ORCID Identifier(s)


Graduation Semester and Year




Document Type


Degree Name

Master of Science in Computer Science


Computer Science and Engineering

First Advisor

Hwa Won Kim


Missing data problem is inevitable in mostly all research areas including Artificial Intelligence, Machine Learning and Computer Vision where we have modicum knowledge about the complete dataset. One of the key reasons of missing data in AI is insufficiency of accurately labeled data. To solve a classification problem using ML or training a Deep Neural Network model, we need a huge amount of labeled data. It is difficult to get labeled data but unlabeled data is inexpensive and available easily. It is usual that we get no more than a single element per class to train our models due to unavailability of enough labeled training data. Strict privacy control or accidental loss may also cause missing data problem. One of the ways of getting training data labeled is using human-in-the-loop, but budget constraints can prevent that option. The objective of this research is to recover the complete signal or missing labels of the dataset using state-of-the-art Machine Learning and Computer Vision techniques. We propose a novel network trained with a few instances of a class to perform Metric Learning. We then convert our dataset to a graph signal and recover the graph completely using Recovery algorithm in Graph Fourier Transform. Our approach performs significantly better than Graph Neural Network and other state-of-the-art techniques.


Semi-supervised learning, Metric learning, Spectral graph theory, Harmonic analysis of graph, Graph fourier transform, Graph signal recovery, Siamese networks, Convolutional neural networks


Computer Sciences | Physical Sciences and Mathematics


Degree granted by The University of Texas at Arlington