Reasoning Polus Framework Learning

Conjunction of specifications and learning

The attributes of the action object are derived using a combination of specifications and learning. In the current implementation of Polus, the rule-of-thumb specifications forms the static part of the knowledge base i.e., the system does not discover additional relationships. The information associated with the learning algorithms forms the dynamic part of the knowledge base as this information is constantly updated by monitoring the system and refining the interpolation functions. The reasoning engine accesses the information in the knowledge base. The access is a query of the form Does Action X affect throughput and if yes, by how much?. The query is translated into a sub-query which is first handled by the specification sub-part of the knowledge base that looks-up to see if there is a relation defined between Action X and throughput. If yes, it generates another query for the learning sub-part that contains the interpolation function for action X, current-state and throughput. Figure 7 illustrates this in the context of prefetching.

Figure: Conjunction of knowledge with the specifications and learning module

Reasoning Polus Framework Learning