What the study found
Hebbian learning, meaning reinforcement of a connection after successful activation, can produce opposite global effects from local changes. The authors report that positive reinforcement tends to eliminate the active phase, while negative reinforcement can turn an inactive phase into a globally active one.
Why the authors say this matters
The study suggests that models of infection spreading, inter-regional brain activity propagation, and population spreading may need local learning rules to reflect how activity changes across scales. The findings indicate that local incentives can lead to emergent behaviors that depend on the microscopic learning mechanism and the initial conditions.
What the researchers tested
The researchers extended the contact process, a model of activity spreading, by adding learning as either positive (Hebbian) or negative (anti-Hebbian) reinforcement of activation rates between pairs of sites after each successful activation event. They combined analytical and numerical results to study the resulting behavior, including in two dimensions and above.
What worked and what didn't
Negative reinforcement was found to promote spreading of activity and to generate effectively immune regions at the same time, leading to two distinct critical points in two dimensions and above. Positive reinforcement could produce Griffiths effects, which are non-universal power-law scaling, associated with the 'ant-mill' phenomenon. The abstract states that rich emergent behavior depends on the details of the learning mechanism and the starting conditions.
What to keep in mind
The abstract does not provide detailed limitations beyond noting dependence on microscopic details and initial conditions. No additional caveats, parameter ranges, or application-specific constraints are described in the available summary.
Key points
- Positive reinforcement tends to remove the active phase.
- Negative reinforcement can make an inactive phase become globally active.
- In two dimensions and above, negative reinforcement can create two distinct critical points.
- Positive reinforcement can lead to Griffiths effects with non-universal power-law scaling.
- The abstract says the behavior depends on the learning details and initial conditions.
Disclosure
- Research title:
- Hebbian learning can shift activity spreading in opposite directions
- Authors:
- Will T. Engedal, Róbert Juhász, I. Kovács
- Institutions:
- Northwestern University, HUN-REN Wigner Research Centre for Physics
- Publication date:
- 2026-04-27
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.