Aktivno učenje dubokih konvolucijskih modela

undergraduate thesis

• Title (english): Active learning for deep convolutional models
• Author: Tin Reiter
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Mladen Karan (član povjerenstva)
• Committee member: Josip Šarić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2020-07-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Cilj ovog rada je pokazati aktivno učenje dubokih konvolucijski modela te usporediti rezultate takovog učenja sa rezultatima dobivenim nasumičnim odabirom podataka. Želimo vidjeti da li možemo dobiti bolju performansu odabirom podataka za označavanje pomoću aktivnog učenja u usporedbi s performansom dobivenom nasumičnim odabirom.
U radu prvo opisujemo strukturu umjetnih neurona i dubokih modela. Zatim definiramo aktivno učenje i opisujemo razne metode koje koristimo da ga postignemo. Nakon toga opisujemo korištenu programsku podršku te skupove podataka koje koristimo. Ovaj rad koristi skupove podataka MNIST i PASCAL VOC 2007. Na kraju prikazujemo dobivene eksperimentalne rezultate na dva prijašnje navedena skupa podataka koji nam pokazuju da koristeći aktivno učenje za odabir podataka doista dobivamo bolje performanse nego da koristimo nasumični odabir podataka.

Abstract (english)

The goal of this paper is to demonstrate active learning of deep convolutional models and to compare such learning with the results gained by randomly selecting data to label. We want to see if we can gain better performance by selecting data to label using active learning compared to the performance gained by randomly selecting data.
In this paper we first describe the structure of artificial neurons and deep models. Then we define active learning and describe the various methods we use to achieve it. After that we describe the software and datasets used. This paper uses the MNIST and PASCAL VOC 2007 datasets. In the end we show the experimental results gained on the previous two datasets that show that using active learning to select data to label we gain better performance than randomly selecting data.

• Keywords: Aktivno učenje; Duboki konvolucijski modeli; Računalni vid; MNIST skup podataka; PASCAL VOC skup podataka; Klasificiranje slika
• Keywords (english): Active learning; Deep convolutional models; Computer Vision; MNIST dataset; PASCAL VOC dataset; Image classification
• Language: croatian
• URN:NBN: urn:nbn:hr:168:995046
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 16:02:18