Intra-Family Track Discrimination Using Supervised and Semi-Supervised Statistical Learning Methods - MP 121-07

We are proposing to incorporate support vector machines (SVM’s) as well as related, but more analytically tractable methods such as proximal SVM’s, into a multi-hypothesis tracker (MHT) and to use the combined SVM/MHT algorithm to discriminate launch families into intra-family similarity groups. We also propose to consider the use of semi-supervised statistical learning methods [5] since they exploit both labeled and unlabeled data in the design of the discriminator. Semi-supervised methods take advantage of available archived data which, even if it has limitations, may be still useful to include as unlabeled data. The joint use of both labeled and unlabeled data is generally accepted statistical practice and is an active research area in the machine learning community.