PIVOT: Prompting for Video Continual Learning

Simple item page
dc.catalogadorgrr
dc.contributor.authorVilla Ojeda, Andres Felipe
dc.contributor.authorAlcazar, Juan Leon
dc.contributor.authorAlfarra, Motasem
dc.contributor.authorAlhamoud, Kumail
dc.contributor.authorHurtado, Julio
dc.contributor.authorHeilbron, Fabian Caba
dc.contributor.authorSoto, Alvaro
dc.contributor.authorGhanem, Bernard
dc.date.accessioned2024-02-29T21:18:06Z
dc.date.available2024-02-29T21:18:06Z
dc.date.issued2023
dc.description.abstractModern machine learning pipelines are limited due to data availability, storage quotas, privacy regulations, and expensive annotation processes. These constraints make it difficult or impossible to train and update large-scale models on such dynamic annotated sets. Continual learning directly approaches this problem, with the ultimate goal of devising methods where a deep neural network effectively learns relevant patterns for new (unseen) classes, without significantly altering its performance on previously learned ones. In this paper, we address the problem of continual learning for video data. We introduce PIVOT, a novel method that leverages extensive knowledge in pre-trained models from the image domain, thereby reducing the number of trainable parameters and the associated forgetting. Unlike previous methods, ours is the first approach that effectively uses prompting mechanisms for continual learning without any in-domain pre-training. Our experiments show that PIVOT improves state-of-the-art methods by a significant 27% on the 20-task ActivityNet setup.
dc.fechaingreso.objetodigital2024-02-29
dc.fuente.origenWOS
dc.identifier.doi10.1109/CVPR52729.2023.02319
dc.identifier.eisbn979-8-3503-0129-8
dc.identifier.issn1063-6919
dc.identifier.urihttps://doi.org/10.1109/CVPR52729.2023.02319
dc.identifier.urihttps://repositorio.uc.cl/handle/11534/81424
dc.identifier.wosidWOS:001062531308053
dc.information.autorucEscuela de Ingeniería; Villa Ojeda, Andres Felipe; S/I; 1092267
dc.language.isoen
dc.nota.accesoContenido parcial
dc.pagina.final24223
dc.pagina.inicio24214
dc.publisherIEEE Computer Soc.
dc.relation.ispartofIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2023, Vancouver, Canadá)
dc.revistaIEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)
dc.rightsacceso restringido
dc.subject.ods03 Good Health and Well-being
dc.subject.odspa03 Salud y bienestar
dc.titlePIVOT: Prompting for Video Continual Learning
dc.typecomunicación de congreso
sipa.codpersvinculados1092267
sipa.trazabilidadWOS;2023-11-25
Files
Collections
Artículos de conferencia