Graduate Program Seminars Nov 23, 2017

Yubin Li
M.Sc. Candidate
Supervisor: Dr. Jian Tang

Explore meta-learning via inductive transfer learning methods

Department of Computer Science
Thursday, November 23, 2017, 10:00 a.m., Room EN 2022


Abstract

Meta-learning studies how learning systems can improve themselves through experience, the goal of meta-learning is to figure out how learning itself can become flexible, or how to extract the underlying knowledge in certain domain and use it to improve itself in future learning task. In this project, the plan is to utilized inductive transfer learning method to achieve meta-learning. Candidate
applications include Survival analysis, Phase transition, and Autonomous-driving.



Majid Beheshti Mohtasham
M.Sc. Candidate
Co-Supervisors: Drs. Saeed Samet and Ting Hu

Classification of SNP Genetic Data of Type 2 Diabetes - A Hybrid Approach Using Quantum Evolutionary Algorithm and Support Vector Machine

Department of Computer Science
Thursday, November 23, 2017, 3:00 p.m., Room EN 2022


Abstract

The improvement of information and communication technology in healthcare is undeniable and computer science proves its contribution in this progress. Employing information technology in health sciences leads to produce a huge amount of data. This is the most important reason to show the essential role of data mining. Data scientists should work on wide range of various datasets as
all the diseases produce data. In the same way, diabetes collects large data as a disease. Diabetes is a significant area of focus for health and bioinformatics scientists because its rate is growing in most countries [3]. Data mining and machine learning techniques could be used as a solution [5] for prediction and diagnosis of diabetes.

In this thesis, we work on the binary classification of genetic datasets of type 2 diabetes(T2D) and utilize one of the best machine learning techniques for this type of classification, the Support Vector Machine(SVM). The proposed method uses the SVM for T2D classification by evaluating potential
gene-gene (SNP-SNP) interactions. We also intend to employ the Quantum Evolutionary Algorithm(QEA), which is the most useful evolutionary algorithm, in this research. The plan is applying QEA for the training phase of the SVM. As a result of the proposed approach, high accuracy of predictions in classifying genetic T2D datasets are expected.