In the field of Industry 4.0, where Artificial Intelligence has an important role in business, there is a nontrivial challenge to face in which the algorithm's performance degrades over time; this is known as Catastrophic Forgetting. In the Life Long learning scenario in an industrial environment, the aim is to build a machine learning algorithm which is capable of learning as long as new data from new machines arrive and in contemporary is able to remember the patterns learned previously. Continual Learning aims to learn from a stream of tasks, being able to remember at the same time both new and old tasks. While many approaches for single-class classification were proposed, multi-label classification in the continual scenario remains a challenging problem. In this work, multi-label classification in the Domain Incremental Learning scenario has been studied on a real world Alarm Forecasting problem from the packaging industry.
Nel campo dell'Industria 4.0, in cui l'Intelligenza Artificiale ha un ruolo importante nel business, c'è una sfida non banale da affrontare in cui le prestazioni dell'algoritmo si degradano nel tempo; questo fenomeno è noto come Catastrophic Forgetting. Nello scenario dell'apprendimento continuo in un ambiente industriale, l'obiettivo è quello di costruire un algoritmo di apprendimento automatico che sia in grado di imparare finché arrivano nuovi dati da nuove macchine e che contemporaneamente sia in grado di ricordare i modelli appresi in precedenza. Il Continual Learning mira ad apprendere da un flusso di compiti, essendo in grado di ricordare allo stesso tempo sia i nuovi che i vecchi compiti. Mentre sono stati proposti molti approcci per la classificazione a classe singola, la classificazione multi-label nello scenario continuo rimane un problema impegnativo. In questo lavoro, la classificazione multi-label nello scenario dell'apprendimento incrementale del dominio è stata studiata su un problema reale di previsione degli allarmi nell'industria del packaging.
A continual learning framework to scale deep learning approaches for packaging equipment monitoring
DERONJIC, DENIS
2021/2022
Abstract
In the field of Industry 4.0, where Artificial Intelligence has an important role in business, there is a nontrivial challenge to face in which the algorithm's performance degrades over time; this is known as Catastrophic Forgetting. In the Life Long learning scenario in an industrial environment, the aim is to build a machine learning algorithm which is capable of learning as long as new data from new machines arrive and in contemporary is able to remember the patterns learned previously. Continual Learning aims to learn from a stream of tasks, being able to remember at the same time both new and old tasks. While many approaches for single-class classification were proposed, multi-label classification in the continual scenario remains a challenging problem. In this work, multi-label classification in the Domain Incremental Learning scenario has been studied on a real world Alarm Forecasting problem from the packaging industry.File | Dimensione | Formato | |
---|---|---|---|
Deronjic_Denis.pdf
accesso riservato
Dimensione
12.46 MB
Formato
Adobe PDF
|
12.46 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/30730