Controlling collective dynamics in complex minority-game resource-allocation systems

Abstract

Resource allocation takes place in various kinds of real-world complex systems, such as the traffic systems, social services institutions or organizations, or even the ecosystems. The fundamental principle underlying complex resource-allocation dynamics is Boolean interactions associated with minority games, as resources are generally limited and agents tend to choose the least used resource based on available information. A common but harmful dynamical behavior in resource-allocation systems is herding, where there are time intervals during which a large majority of the agents compete for a few resources, leaving many other resources unused. Accompanying the herd behavior is thus strong fluctuations with time in the number of resources being used. In this paper, we articulate and establish that an intuitive control strategy, namely pinning control, is effective at harnessing the herding dynamics. In particular, by fixing the choices of resources for a few agents while leaving majority of the agents free, herding can be eliminated completely. Our investigation is systematic in that we consider random and targeted pinning and a variety of network topologies, and we carry out a comprehensive analysis in the framework of mean-field theory to understand the working of control. The basic philosophy is then that, when a few agents waive their freedom to choose resources by receiving sufficient incentives, majority of the agents benefit in that they will make fair, efficient, and effective use of the available resources. Our work represents a basic and general framework to address the fundamental issue of fluctuations in complex dynamical systems with significant applications to social, economical and political systems.

Multiagent model and mean field theory of complex auction dynamics

“…The motivation of our work came from the observation that complex dynamical process of online auction bears certain similarity to minority game dynamics, a class of relatively well studied, multi-agent evolutionary game dynamics [21][22][23][24][25][26][27][28][29][31][32][33] that have proven capable of providing great insights into a variety of social and economical processes. Our idea is then to construct a multi-agent game model with the aim to better predict the empirically observed bid price distributions.…”

“…The rational upper bound of bid price k is k V c < -, ensuring positive payoff for the winner. Our construction of a computational model for the LUBA auction process benefits from the fact that a host of social and economical behaviors can be described by the minority game model [21][22][23][24][25][26][27][28][29][30][31][32][33][34][35][36][37], a multi-agent model characterizing a population of selfish individuals competing for limited common resources. The similarity between LUBA auction process and those described by the minority game model suggests strongly the suitability of using multi-agent models to understand the LUBA auction dynamics.…”

Overall fluctuations and fat tails in an artificial financial market: The two-sided impact of leveraged trading

“…The intelligent agents mainly represent institutional investors who can gain access to the use of leverage. When making trading decisions, these intelligent agents use heterogenous strategies which take the form modified from the famous minority game [15][16][17][18][19] small amount. When simulating the model, we only alter the value of leverage ratio of the intelligent agents, while the other market conditions and parameters including trading regulations are fixed.…”

Reinforcement learning meets minority game: Toward optimal resource allocation

“…In the original minority game model, an individual's scheme for state updating (or decision making) is essentially a trial-anderror learning process based on the global historical winning information [4]. In other models, learning mechanisms based local information from neighbors were proposed [11,12,16,17,25,28,[31][32][33][34][35]. The issue of controlling and optimizing complex resource allocation systems was also investigated [32], e.g., utilizing pinning control to harness the herding behavior, where it was demonstrated that a small number of control points in the network can suppress or even eliminate herding.…”

“…In other models, learning mechanisms based local information from neighbors were proposed [11,12,16,17,25,28,[31][32][33][34][35]. The issue of controlling and optimizing complex resource allocation systems was also investigated [32], e.g., utilizing pinning control to harness the herding behavior, where it was demonstrated that a small number of control points in the network can suppress or even eliminate herding. A theoretical framework for analyzing and predicting the efficiency of pinning control was developed [32], revealing that the connecting topology among the agents can play a significant role in the control outcome.…”