Supervised Sequential Classification Under Budget Constraints

摘要: In this paper we develop a framework for sequential decision making under budget constraints multi-class classification. many classification systems, such as medical diagnosis and homeland security, decisions are often warranted. For each instance, sensor is first chosen acquiring measurements then based on the available information one decides (rejects) to seek more from new sensor/modality or terminate by classifying example information. Different sensors have varying costs acquisition, these account delay, throughput monetary value. Consequently, methods maximizing performance of system subject constraints. We formulate multi-stage empirical risk objective learn functions training data. show that reject at stage can be posed supervised binary derive bounds VC dimension quantify generalization error. compare our approach alternative strategies several real world datasets.