Learning in networks

Yu-Han Chang
MIT

A communication network can be thought of as a collection of agents
with a specific task: to route information from senders to receivers
within the network.  Routing algorithms for static networks work
fairly well and are widely deployed.  Recently mobile ad-hoc networks
have been getting more attention in the research community, in part
due to the widespread popularity of 802.11b and Bluetooth.  It has
been shown that mobility within a network can be used to increase
bandwidth and decrease latency.  However, simple algorithms for this
type of movement and routing do not attain the optimal benefit.  Using
techniques from learning, we might hope to attain better solutions for
this complex domain.  We will discuss various applications of
multi-agent learning to the mobile ad-hoc networking domain and
demonstrate a distributed algorithm that learns to cooperatively move
and route information.