Summary: Researchers used fMRI to show that the brain handles pattern learning differently from probabilistic learning.
Source: Ohio State University.
Detecting patterns is a fundamental part of human learning and decision-making. A new study reveals how the brain responds when people discover a deterministic pattern within a sequence of sensory information.
Researchers found that pattern learning engages distinct neural systems from those involved in probabilistic learning, suggesting the brain treats rule-based sequence detection differently than learning based on outcome probabilities.
In the experiment, 26 adult participants viewed 50 sequences of 12 gradually revealed images while undergoing functional magnetic resonance imaging (fMRI). Each sequence contained various arrangements of three image categories — a hand, a face and a landscape — presented either in a repeating pattern or in random order. Participants were instructed to press a button as soon as they recognized which of the three images was being revealed, with faster correct responses earning higher monetary rewards.
“We could see which brain regions became active when participants discovered that a sequence followed a pattern, and when they realized that no pattern was present,” said Ian Krajbich, co-author and assistant professor of psychology and economics at The Ohio State University.
The task required participants to balance two kinds of uncertainty. One uncertainty concerned which specific image would appear next in the sequence, a form of prediction that relates closely to probabilistic learning models previously studied. The other uncertainty involved whether an underlying pattern governed the sequence at all — a structural or rule-based uncertainty.
When participants were uncertain about which image would appear next, brain activity matched regions previously linked to probabilistic prediction. However, when participants were resolving whether a pattern existed, the ventromedial prefrontal cortex (vmPFC) showed stronger activation. The vmPFC has been associated in prior research with value and reward processing, raising the possibility that identifying a pattern carries an intrinsic reward signal for the brain.
The study also highlighted a role for the hippocampus: greater hippocampal activity was associated with faster pattern learning among participants. This finding aligns with the hippocampus’s known involvement in memory formation and sequence learning, indicating its contribution when the brain extracts deterministic structure from sensory input.
Researchers modeled participants’ behavior with a Bayesian pattern-learning framework that tracks beliefs both about the current state (which image is likely next) and about the underlying structure (whether a deterministic rule governs the sequence). Reaction times and choice behavior were well predicted by this model, supporting the idea that human sequence learning can involve Bayesian inference over possible patterns as well as over likely outcomes.
“People in our study were not only estimating the odds of a particular image appearing next,” said Arkady Konovalov, a postdoctoral researcher who collaborated on the study. “They were discovering rules and using those rules to make faster, more accurate responses.”
Overall, the results demonstrate that the brain monitors multiple forms of uncertainty during sequence learning and recruits different neural networks for probabilistic prediction versus rule discovery. The distinction matters because rule detection allows the brain to move beyond estimating probabilities into forming explicit structures that support rapid and reliable predictions.
These insights broaden understanding of how humans learn structured relationships from limited evidence, with implications for models of decision making, memory, and adaptive behavior. By separating the neural signatures of pattern detection from those of probabilistic learning, the study provides a clearer picture of the computations the brain uses when searching for rules that simplify prediction and guide action.
Source: Ian Krajbich, Ohio State University
Publisher: NeuroscienceNews.com (organized content)
Image source: Public domain
Original research: “Neurocomputational Dynamics of Sequence Learning” by Arkady Konovalov and Ian Krajbich, published in Neuron (May 31, 2018).
DOI: 10.1016/j.neuron.2018.05.013
Neurocomputational Dynamics of Sequence Learning — Highlights
- Human subjects learned to detect patterns in sequences of images while undergoing fMRI.
- Behavior was well captured by a Bayesian pattern-learning model that tracks both state predictions and beliefs about underlying structure.
- Distinct neural networks reflect uncertainty about the next predicted image versus uncertainty about the presence of a pattern.
Summary
The brain can infer complex sequential structure from limited data, which supports model-based planning and reliable prediction. This study shows that humans form beliefs about both the current state and the hidden structure of sequences, and that these beliefs are reflected in separate neural signatures. The findings support the hypothesis that structure learning in the brain relies on Bayesian-like inference processes that integrate prior expectations about possible patterns with incoming evidence.