News of the Institute of Computer Science Polish Academy of Sciences

New classification methods for data with incomplete observability

The IPI PAN's Statistical Analysis and Modeling Group achieved significant results regarding two new machine learning methods for positive-unlabeld data in an important case when their availability depends on the characteristics of the examined entities. The first proposed method examined a new parametric model for generating this type of data and an iterative procedure for estimating its parameters. The second method is based on the concept of variational autoencoders combined with outlier detection techniques. In addition to the theoretical justification of the methods, a significantly better quality of the resulting classifiers for tabular and image data was experimentally demonstrated.