Related Experiment Video
Updated: Sep 19, 2025

The HoneyComb Paradigm for Research on Collective Human Behavior
Published on: January 19, 2019
Heterogeneity, reinforcement learning, and chaos in population games
Jakub Bielawski1, Thiparat Chotibut2, Fryderyk Falniowski1
1Department of Mathematics, Krakow University of Economics, Kraków 31-510, Poland.
Multiagent reinforcement learning dynamics in congestion games can become chaotic with differing agent learning rates, increasing social costs. However, average behavior still converges to stable Nash equilibria.
Area of Science:
- Game Theory
- Machine Learning
- Dynamical Systems
Background:
- Congestion games are a class of potential games with well-understood Nash equilibria.
- Agents in these games have negligible individual impact and aligned incentives.
Purpose of the Study:
- Investigate multiagent reinforcement learning (MARL) dynamics in congestion games with heterogeneous learning rates.
- Analyze deviations from Nash equilibria and resulting social costs.
- Explore the link between microscopic dynamics and macroscopic outcomes.
Main Methods:
- Studied discrete-time MARL dynamics in nonatomic congestion games.
- Employed dynamical systems techniques to analyze agent adaptation and population-level outcomes.
- Examined scenarios with diverse agent beliefs and varying learning rates.
Main Results:
- MARL dynamics with heterogeneous learning rates can exhibit instability and chaotic behavior, deviating from static Nash equilibria.
- These chaotic regimes lead to increased social costs.
- Time-averaged macroscopic behavior converges to exact Nash equilibria despite microscopic complexity.
Conclusions:
- Heterogeneous learning rates in MARL can introduce complex dynamics in congestion games.
- A link exists between chaotic microscopic dynamics and stable macroscopic equilibrium behavior.
- Findings pave the way for studying complex learning dynamics in discrete time.
Related Concept Videos
Mutation, Gene Flow, and Genetic Drift
Genetic Drift
Generalization, Discrimination, and Extinction
Generalization occurs when a behavior reinforced in one context is performed in similar situations. For instance, a student who studies diligently for calculus and receives excellent grades might apply the same study habits to psychology and history, expecting similar results. Generalization shows how learning in one setting can influence behavior in...
What is Population Genetics?
Analysis of Population Pharmacokinetic Data
Mechanistic Models: Compartment Models in Individual and Population Analysis

