University of Twente Student Theses

Login

Learning on a budget : tackling the cold-start problem in active learning for livestock behavior monitoring

Vanwinsen, Wannes (2024) Learning on a budget : tackling the cold-start problem in active learning for livestock behavior monitoring.

[img] PDF
8MB
Abstract:This study explores cold start effects in active learning for livestock behavior monitoring and introduces a novel framework that combines self-supervised contrastive learning with active learning. The framework offers a structured way to select a representative and diverse initial training set for labeling by pre-training a model on a contrastive learning objective. Experiments were conducted on the CVB dataset, a benchmark for cattle behavior classification. Results reveal that active learning strategies with and without self-supervised pre-training are sensitive to the size and diversity of the initial training set. The proposed framework failed to consistently outperform baseline models without pre-training and pre-training was found to have detrimental effects on model performance in most cases. The findings highlight that active learning’s effectiveness depends on careful initial data selection and the selection of different sampling strategies across learning stages, emphasizing that the cold start problem remains a challenging problem in active learning.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science MSc (60300)
Link to this item:https://purl.utwente.nl/essays/104742
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page