Institutional Repository of Vilnius University
Mokymosi iš nedaug duomenų metodų tyrimas ir jų taikymas marketingo vaizdams vertinti /
Abstract
dc:descriptionThis bachelor thesis explores few-shot learning (FSL) methods and their application in predicting the performance of marketing images. The aim is to evaluate whether deep learning techniques, such as transfer learning, prototypical networks, and MAML, can be effective under data-scarce conditions. The study utilizes pretrained architectures (ResNet18, ResNet50, EfficientNetV2, DeiT) with frozen layers and partial finetuning. Experiments showed that standard approaches failed to produce reliable predictions of CTR or click count. The coefficient of determination R2 was often negative. Better results were achieved using prototypical networks, which managed to separate classes even with a highly imbalanced and limited dataset. The MAML model showed high metric volatility, with only a short period of acceptable performance, indicating that it requires further tuning and stability improvements.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Skripkiūnas, Benas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD239032315&prefLang=en_US