Institutional Repository of Vilnius University
Didelių kalbos modelių taikymas sveikatingumo rekomendacijoms /
Abstract
dc:descriptionArtificial intelligence has long been used for medical purposes. The usage spans from notifying medical staff about irregularities in data derrived from heart monitors to medical image and person's medical history analysis, which helps medical staff to detect diseases such as cancer earlier, faster and easier. With the rise of artificial intelligence technologies and the rise of increasingly sophisticated wearable device use, the question comes up: how can artificial intelligence help people seeking medical recommendations? Can easily accessible devices such as smartwatches provide enough information to provide complex recommendations based on rich descriptions? This paper will attempt to compare various deep learning models for creating medical descriptions using physiological data from publicly available wearable devices, emphasizing the current ability of large language models in creating these descriptions. First, optimal architectures of convolutional neural networks and recurrent neural networks will be proposed. Then, capability to fine-tune a large language model designed for sequence classification will be analysed, additionally a regular large language model will be analysed to find out the capability of predicting continuous values which can be used for medical recommendations. The study found that when analyzing physiological data with limited data, large language models perform considerably worse than convolutional and recurrent neural networks, even when their architecture is modified to perform specific tasks such as classification.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kakčiukas, Domantas,
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:ELABAETD239030896&prefLang=en_US