The key issues in network management are the representation and sharin
g of management information and the automatic management mechanisms ba
sed on the underlying information infrastructure. In this paper, we pr
opose a framework, which operates on the standard MIB's and CMIP, for
the network management system with learning and inference as its manag
ement engines. In addition to the general domain knowledge, patterns r
elated to the managed network are learned to enhance the understanding
of the network and refine the knowledge base. Facts in object-oriente
d databases or queries from management applications trigger the infere
nce process on logical rules which are either prespecified knowledge o
r learned network patterns. Forward inference drives prediction and co
ntrol, while backward inference directs diagnosis and supports view ab
straction. A case study on ATM network topolgy tuning is presented.