Practical Estimation of Ensemble Accuracy

Ensemble learning combines several individual models to obtain better generalization performance. In this work we present a practical method for estimating the joint power of several classifiers which differs from existing approaches by {\em not relying on labels}, hence enabling the work in unsupervised setting of huge datasets. It differs from existing methods which define a "diversity measure". The heart of the method is a combinatorial bound on the number of mistakes the ensemble is likely to make. The bound can be efficiently approximated in time linear in the number of samples. Thus allowing an efficient search for a combination of classifiers that are likely to produce higher joint accuracy. Moreover, having the bound applicable to unlabeled data makes it both accurate and practical in modern setting of unsupervised learning. We demonstrate the method on popular large-scale face recognition datasets which provide a useful playground for fine-grain classification tasks using noisy data over many classes. The proposed framework fits neatly in trending practices of unsupervised learning. It is a measure of the inherent independence of a set of classifiers not relying on extra information such as another classifier or labeled data.