diplomsko delo
Gal Šubic (Author), Marko Robnik Šikonja (Mentor)

Abstract

Razlaga s protiprimeri je metoda razlage napovednih modelov strojnega učenja. V nalogi opišemo več načinov generiranja protiprimerov pri besedilnih klasifikatorjih, LIME-C, Polyjuice in ChatGPT, ter izpostavimo njihove ključne značilnosti. Uporabimo jih na treh različnih besedilnih podatkovnih množicah. Uporabljene metode in pridobljene protiprimere primerjamo in jih ocenimo po kriterijih za ocenjevanje kakovosti protiprimerov. Ugotovimo, da ni ene same najboljše rešitve in da ima vsak pristop prednosti in slabosti. Kljub temu se izkaže, da so trenutno najsplošnejša in uporabna rešitev protiprimeri, generirani z velikim jezikovnim modelom ChatGPT.

Keywords

razlaga napovednega modela strojnega učenja;protiprimeri;razlaga s protiprimeri;LIME-C;Polyjuice;ChatGPT;univerzitetni študij;diplomske naloge;

Data

Language: Slovenian
Year of publishing:
Typology: 2.11 - Undergraduate Thesis
Organization: UL FRI - Faculty of Computer and Information Science
Publisher: [G. Šubic]
UDC: 004.85:81'322(043.2)
COBISS: 170240003 Link will open in a new window
Views: 75
Downloads: 14
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Counterfactual explanation of text classifiers
Secondary abstract: Counterfactual explanations are used for interpreting predictive machine learning models. We describe three methods for generating counterfactual examples, LIME-C, Polyjuice and ChatGPT, and highlight their key features. We apply them to three different text datasets. We compare the methods used and the obtained counterfactual examples and evaluate them according to the quality criteria of counterfactual examples. We conclude that there is no single best solution and that each approach has advantages and disadvantages. Nevertheless, the most general and useful solution at the moment are the counterfactual examples generated with ChatGPT large language model.
Secondary keywords: machine learning;explaining machine learning prediction model;counterfactual examples;LIME-C;Polyjuice;ChatGPT;computer science;diploma;Strojno učenje;Računalniško jezikoslovje;Računalništvo;Univerzitetna in visokošolska dela;
Type (COBISS): Bachelor thesis/paper
Study programme: 1000468
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages: 43 str.
ID: 20325615