Implementation and Analysis of Contextual Neural Networks in H2O Framework

• Published in: Intelligent Information and Database Systems Vol. 11432; pp. 429 - 440
• Main Authors: Wołk, Krzysztof, Burnell, Erik
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2019; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Aggregation functions; GBP; H2O.ai; Scan-paths; Sigma-if
• ISBN: 9783030148010; 3030148017
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-14802-7_37

Contextual neural networks utilizing conditional multi-step aggregation functions have many useful properties. For example, their ability to decrease the activity between internal neuron connections may decrease computational costs, whereas their built-in automatic selection of attributes required for proper classification can simplify problem setup. The research of contextual neural networks was motivated by a limited number of satisfactory machine learning solutions providing these features. An implementation of the CxNN model in the H2O.ai machine learning framework was also developed to validate the method. In this article we explain relevant terms and the implementation of contextual neural networks as well as conditional multi-step aggregation functions. To validate the solution, experiments and their results are presented for selected UCI benchmarks and Cancer Gene Expression Microarray data.