Smooth Dynamics, Good Performance in Cognitive-Agent Congestion Problems

Abstract

In a congestion game, individuals exhaust a common resource out of selfish behavior. In this scenario, drivers create traffic jams by choosing the shortest route according to their individual knowledge. They can avoid them by communicating their belief states about the traffic situation in real-time through a peer-to-peer network, assuming unlimited bandwidth. We study two potential, cognitively inspired models of human behavior: 1) categorization (quantized memorization and communication), which dampens communication and belief adoption, but leads to undesired oscillations and lower performance. 2) Instance-based blending with memory decay, which achieves good dynamics and near-optimal performance without the same bandwidth needs. We argue that this supports our hypothesis of co-adaptation of cognitive function and communicating communities.


Back to Table of Contents