Generativni duboki model za izvještaje o fiktivnim nogometnim utakmicama

undergraduate thesis

• Title (english): Generative deep learning model for generating reports about fictional football matches
• Author: Luka Hobor
• Mentor: Mario Brčić (mentor)
• Committee member: Mario Brčić (predsjednik povjerenstva)
• Committee member: Mihael Kovač (član povjerenstva)
• Committee member: Kristijan Poje (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2023-07-05, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Generativni modeli umjetne inteligencije su moderni pristup za stvaranje sintetičkih podataka za različite svrhe. U ovom radu opisat ćemo postupak finog podešavanja GPT-2 modela za potrebe generiranja fiktivnih nogometnih sažetaka iz statistika utakmica. Pretpostavljamo da se manji modeli fino podešeni za specifične zadatke mogu dobro nositi s velikim generativnim modelima poput GPT-3 i GPT-4. Nakon provodenog treniranja modela uočavamo da je model naučio stil izražavanja prisutan u skupu podataka za treniranje, ali da semantička smislenost priče nije zadovoljiva. Daljnja unaprjeđenja modela mogla bi se ostvariti povećanjem skupa podataka za treniranje ili korištenjem većeg modela.

Abstract (english)

Generative AI models are a modern approach to creating synthetic data for various purposes. In this paper, we will describe the procedure for fine-tuning the GPT-2 model for the purposes of generating fictitious football summaries from match statistics. We assume that smaller models fine-tuned for specific tasks can cope well with large generative models such as GPT-3 and GPT-4. After training the model, we notice that the model has learned the style of expression present in the training dataset, but the semantic meaningfulness of the story is not satisfactory. Further improvements to the model could be achieved by increasing the training data set or using a larger model.

• Keywords: GPT-2; generativni modeli; fino podešavanje; skup podataka; nogometni sažetci; transformatori; slojevi pažnje
• Keywords (english): GPT-2; generative models; fine-tuning; dataset; football summaries; transformers; attention layers
• Language: croatian
• URN:NBN: urn:nbn:hr:168:154769
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-10-11 09:01:55