose, deduce, and dispose: A memory-limited, metacognitive model of human problem-solving

Cheyette, S.J., Chen, T., Hofer, M., Callaway, F., Bramley, N., & Tenenbaum, J.B.

Proceedings of the 47th Annual Cognitive Science Conference (2025)

Limited information-processing capacity in vision explains number psychophysics

Cheyette, S. J., Wu, S., & Piantadosi, S. T.

Psychological Review (2024)

PDF →

Response to difficulty drives variation in IQ test performance

Cheyette, S. J. & Piantadosi, S. T.

Open Mind (2024)

PDF →

Spatiotemporal pattern learning as probabilistic program synthesis

Mills, T., Tenenbaum, J. B., & Cheyette, S. J.

Neural Information Processing Systems (2024)

PDF →

A unified account of numerosity perception

Cheyette, S. J. & Piantadosi, S. T.

Nature Human Behaviour (2020)

PDF →

Recursive sequence generation in monkeys, children, and native Amazonians

Ferrigno, S., Cheyette, S. J., Piantadosi, S. T., & Cantlon, J.

Science Advances (2020)

PDF →

A primarily serial, foveal accumulator underlies approximate numerical estimation

Cheyette, S. J. & Piantadosi, S. T.

Proceedings of the National Academy of Sciences (2019)

PDF →

Modeling the N400 ERP component as transient semantic over-activation within a neural network model of word comprehension

Cheyette, S. J., Plaut, D. C.

Cognition (2017)

PDF →