CS Other Presentations

Department of Computer Science - University of Cyprus

Besides Colloquiums, the Department of Computer Science at the University of Cyprus also holds Other Presentations (Research Seminars, PhD Defenses, Short Term Courses, Demonstrations, etc.). These presentations are given by scientists who aim to present preliminary results of their research work and/or other technical material. Other Presentations serve as a forum for educating Computer Science students and related announcements are disseminated to the Department of Computer Science (i.e., the csall list):
rss RSS Directions Directions

Presentations Coordinator: Demetris Zeinalipour

PhD Defense: Conditional Generative Denoising Autoencoder, Mr. Savvas Karatsiolis (University of Cyprus, Cyprus), Friday, October 25, 2019, 10:00-11:00 EET.


The Department of Computer Science at the University of Cyprus cordially invites you to the PhD Defense entitled:

Conditional Generative Denoising Autoencoder

Speaker: Mr. Savvas Karatsiolis
Affiliation: University of Cyprus, Cyprus
Category: PhD Defense
Location: Room 148, Faculty of Pure and Applied Sciences (FST-01), 1 University Avenue, 2109 Nicosia, Cyprus (directions)
Date: Friday, October 25, 2019
Time: 10:00-11:00 EET
Host: Prof. Chris Christodoulou (cchrist-AT-cs.ucy.ac.cy)
URL: https://www.cs.ucy.ac.cy/colloquium/presentations.php?speaker=cs.ucy.pres.2019.karatsiolis

Abstract:
Machine Learning’s unsupervised learning paradigm is primarily concerned with the development of generative models that learn the data distribution of real-world problems. In this context we present a conditional generative model, the Conditional Generative Denoising Autoencoder that relies its mathematical foundation on the theoretical framework of the conventional Denoising Autoencoder. The model generates images according to the user’s preferences that define both the desired and the undesired image characteristics that are evident in the training dataset. For example, training on a dataset containing human faces that may or may not process some specific characteristics like eyes and hair color, face shape, gender, presence of a beard, wearing a hat or glasses, being old or young etc., the model is able to generate images of faces either possessing or not possessing such characteristics. Selected characteristics may constitute a broader categorical meaning in which case the model generates images corresponding to one or more specific problem categories. Besides the structures trained with unsupervised learning, the model also contains a classifier structure trained with supervised learning. The proposed model’s functionality relies on the concatenation of the features produced by the classifier and the features calculated by the model’s structures trained with unsupervised learning. Unifying the two learning regimes (supervised and unsupervised learning) in order to enhance unsupervised learning has not generally been explored by the research community. On the contrary, the use of unsupervised learning for improving supervised learning is widely known as semi-supervised training and is being researched to a greater extend. The value of the proposed model rises from its ability to generate good quality images according to predefined conditions (labels), while the available Machine Learning models performing the specific task are very few in number, generally perform moderately and have important disadvantages. Generative models are very important in the Machine Learning field and the Artificial Intelligence field in general, mainly because of the belief that unsupervised learning and its variants occupy a large portion of human brain functionality. The proposed methodology besides constituting a novel generative model, also establishes the argument that supervised learning may assist unsupervised learning through an interaction of the two learning paradigms.

Short Bio:
Savvas Karatsiolis is a PhD candidate at the Department of Computer Science under the supervision of Professor Schizas N. Christos. His research interests lie in the area of deep learning and especially unsupervised learning, self-supervised learning and recurrent neural networks. He holds a HND in Electrical Engineering from Higher Technical Institute, a BSc in Computer Engineering from Intercollege Nicosia and a MSc in Information and Communication Systems from Open University Cyprus.

  Other Presentations Web: https://www.cs.ucy.ac.cy/colloquium/presentations.php
  Colloquia Web: https://www.cs.ucy.ac.cy/colloquium/
  Calendar: https://www.cs.ucy.ac.cy/colloquium/schedule/cs.ucy.pres.2019.karatsiolis.ics