Exploiting micro-clusters to close the loop in data-mining robots for human monitoring

This paper describes our approach to integrating representation, reasoning, learning, and execution in our data-mining robots by exploiting micro-clusters to close the loop of the KDD process model. Based on our several kinds of autonomous mobile robots that monitor humans with Kinect and discover patterns, we are working on designing data-mining robots, each of which makes trials and errors in its data observation, data processing, pattern extraction, and mobile explorations. In other words, the robots continuously refine their goals at the micro-cluster level. We briefly discuss our four research directions, i.e., the balance between the exploitation and the exploration, the use of weak labels, the anytime algorithm, and the countermeasure to the concept drift, and describe potential, promising approaches for some of them.