Adaptive Clustering: Better Representatives with Reinforcement Learning

Adaptive clustering uses reinforcement learning to learn the reward values of successive data clusterings. Adaptive clustering applies when external feedback exists for a clustering task. It supports the reuse of clusterings by memorizing what worked well in a previous context. It explores multiple paths in a reinforcement learning environment when the goal is to find better cluster representatives based on arbitrary environmental feedback. Our experiments apply adaptive clustering to instance-based learning relying on a distance function modification approach. The results show that adaptive clustering can find better representatives, if compared with traditional instance-based learning, such as k-nearest neighbor classifiers. Moreover, we introduce as a by-product a new instance-based learning technique that classifies examples by solely using cluster representatives; the technique shows high promise in our experimental evaluation.