Data acquisition for improving machine learning models

The vast advances in Machine Learning over the last ten years have been powered by availability of suitably prepared data for training purposes. future ML-enabled enterprise hinges on data. As such, there is already a vibrant market offering annotation services to tailor sophisticated ML models. In this paper, we present research practical problem obtaining order improve accuracy We consider an environment which consumers query enhance their models and providers who possess make them available first formalize interaction process laying out suitable framework associated parameters exchange. then propose two acquisition strategies that trade-off between exploration during obtain learn about distribution provider's exploitation optimize our inquiries utilizing gained knowledge. strategy, Estimation Allocation, utilize queries estimate utilities various predicates while learning data; proceed allocation stage those learned utility estimates inform decisions. second algorithmic proposal, named Sequential Predicate Selection, utilizes sampling strategy explore data, adaptively investing more resources parts space are statistically promising overall model accuracy. detailed experimental evaluation proposals variety real sets exploring all applicable interest. identify trade-offs highlight relative benefits each algorithm further