We study opinion dynamics in a population of $K$ interacting adaptive agents voting on a set of issues. The agents are modeled as Boolean Perceptrons that have to classify a set of $N$ dimensional random vectors ${\mathbfx_a}_a=1^P$ representing the issues being debated. At each interaction, the agents react to the classification of their neighbors in a social network by trying to learn the rule used to vote. The learning algorithm is parametrized in terms of how agents weight their neighbors agreement in relation to disagreement ($δ$) and of a learning rate ($η$). By simulating the model on a ring we observe that, if information is exchanged asynchronously, consensus only emerges for $δ=0$ and P=1, being in this case equivalent to the Voter model. For $δ>0$ and P=1 factions with extreme opposite beliefs emerge. Random moderate opinions are observed as the number $P$ of issues debated increases. The synchronous case is studied semi-analytically for $P\to∞$ and unidirectional information flow to show that consensus is a fixed point for $δ=0$ and $η\to0$. For $δ>0$ consensus is also a fixed point but the dynamics becomes glassy as $K$ grows larger.
Search
Reviews
[Write a review of this article]
Find related articles from these CiteULike users
