Primjena evaluacijskih tehnika na primjeru konvolucijske neuronske mreže

undergraduate thesis

• Title (english): Application of evaluation techniques on the example of a convolutional neural network
• Author: Lora Lovreković
• Mentor: Tomislav Stipančić (mentor)
• Granter: University of Zagreb; Faculty of Mechanical Engineering and Naval Architecture; Zagreb
• Defense date and country: 2021-09-29, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing; Artificial Intelligence

Abstract

U ovome je radu trenirana konvolucijska neuronska mreža na CIFAR-10 bazi slika i evaluiran je rad mreže primjenom evaluacijskih funkcija. U teorijskom dijelu objašnjen je princip rada umjetnih neuronskih mreža te arhitektura i funkcija konvolucijskih neuronskih mreža. Uveden je pojam matrica konfuzije kako bi se mogle objasniti evaluacijske funkcije točnost, preciznost i opoziv, a opisana je i struktura CIFAR-10 skupa podataka.
Kod je pisan u programskom jeziku Python (verzija 3.7), a za izradu koda korištene su knjižnice tensorflow, matplotlib, numpy i scikit-learn. Uspoređen je rad mreže na serijama za trening i na testnoj seriji te je model mreže evaluiran i koristeći predviđanja. Predložena su i moguća poboljšanja mreže.

Abstract (english)

In this thesis convolutional naural network is trained on CIFAR-10 database and it's work is evaluated using evaluation functions. Theoretical part of the thesis explains artificial neural networks work principles along with convolutional neural networks architecture and funcion. The concept of confusion matrix is introduced to help with explanation of evaluation functions accuracy, precision and recall, and the structure of CIFAR-10 database is described as well.
The code is written in Python programming language (version 3.7), and libraries tensorflow, matplotlib, numpy and scikit-learn were used to write the code. Network performance on training batches was compared with the performance on a test bach, but the network model was evaluated using predictions too. Possible network improvemets are also sugested in this thesis.

• Keywords: konvolucijska neuronska mreža; CIFAR-10; matrica konfuzije; Accuracy; Precision; Recall; Python
• Keywords (english): convolutional neural network; CIFAR-10; confusion matrix; Accuracy; Precision; Recall; Python
• Language: croatian
• URN:NBN: urn:nbn:hr:235:669818
• Study programme: Title: Mechanical Engineering; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka strojarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka strojarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2021-09-22 14:14:00