Export Reviews, Discussions, Author Feedback and Meta-Reviews

Resource constraint prediction has to select the smallest/ cheapest set of sensors in order to make accurate decisions. A new approach to this problem is proposed in this paper where the authors basically define a deterministic Markov decision process and learn its policy applying cost sensitive learning in every state of the MDP. Given a new testing example to classify, the policy has to decide whether a new sensor should be queried (if yes, which sensor), or whether the prediction should be made with the current set of sensors measured. In line 44, the authors mention the "expected budget constraint". Please be clear about that because your paper is not budgeted learning; budgeted learning is more challenging, please see, e.g.