Cognitive Networks: Towards Self-Aware Networks

9.3. Distributed Learning and Reasoning within an MAS Framework

Having established the framework within which we will treat learning and reasoning for a cognitive network, we turn our attention to methods that allow agents to learn and reason and the factors involved in selecting a method. The number of methods that we could discuss is much larger than this chapter can accommodate, so we have selected methods for which we can provide some motivation for use in cognitive network development.

9.3.1. Elements of Distributed Reasoning

The primary objective of reasoning within a cognitive network is to select an appropriate set of actions in response to perceived network conditions. This selection process ideally incorporates historical knowledge available in the knowledge base (often referred to as short-term and long-term memories) as well as current observations of the network's state.

Often reasoning is categorized as either inductive or deductive. Inductive reasoning forms hypotheses that seem likely based on detected patterns whereas deductive reasoning forgoes hypotheses and only draws conclusions based on logical connections. Due to the size of the cognitive network state space, which grows combinatorially with the number of network nodes, the cognitive process must be capable of working with partial state information. Since the cognitive process always sees a limited view of the network state, it is difficult to draw certain conclusions as required by deductive reasoning. ...