Thesis Defense - Barış Büyüktaş (MSEE)

Barış Büyüktaş - M.Sc. Electrical and Electronics Engineering

Prof. Tanju Erdem - Advisor 

Prof. Çiğdem Eroğlu Erdem - Co-advisor

Date: 11.08.2020

Time: 14:00

Location: This meeting will be held ONLINE. Please send an e-mail to gizem.bakir@ozyegin.edu.tr in order to participate in this defense.

  CURRICULUM AND ACTIVE SELF-PACED LEARNING WITH

MINIMUM SPARSE RECONSTRUCTION FOR FACE RECOGNITION

Thesis Committee:

Prof. Tanju Erdem, Ozyegin University

Prof. Çiğdem Eroğlu Erdem, Marmara University

Prof. Fatih Uğurdağ, Ozyegin University

Asst. Prof. Furkan Kıraç, Özyeğin University

Asst. Prof. Nihan Kahraman, Yıldız Technical University

Abstract:

In this thesis, we present two different frameworks for face recognition. The first one includes a novel curriculum learning (CL) algorithm for face recognition using convolutional neural networks. Curriculum learning is inspired by the fact that humans learn better when the presented information is organized in a way that covers the easy concepts first, followed by more complex ones. It has been shown in the literature that that CL is also beneficial for machine learning tasks by enabling convergence to a better local minimum. In the proposed CL algorithm for face recognition, we divide the training set of face images into subsets of increasing difficulty based on the head pose angle obtained from the absolute sum of yaw, pitch and roll angles. These subsets are introduced to the deep CNN in order of increasing difficulty. Experimental results on the large-scale CASIA-WebFace-Sub dataset show that the increase in face recognition accuracy is statistically significant when CL is used, as compared to organizing the training data in random batches.

The second framework is a novel framework, which can progressively train classifiers for face recognition when the available face images of subjects increase gradually. The framework starts with a small set of annotated images and includes new face images into the training set with minimum expert annotation effort using two different strategies: self-paced learning (SPL) and active learning with minimum sparse reconstruction (AL-MSR). SPL is used for automatic annotation of easy images, for which the classifiers are highly confident. AL-MSR is used to request the help of an expert for annotating difficult or low confidence images, which are further filtered for diversity using minimum sparse reconstruction. We combine self-paced and active learning with minimum sparse reconstruction in a single framework, namely ASPL-MSR. We improve the recently proposed ASPL framework [1] by introducing MSR for eliminating “similar” images from the set selected by AL, which significantly reduces the required cost of expert annotation effort while providing a comparable recognition performance. The proposed framework is also robust against incorrectly annotated images in the training set and shows faster convergence properties. The proposed method (ASPL-MSR) is tested using two large-scale data sets (CASIA-WebFace and CACD) and promising results have been obtained. We can achieve the same face recognition accuracy by using less expert-annotated data as compared to the state-of-the-art.

.

Bio: 

Barış Büyüktaş received the B.Sc. degree in Electrical & Electronics Engineering Department from Ozyegin University in 2018. He started his M.Sc. as a full scholarship student at Özyeğin University in 2018. He was funded as a research assistant by the Scientific and Technological Research Council of Turkey (TÜBİTAK) and he worked as a graduate teaching assistant at Özyeğin University. He has been doing research under the supervision of Prof. Çiğdem Eroğlu Erdem and Prof. Tanju Erdem. His research interests include image processing, computer vision, machine learning, human-computer interaction.

--------------------------------

Kişisel Verilerin Korunması/Aydınlatma Metni

Özyeğin Üniversitesi Fen Bilimleri Enstitüsü bünyesindeki programlarda, izleyici olarak katılınan online platformlarda gerçekleştirilen aleni akademik faaliyetlerin sadece idari ve yasal yükümlülüklerin yerine getirilmesi amacıyla kayıt altına alındığı, söz konusu kayıt ile elde edilen kişisel verilerin ilgililer tarafından başka bir amaçla kullanılamayacağı hususunda şahsınızı aydınlatır, toplantıya gönüllü katılım ile bu hususu beyan ve taahhüt etmiş olduğunuzu belirtir, gereğini önemle rica ederiz. 

Fen Bilimleri Enstitüsü