Collective Behavior Of Resource Allocation And Robust Analysis In Complex System

Complex resource allocation systems provide the fundamental support for the normal functioning and well being of the modern society and present rich and inter-esting physical phenomena(herding,crowd panic,stampede).These phenomena have been extensively studied by researchers in recent years.Computationally and mathematically,such systems can be modeled as minority games.A ubiq-uitous dynamical phenomenon is the spontaneous emergence of herding,where a vast maj ority of the users concentrate on a small number of resources.From an op-erational point of view,herding is of grave concern as the few overused resources can be depleted quickly,directing users to the next few resources and causing them to fail,and so on,and eventually leading to a catastrophic collapse of the whole system in short time.To devise strategies to prevent herding from occurring is thus of interest.Previous works focused on control strategies that rely on external in-terventions,such as pinning control where a fraction of users are forced to choose a certain action.But its disadvantage is that it is expensive and difficult to oper-ate.Therefor,is it possible to eliminate herding without any external control?We provide an affirmative answer and give a detailed discussion in the third chapter of this paper.The second work of this paper is the robust research of supply-demand sys-tem.As everyone knows,supply-demand processes take place on a large variety of real-world networked systems ranging from power grids and the internet to so-cial networking and urban systems.With the rapid development of human society,supply-demand systems are constantly expanding,leading to constant increase in load requirement for resources and consequently,to problems such as low efficien-cy,resource scarcity,and partial system failures.Under certain conditions global catastrophe on the scale of the whole system can occur through the dynamical pro-cess of cascading failures.The primary results and research innovations of this paper are as follows:Chapter 1:At first,we introduce the history of complex network,and describe the conceptual framework and mathematics formulation.Moreover,we summa-rize the different of complex network models basing on real system and the cor-responding character to the present time.Secondly,with the development of net-work science,because modeling real system is more accurate,here we summarize the different of supply-demand network models and some important conclusions.Thirdly,in previous complex network works,artificial intelligence is rarely com-bined.However,with development of social science and technology,it is possible for artificial intelligence(AI)to penetrate into every aspect of the human society?Here,we given a brief description of the foundation frame on machine learning.Forth,we introduce also the relevant work and research status on complex resource allocation system.Chapter 2:We investigate optimization and resilience of time-varying supply-demand systems by constructing network models of such systems,where resources are transported from the supplier sites to users through various links.Here by op-timization we mean minimization of the maximum load on links,and system re-silience can be characterized using the cascading failure size of users who fail to connect with suppliers.Our findings are:at firstly,optimized systems are more robust since relatively smaller cascading failures occur when triggered by external perturbation to the links;Secondly,a large fraction of links can be free of load if resources are directed to transport through the shortest paths;Redundant links in the performance of the system can help to reroute the traffic but may undesir-ably transmit and enlarge the failure size of the system;Thirdly,the patterns of cascading failures depend strongly upon the capacity of links;In particular,the specific location of the trigger determines the specific route of cascading failure,but has little effect on the final cascading size;In addition,we propose two rep-resentative classes of supply schemes:load driven supply and fix fraction supply.we find that system expansion typically reduces the efficiency,and when the loca-tions of the suppliers are optimized over a long expanding period,fewer suppliers are required.These results hold for heterogeneous networks in general,providing insights into designing optimal and resilient complex supply-demand systems that expand constantly in time.Chapter 3:Is it possible to eliminate herding without any external control?The main point of this paper is to provide an affirmative answer through exploiting artificial intelligence(AI).In particular,we demonstrate that,when agents are em-powered with reinforced learning(e.g.,the popular Q-learning in AI)in that they get familiar with the unknown game environment gradually and attempt to deliver the optimal actions to maximize the payoff,herding can effectively be eliminated.Furthermore,computations reveal the striking phenomenon that,regardless of the initial state,the system evolves persistently and relentlessly toward the optimal state in which all resources are used efficiently.However,the evolution process is not without interruptions:there are large fluctuations that occur but only inter-mittently in time.The statistical distribution of the time between two successive fluctuating events is found to depend on the parity of the evolution,i.e.,whether the number of time steps in between is odd or even.At the same time,we find macro-physical phenomena such as abrupt change of system state and emergence of collective behavior and give the micro-mechanism of system evolution at the in-dividual level.We develop a physical analysis and derive mean-field equations to gain an understanding of these phenomena.As minority game dynamics and the phenomenon of herding are common in social,economic,and political systems,and since AI is becoming increasingly widespread,we expect our AI empowered minority game system to have broad applications.Chapter 4:We investigate the dynamical behaviors of AI system numerical-ly and analytically for different game setting,including homogeneous population of artificial intelligence agents,or combination of different types of agents which mimics the diversified situations.It is found that,through short-term training,the AI system adopting Q-learning algorithm relaxes to the optimal solution of the game.Moreover,one striking phenomena is the transition of interaction mechanis-m from self-organized optimization to game situation in AI system through tuning the fraction of AI agent.Simultaneously,we given the transition critical curve an-alytically from one interaction mechanism(self-organized optimization)to another one(game)in phase space.The adaptability of the AI agents population against the time variable environment is also discussed.We develop a physical analysis and derive mean-field equations to gain an understanding of these phenomena.Our findings from the simplified AI model may give new enlightenment to how would the reconciliation and optimization can be breed in the coming physics mechanism of AI era.Chapter 5:We summarized our above works and discussed the research ori-entation and prospective basing on previous exploration results.