Structure learning and context inference emerge from next-state prediction in a cerebellar-medial temporal lobe circuit

Humans learn many context-dependent behaviors; yet the contexts driving these behavior are rarely explicit and must instead be inferred. Further, we are rarely exposed to the same context multiple times, so we must learn how to generalize to new contexts from our experiences in prior ones. Motor learning relies on both. People can hold multiple conflicting visuomotor mappings and recall the appropriate mapping when needed. Much as cricket is easier to learn for someone who already plays baseball, experience with previous mappings accelerates learning of others. Indeed, human motor learning is well described by context inference, in which the learner infers the context and uses it to update and recall the appropriate memory. How the brain implements this inference, and which circuits perform it, however, is unknown. Here we introduce a circuit in which a cerebellar forward model predicts the next state conditioned on a latent from the medial temporal lobe. At the same time, the medial temporal lobe updates that latent from the cerebellum's prediction error. While we only trained the circuit to predict next state, it learns to infer context and represents contexts in a structured latent space. Coupled with this latent, a motor cortex module reproduces context-dependent retrieval, savings after washout, spontaneous recovery, and evoked recovery, and compensates for perturbations it has never experienced. The same circuit reproduces cerebellar-dependence of object-location memory and independence in object-recognition memory, loss of place-cell stability in the dark when cerebellar plasticity is blocked, and the abstract geometry of hippocampal activity in a reversal task. Cerebellar–medial temporal lobe interaction is therefore a candidate locus for context inference and structure learning in the brain.

Authors

Institutions

Publication Details

Journal
bioRxiv (Cold Spring Harbor Laboratory)
Published
2026-11-16
DOI
https://doi.org/10.64898/2026.11.14.765964
Primary Topic
Memory and Neural Mechanisms
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Structure learning and context inference emerge from next-state prediction in a cerebellar-medial temporal lobe circuit

Elom A. Amematsro, Samuel Joseph Gershman, Cheshta Bhatia
1 links
bioRxiv (Cold Spring Harbor Laboratory)
Memory and Neural Mechanisms
preprint

Structure learning and context inference emerge from next-state prediction in a cerebellar-medial temporal lobe circuit

Elom A. Amematsro, Samuel Joseph Gershman, Cheshta Bhatia
preprint en

Abstract

Humans learn many context-dependent behaviors; yet the contexts driving these behavior are rarely explicit and must instead be inferred. Further, we are rarely exposed to the same context multiple times, so we must learn how to generalize to new contexts from our experiences in prior ones. Motor learning relies on both. People can hold multiple conflicting visuomotor mappings and recall the appropriate mapping when needed. Much as cricket is easier to learn for someone who already plays baseball, experience with previous mappings accelerates learning of others. Indeed, human motor learning is well described by context inference, in which the learner infers the context and uses it to update and recall the appropriate memory. How the brain implements this inference, and which circuits perform it, however, is unknown. Here we introduce a circuit in which a cerebellar forward model predicts the next state conditioned on a latent from the medial temporal lobe. At the same time, the medial temporal lobe updates that latent from the cerebellum's prediction error. While we only trained the circuit to predict next state, it learns to infer context and represents contexts in a structured latent space. Coupled with this latent, a motor cortex module reproduces context-dependent retrieval, savings after washout, spontaneous recovery, and evoked recovery, and compensates for perturbations it has never experienced. The same circuit reproduces cerebellar-dependence of object-location memory and independence in object-recognition memory, loss of place-cell stability in the dark when cerebellar plasticity is blocked, and the abstract geometry of hippocampal activity in a reversal task. Cerebellar–medial temporal lobe interaction is therefore a candidate locus for context inference and structure learning in the brain.

bioRxiv (Cold Spring Harbor Laboratory)
Harvard University (US)
U.S. Department of Defense, Multidisciplinary University Research Initiative, Army Research Office
Memory and Neural Mechanisms
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon