Karen Seidel (UP)
Abstract: In machine learning, algorithms generalize from available training data to unseen situations. The engineering practices used in the respective technologies are far from understood. Research in theoretical machine learning analyses concrete mathematical models for this complex subject. We introduce a class of sequential learning algorithms suitable for improving Data Sampling. Bandit Algorithms have been developed for resolving the exploration versus exploitation trade-off when navigating with incomplete information in an uncertain environment. We consider discrete and continuous versions in the context of dependencies in the sampling process.
Zoom-link on request
