Analysis of Network Evolution and Learning
Abstract
A (large) network science literature studies networks that have already formed. A (smaller) microeconomics literature studies the formation of networks – but makes strong assumptions (e.g., homogeneous agents/entities, complete information about other agents) that are far from being met in reality. Neither the network science literature nor the microeconomics literature take into account that agents behave strategically in deciding what information to consume, produce and share (in addition to deciding what links to form/maintain/break) and that agents begin with incomplete information (about others). As a result, neither network science nor microeconomics provide a useful methodology for understanding, predicting and guiding the formation (and evolution) of real networks and the consequences of network formation. The overarching goal of this project is to develop such a methodology. The research will take into account that agents behave strategically and that they begin with incomplete information about each other and thus, must learn over time what information to produce and consume, and which connections to form and maintain and which to break. The approach is through social norms; the previous work demonstrated the usefulness of social norms to promote/regulate anonymous one-time interactions (such as exchange of services) on the web. A key goal of this project is to develop a systematic methodology for understanding how social norms affect (positively and negatively) the formation and functioning of informational networks as well as a theory and methods to develop social norms that lead agents to form networks with desired properties. Objective Learning and network formation are intertwined and this co-evolution is a central theme of the proposed research. The first objective is to determine how the process of learning about others directs the formation/evolution of the network and how in turn the formation/evolution of the network directs the process of learning. We will determine which network topologies emerge and persist and whether these network topologies are desirable. We will investigate the way in which learning (about others) interacts with the evolution of the network. Particular aspects that are important are the tradeoffs between direct and indirect connections, the complementarity of information and how all of this affects the evolution and structure of the network. The second objective is to determine the effects of strategic choices of agents and how social norms influence these choices and their consequences – both the choices of what links to form/break and also what information to produce and where to disseminate it and what information to acquire and from where. The PI will determine how social norms influence the topologies of networks that emerge and persist and which social norms promote social welfare (through the production and dissemination of the most useful quantity and variety of information) and learning. Approach The PI will model what agents learn about others in terms of reputation. Reputation evolves because knowledge arrives gradually and is noisy, and also because the network evolves as individuals form and break links. The PI will model the information process in terms of Brownian motion. This will capture the idea that information about agents is disseminated gradually and is intrinsically noisy. When agents have more links, more information is disseminated and learning is more rapid. Rapid learning is both good and bad: it is good because low quality or malicious agents can be quickly discovered and ostracized; it is bad because high quality agents may produce an unlucky string of bad signals and so be mistakenly thought to be bad and treated as such. The PI will address this tension; a specific approach is to determine how a network populated by agents with specific characteristics will evolve and whether the network that emerges will be optimal for these agents.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141512038
Entities
People
- Mihaela Van Der Schaar
Organizations
- Office of Naval Research
- United States Navy
- University of California, Los Angeles