THE SAFE USE OF SYNTHETIC DATA IN CLASSIFICATION Program Proposal for Ph.D. in Computer Science Draft for Comments

When is it safe to use synthetic data in supervised classification? Trainable classifier technologies require large representative training sets consisting of samples labeled with their true class. Acquiring such training sets is difficult and costly. One way to alleviate this problem is to enlarge training sets by generating artificial, synthetic samples. Of course this immediately raises many additional questions, perhaps the first being “Why should we trust artificially generated data to be an accurate representative of the real distributions?” Other questions include “When will the use of synthetic data work as well as real data ?” and “Can synthetic data produce better results than training only on real data?”. We distinguish between sample space (the set of real samples), generator space (all samples that can be generated synthetically), and finally, feature space (the set of samples in terms of finite numerical values). In this proposal, we discuss a small experiment, in which we produced the synthetic data in feature space, that is, by varying three of the features (the RG and B components) of the data files.We showed that for some optimal number of seed files, the performance of the synthetic data matched that of the real data. However, there