Korištenje jezičnih modela za otkrivanje anomalija u vremenskim serijama

master's thesis

• Title (english): Using language models for anomaly detection in time series
• Author: Vid Rebernak
• Mentor: Vladimir Čeperić (mentor)
• Committee member: Vladimir Čeperić (predsjednik povjerenstva)
• Committee member: Zoran Kalafatić (član povjerenstva)
• Committee member: Marko Koričić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2025-02-25, Croatia

Abstract

U ovom diplomskom radu istražuju se i analiziraju mogućnosti korištenja velikih jezičnih modela (eng. Large Language Models - LLMs) za otkrivanje anomalija u vremenskim serijama. Cilj rada je procijeniti učinkovitost jezičnih modela u prepoznavanju neuobičajenih obrazaca u podacima te analizirati kako se ti modeli ponašaju u usporedbi s tradicionalnim tehnikama otkrivanja anomalija. Teoretski dio rada uključuje pregled relevantne literature o vremenskim serijama, otkrivanju anomalija i korištenju LLM-ova u analitičkim zadacima. U praktičnom dijelu rada implementirana je prilagodba četiri LLM modela zadatku otkrivanja anomalija u vremenskim serijama. Eksperimentalni rezultati prikazuju usporedbu točnosti i računalnih zahtjeva između LLM-ova i drugih uobičajenih metoda za otkrivanje anomalija.

Abstract (english)

This thesis explores and analyzes the potential of Large Language Models (LLMs) for anomaly detection in time series. The goal of the research is to assess the effectiveness of language models in identifying unusual patterns in data and to compare their performance with traditional anomaly detection techniques. The theoretical part of the thesis includes a review of relevant literature on time series, anomaly detection, and the use of LLMs in analytical tasks. In the practical part, four LLM models were adapted for the task of anomaly detection in time series. The experimental results present a comparison of accuracy and computational requirements between LLMs and other commonly used anomaly detection methods.

• Keywords: veliki jezični modeli; otkrivanje anomalija; vremenske serije; strojno učenje; prirodna obrada jezika; analiza podataka; jezično modeliranje; tradicionalne metode otkrivanja anomalija; eksperimentalna evaluacija
• Keywords (english): large language models; anomaly detection; time series; machine learning; natural language processing; data analysis; language modeling; traditional methods for anomaly detection; experimental evaluation
• Language: croatian
• URN:NBN: urn:nbn:hr:168:533726
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2025-03-12 11:17:34