Stabilization of Dataset Matrix Form for Classification Dataset Generation and Algorithm Selection

• Ilya Sahipov ^[1] ; Alexey Zabashta ^[1] ; Andrey Filchenkov ^[1]
  1. [1] Saint Petersburg State University of Information Technologies, Mechanics and Optics

     Saint Petersburg State University of Information Technologies, Mechanics and Optics

     Rusia

     
• Localización: Intelligent Data Engineering and Automated Learning – IDEAL 2020. 21st International Conference: Guimarães, Portugal; November 4–6, 2020. Proceedings / Cesar Analide (ed. lit.), Paulo Novais (ed. lit.), David Camacho Fernández (ed. lit.), Hujun Yin (ed. lit.), Vol. 2, 2020 (Part II), ISBN 978-3-030-62365-4, págs. 66-75
• Idioma: inglés
• Enlaces
  • Texto completo
• Resumen

  • Datasets for the classification task are usually encoded by a matrix of numbers, the order of rows and columns does not matter. Swapping any two objects or features in it does not change the hidden target function and performance of the machine learning algorithms train of the dataset. However, in the dataset generation problem solution such symmetry is an obstacle. In this paper, we study several methods of the inverse transformation of classification dataset aiming to break the symmetry. We experimented with it in the meta-learning problems of datasets generation and algorithm selection which were solved by conditional generative adversarial nets with convolutional networks.