what is adaptive clustering?

Answer 1

A clustering algorithm fnds groups of objects in a pre-defned attribute space. Since the objects have no known
group membership, clustering is an unsupervised learning technique. A clustering algorithm optimizes some explicit
or implicit criterion inherent to the data. The squared summed error, for example, is a criterion whose optimization
is the primary concern of the k-medioids [12] clustering algorithm. A large amount of work (Section II) details
the limitations of these algorithms: the main criticism is that these criteria are excessively simplistic and do not
accurately capture the user's conception of the true data essence. The purpose of this paper is to introduce adaptive
clustering as a solution to this problem.
Adaptive clustering applies when one seeks a clustering algorithm that can meet objectives beyond the simple
least-squares criteria. Adaptive clustering makes the clustering process sensitive to external rewards though the use
of reinforcement learning [11]. Moreover, it supports memorizing clusterings that worked well in a given context,
supporting the reuse of “good” clusterings that were found in the past. This approach differs from that of both
unsupervised and supervised learning. Although the objects have no class label, adaptive clustering seeks to nd
good clusters based on feedback with respect to previously found clusterings.
Adaptive clustering applies when external feedback exists for a clustering task. Applications include interactive
clustering for summary generation, clustering for streaming data and multi-agent coordination and control. In
interactive summary generation, such as spatial clustering, adaptive clustering improves the clustering in response
to user-provided feedback regarding the quality of the obtained clusters. When mining data streams, adaptive
clustering retains a model of the clustering that it can reuse as new objects arrive without re-clustering all the
data. For multi-agent control, a clustering represents the properties of groups of agents, and the adaptive clustering
algorithm uses external feedback to modify the clusters and thereby the agents toward some externally specified
goal. A relatively simple but illustrative application is in classication where adaptive clustering is used to enhance
an instance-based classifier.