Robust multiple-instance learning ensembles using random subspace instance selection

摘要: Many real-world pattern recognition problems can be modeled using multiple-instance learning (MIL), where instances are grouped into bags, and each bag is assigned a label. State-of-the-art MIL methods provide high level of performance when strong assumptions made regarding the underlying data distributions, proportion positive to negative in bags. In this paper, new method called Random Subspace Instance Selection (RSIS) proposed for robust design ensembles without any prior on structure First, instance selection probabilities computed based training clustered random subspaces. A pool classifiers then generated subsets created with these probabilities. By RSIS, more many distributions noise, not adversely affected by bags because repeatedly selected probabilistic manner. Moreover, RSIS also allows identification an individual basis, as required practical applications. Results obtained several synthetic databases show robustness designed over range witness rates, noisy features compared reference literature. HighlightsA method, Selection, ensembles.The yields that variations rate, noise.The state-of-the-art results benchmark sets.