Bina Javed
M.Sc. Candidate
Supervisor: Dr. Antonina Kolokolova

Chatbot - Let's talk to machines
Survey, Architecture and Simulation

Department of Computer Science
Friday, November 24, 2017, 10:00 am, Room: EN 2022

---

Abstract

Technological advancements in the field of Artificial Intelligence (AI) have made machines intelligent almost like humans. Extensive AI research in Natural Language Processing (NLP) has closed the gap for Natural Language Understanding for machines. Furthermore, AI advances in deep learning allowed machines to handle complex user queries. The simplest definition of a Chatbot is a bot conversing with a human through a chatting platform (e.g. Facebook Messenger, Slack, Skype, etc).

In this project, we have developed a prototype for a chatbot to handle FAQs for the Registrar's office of Memorial University of Newfoundland using Facebook platform tools. We have developed a chatbot using Chatfuel platform tool and also coded a chatbot from scratch; both chatbots run on Facebook Messenger chatting application. We have trained our chatbots on the FAQs available on the MUN registrar's website. The resulting chatbots serve as an automated virtual helpdesk agents which are capable of answering any general purpose queries such as "add or drop a course, audit a course and apply for coop permit, etc". Furthermore, this project surveys the chatbot technology and available platforms.

---

---

Nnamdi Wilson Ozah
M.Sc. Candidate
Supervisor: Dr. Antonina Kolokolova

When AI's vision gets blurry: the effect of compression on image recognition

Department of Computer Science
Friday, November 24, 2017, 10:45 am, Room: EN 2022

---

Abstract

JPEG and Singular Value Decomposition (SVD) are common techniques used for image compression. These techniques adopt different modes of isolating some pixels that constitute an image, to provide a smaller sized image with little visual difference. However, as compression increases further, the qualityof the images noticeably deteriorates, with images eventually becoming unrecognizable. However, quality is not directly proportional to size and humans’ evaluation of an image quality is very subjective.

In order to find a more objective way of comparing quality of images compressed using different algorithms and parameters, we turn to AI: more specifically, to Inception-v3, a TensorFlow-based image classifier, which is a convolutional neural network trained on ImageNet database. In this talk, we present experimental results showing how compression with SVD and JPEG algorithms affects the ability of Inception-v3 to classify a number of standard images used in image compression literature.

---

---

Oluwatosin Ifeoluwa Adelegan
M.Sc. Candidate
Supervisor: Dr. Ting Hu

Identification of metabolic markers for osteoarthritis and diabetes

Department of Computer Science
Friday, November 24, 2017, 11:30 am, Room: EN 2022

---

Abstract

Osteoarthritis (OA) and Diabetes (DM) are two prevalent diseases in the world today with close association due to their shared risk factors. In clinical practice, medical practitioners have had to rely on the use of conventional apparatus (e.g. radiography) in the diagnosing and monitoring of diseases in patient. However, due to the insensitivity of conventional methods in determining the sudden changes in a biological system, early detection of diseases remains a growing concern. In recent years, research has discovered an association between diseases and metabolic compounds. Hence, the application of metabolomics study in the identification of potential biomarkers for the diagnosis of diseases. In this project, our goal is to apply machine learning techniques to metabolomics data for the prediction of two phenotypes (OA and DM) and the identification of important metabolites.

Keywords: metabolic, markers, metabolomics, machine learning.