A new method called PEM-UDE can discover the hidden mathematical rules behind chaotic systems — including networks of brain cells — even when experimental data is extremely noisy.
The research
A team led by Anthony G. Chesebro and Helmut H. Strey at Stony Brook University, with collaborators including Earl K. Miller at MIT and Christopher V. Rackauckas, published the study on arXiv (submitted July 4, 2025; updated August 17, 2026). They combined two existing approaches: prediction-error methodology, which smooths the tricky process of fitting chaotic systems, and universal differential equations, which let a computer search for the missing mathematical terms in a model.
The researchers tested PEM-UDE on two benchmark chaotic systems: the Rössler attractor, a standard mathematical model, and a real electrical circuit. In both cases, the method recovered the correct equations — even when one observed dimension contained noise five times larger than the signal itself.
For the neuroscience application, the team applied prior knowledge of how neurons connect and fire. They used a population of Izhikevich neurons, a common model that mimics real neuron behavior. PEM-UDE produced a multi-scale neural mass model that links single-neuron properties to network-level dynamics. The model predicts how connection density relates to the brain's dominant oscillation frequency and synchrony — the tendency of neurons to fire together.
The team checked these predictions against three intracranial recording datasets from rat and human cortices. Importantly, the recordings serve as an indirect consistency check of the predicted trends, not a direct fit. The learned equations are a reduced-order closure for a specific simulated network family, meaning they summarize complex neuron behavior in a simpler form.
Why it matters
Your brain is a chaotic, noisy system. Understanding its underlying rules could help researchers build better models of attention, memory, and cognitive flexibility — the very processes that support learning and problem-solving. This study shows that machine learning can extract interpretable equations from messy data, which is a step toward personalized brain models. For anyone interested in cognition, it highlights that complexity and noise don't have to hide the signal. With the right mathematical tools, meaningful patterns can emerge from chaos.
What you can do
- Embrace noise: Your cognitive performance varies day to day. Track your focus, mood, and sleep to find patterns despite the ups and downs.
- Look for hidden structure: When learning a new skill, pay attention to underlying rules — like grammar in a language or chord progressions in music — rather than memorizing isolated facts.
- Train adaptively: Just as PEM-UDE refines equations, brain training that adapts to your performance can strengthen specific cognitive skills.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.