Towards continual learning with the artificial neural twin applied to recycling processes

Mendez R, Maier A, Emmert J (2025)


Publication Type: Conference contribution

Publication year: 2025

Publisher: KIT Scientific Publishing

Pages Range: 295-307

Conference Proceedings Title: Optical Characterization of Materials

Event location: Karlsruhe, DEU

ISBN: 9783731514084

Abstract

With the increasing use of AI models in industrial processes, comes the need for more training data. Further, material packaging industry is an ever evolving sector, which makes the task of the AI models used by post-consumer package sorting facilities equally dynamic, rapidly outdating the training data, and requiring the generation of new expensive datasets. We propose to apply continual learning in combination with the Artificial Neural Twin (ANT), to continually train models without generating new data manually. We initially train with a small dataset, then, apply Orthogonal Weight Modification with training stimuli from quality control measurements collected by the ANT, and poof through experiments that this can replace the expensive process of dataset generation.

Involved external institutions

How to cite

APA:

Mendez, R., Maier, A., & Emmert, J. (2025). Towards continual learning with the artificial neural twin applied to recycling processes. In Jürgen Beyerer, Thomas Längle, Jürgen Beyerer, Michael Heizmann (Eds.), Optical Characterization of Materials (pp. 295-307). Karlsruhe, DEU: KIT Scientific Publishing.

MLA:

Mendez, Ronald, Andreas Maier, and Johannes Emmert. "Towards continual learning with the artificial neural twin applied to recycling processes." Proceedings of the 7th International Conference on Optical Characterization of Materials, OCM 2025, Karlsruhe, DEU Ed. Jürgen Beyerer, Thomas Längle, Jürgen Beyerer, Michael Heizmann, KIT Scientific Publishing, 2025. 295-307.

BibTeX: Download