"Explaing the impact of data augmentation with Quantum Harmonic Analysis (QHA) Tools"Dörfler, MonikaData, which we strive to use, interpret, classify, usually undergoes some kind of representation; that is, each data point is written in some dictionary of building blocks by means of coefficients. The building blocks may or may not be derived from the data themseleves, they may or may not be physically meaningful and the may or may not depend on feq fixed parameters. In any case, we are often interested in using augmentation, which can happen along the dimensions given by the representation coefficients. Augmentation is a well-established and successful tool in many data science applications, however, often heuristic and not well-studied mathematically. We give new insights on the impact of augmentation, which we interpret as a generalized convolution in a QHA context. Using surprsing insights from QHA, we provide a possible explanation of the benefits of augmentation. (Joint Work with Henry McNulty) |
« back