---

Presented By:
David Legge; Peter Baxendale
Centre for Telecommunication Networks,

---

Abstract;
An Ant-Based Routing System has been constructed and it has been shown to find optimum routing strategies un- der static conditions. With the introduction of dynamic traffic situations onto the network, the ants do not always display optimum behaviour. It has been shown that parameters cannot be set to a single value to satisfy both steady state and transitory conditions optimally. It is therefore proposed to use an Agent on each node to adapt to changing conditions on the network. This can be achieved by manipulating the parameters with which the ants operate. Since the agent is classifying different states and the associated actions best suited to them, methods of model-free Rein- forcement Learning, such as Temporal Difference and Q-Learning are discussed, along with the problems of a con- tinuous domain.

read full report
http://citeseerx.ist.psu.edu/viewdoc/dow...1&type=pdf