magistrsko delo
Timen Stepišnik Perdih (Author), Marko Robnik Šikonja (Mentor), Senja Pollak (Co-mentor)

Abstract

S porastom uporabe modelov strojnega učenja in pospešenim razvojem kompleksnih modelov se pojavlja tudi potreba po učinkovitih rešitvah razložljive umetne inteligence, ki poskuša približati delovanje modelov človeškemu razumevanju. V tem delu predstavimo novo metodologijo razlaganja modelov, ki povezuje že uveljavljene metode luščenja najpomembnejših značilk z metodami gručenja in novim pristopom posploševanja razlag na podlagi strojno berljivega predznanja. Za demonstracijo delovanja zasnovane metode ustvarimo umetno domeno in jo ovrednotimo na dveh različnih realnih javno dostopnih podatkovnih množicah, s čimer prikažemo uporabnost metode in kompatibilnost z različnimi tipi podatkov. Za ocenjevanje kakovosti razlag pridobimo mnenje domenskih strokovnjakov. Posplošitve so ocenjene kot smiselne, vendar za potrebe strokovnjakov presplošne, zato predlagamo možnosti uporabe za druge skupine uporabnikov. Metodo implementiramo v obliki javno dostopne knjižnice Python.

Keywords

razložljiva umetna inteligenca;razlage modelov;posploševanje razlag;gručenje;procesiranje naravnega jezika;magisteriji;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FRI - Faculty of Computer and Information Science
Publisher: [T. Stepišnik Perdih]
UDC: 004.85(043.2)
COBISS: 210383619 Link will open in a new window
Views: 94
Downloads: 108
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: Explaining machine learning models using background knowledge generalization
Secondary abstract: With the growing use of machine learning models and the accelerated development of complex models, there is a need for efficient solutions in the field of explainable artificial intelligence. In this work, we propose a new method of explaining models that couples state-of-the-art "most important" feature extraction methods with clustering methods and a new approach to feature generalization using machine-readable domain knowledge. We demonstrate the methodology on an artificially created domain and two different real domains. We evaluate the quality of explanations with the help of domain experts. Generalizations are evaluated as sensible but too general for researchers. We propose potential use cases for other groups of users. The implemented method is created as a publicly available Python library.
Secondary keywords: explainable AI;model explanations;explanation generalization;machine learning;clustering;natural language processi;computer science;master's degree;Umetna inteligenca;Računalništvo;Univerzitetna in visokošolska dela;
Type (COBISS): Master's thesis/paper
Study programme: 1000471
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages: 1 spletni vir (1 datoteka PDF (69 str.))
ID: 25041095