izpis_h1_title_alt

Argumentirano strojno učenje za klasifikacijo slik
ID Križnar, Anton (Author), ID Žabkar, Jure (Mentor) More about this mentor... This link opens in a new window

.pdfPDF - Presentation file, Download (948,84 KB)
MD5: CE94E7DA7FBA23BC032AEA522700A30D

Abstract
Argumentirano strojno učenje (ABML) (angl. argument based machine learning) je razširitev metod strojnega učenja, kjer s podajanjem argumentov pri napačno klasificiranih primerih dodajamo znanje domenskega strokovnjaka. ABML običajno razširja metode, ki se učijo klasifikacijska pravila. Sodobne metode za nalogo klasifikacije slik uporabljajo konvolucijske nevronske mreže (CNN) (angl. convolutional neural network), ki jih nadgrajujemo s principi ABML. Glavni izziv je vključevanje principov ABML v model globokega strojnega učenja, saj so ti modeli težko razložljivi. Za razlago modelov CNN je primerna tehnika pridobivanja aktivacijske slike razreda (CAM) (ang. class activation mapping). Težko razložljivo pa je tudi domensko znanje za klasifikacijo slik. Argumente domenskega strokovnjaka lahko zajamemo kot točke na sliki. S primerjavo slik CAM in formalnih argumentov lahko v model vključimo domensko znanje. Oznake za konstrukcijo argumentov so bile zajete z razvitim programom in računalniško miško. Analiza delovanja razvitega modela je narejena na umetnih slikah likov. Pokazali smo, da z uporabo ABML za klasifikacijo slik lahko izboljšamo uspešnost modela tudi z različnimi vrstami argumentov.

Language:Slovenian
Keywords:argumentirano strojno učenje, konvolucijske nevronske mreže, klasifikacija slik
Work type:Bachelor thesis/paper
Organization:FRI - Faculty of Computer and Information Science
Year:2024
PID:20.500.12556/RUL-161309 This link opens in a new window
Publication date in RUL:09.09.2024
Views:26
Downloads:4
Metadata:XML RDF-CHPDL DC-XML DC-RDF
:
Copy citation
Share:Bookmark and Share

Secondary language

Language:English
Title:Argument Based Machine Learning for Image Classification
Abstract:
Argument based machine learning (ABML) is an extension of machine learning methods in which domain experts provide additional knowledge by giving arguments for incorrectly classified cases. ABML typically extends methods that learn classification rules. The most advanced techniques for image classification tasks use convolutional neural networks (CNNs), which we are enhancing with ABML principles. The main challenge is incorporating ABML principles into a deep learning model, as these models are often difficult to interpret. Moreover, domain knowledge for image classification is challenging to interpret. One suitable technique for explaining CNN model decisions is class activation mapping (CAM). Domain expert arguments can be captured using a computer mouse and then incorporated into the model. By comparing CAMs with formal arguments, domain knowledge can be integrated into the model. Labels for constructing arguments were captured using a developed program and a computer mouse. The analysis of the model's performance was conducted using synthetic images of geometric shapes. We showed an improvement in performance for the ABML model for image classification even with different kinds of arguments.

Keywords:argument based machine learning, convolutional neural networks, image classification

Similar documents

Similar works from RUL:
Similar works from other Slovenian collections:

Back