On demand and autonomic computing will benefit from policy-based management systems which are responsive to new and ambiguous situations and learn from them. In a typical data center, there are thousands of different events reporting system faults, status, and performance information. Their occurrences are unpredictable. In addition, new events and conditions can occur as operating environment changes. Traditional approaches of authoring policies and techniques of implementing policy-based management systems, such as relying entirely on static authoring of simple "if [condition] then [actions]" rules, become insufficient. Hence, new approaches, such as goal policy, utility function etc., to the design and implementation of policybased management systems have emerged. However, none of these approaches provides a systematic way to enable policies in a policybased management system to be responsive to new and ambiguous situations.

In this paper, we describe a novel method by which policies can be selected or created automatically based on events observed and knowledge learned. This new approach treats the observed event-policy relationship represented by an event-policy matrix as a statistical problem. Using Singular Value Decomposition (SVD) technique, implicit higher order correlations among policies and their associated events are used to estimate the selection or creation of recommended policies based on events found in the observed event set. Initial results have indicated that this approach to policy-based management system is very promising.