Fast, Visual and Interactive Semi-supervised Dimensionality Reduction

Dimitris Spathis, Nikolaos Passalis, Anastasios Tefas; Proceedings of the European Conference on Computer Vision (ECCV) Workshops, 2018, pp. 0-0

Recent advances in machine learning allow us to analyze and describe the content of high-dimensional data ranging from images and video to text and audio data. In order to visualize that data in 2D or 3D, usually Dimensionality Reduction (DR) techniques are employed. Most of these techniques produce static projections without taking into account corrections from humans or other data exploration scenarios. In this work, we propose a novel interactive DR framework that is able to learn the optimal projection by exploiting the user interactions with the projected data. We evaluate the proposed method under a widely used interaction scenario in multidimensional projection literature, i.e., project a subset of the data, rearrange them better in classes, and then project the rest of the dataset, and we show that it greatly outperforms competitive baseline and state-of-the-art techniques, while also being able to readily adapt to the computational requirements of different applications.