Neural Architecture Search for Extreme Multi-label Text Classification
| Autor: | Rohit Babbar, Loïc Pauletto, Nicolas Winckler, Massih-Reza Amini |
|---|---|
| Přispěvatelé: | Université Grenoble Alpes (UGA), Atos, Laboratoire d'Informatique de Grenoble (LIG), Centre National de la Recherche Scientifique (CNRS)-Université Grenoble Alpes (UGA)-Institut polytechnique de Grenoble - Grenoble Institute of Technology (Grenoble INP ), Aalto University, ANR-19-P3IA-0003,MIAI,MIAI @ Grenoble Alpes(2019) |
| Jazyk: | angličtina |
| Rok vydání: | 2020 |
| Předmět: |
Computer science
business.industry Evolutionary algorithm Image processing 02 engineering and technology Machine learning computer.software_genre Extreme classification Variety (cybernetics) Evolution algorithms Machine Learning Task (computing) [INFO.INFO-LG]Computer Science [cs]/Machine Learning [cs.LG] 020204 information systems 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing State (computer science) Artificial intelligence Architecture business computer Scope (computer science) Neural architecture search |
| Zdroj: | Neural Information Processing 27th International Conference, ICONIP 2020, Bangkok, Thailand, November 23–27, 2020, Proceedings Neural Information Processing 27th International Conference, ICONIP 2020, Bangkok, Thailand, November 23–27, 2020, Proceedings, pp.282-293, 2020, ⟨10.1007/978-3-030-63836-8_24⟩ Neural Information Processing ISBN: 9783030638351 ICONIP (3) |
| DOI: | 10.1007/978-3-030-63836-8_24⟩ |
| Popis: | Extreme classification and Neural Architecture Search (NAS) are research topics which have recently gained a lot of interest. While the former has been mainly motivated and applied in e-commerce and Natural Language Processing (NLP) applications, the NAS approach has been applied to a small variety of tasks, mainly in image processing. In this study, we extend the scope of NAS to the task of extreme multilabel classification (XMC). We propose a neuro-evolution approach, which was found to be the most suitable for a variety of tasks. Our NAS method automatically finds architectures that give competitive results with respect to the state of the art (and superior to other methods) with faster convergence. In addition, we perform analysis of the weights of the architecture blocks to provide insight into the importance of different operations that have been selected by the method. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|