About

Master’s of Science Student in Computer Science

matthew.lesko-krleza AT mail.mcgill.ca

Matthew Lesko-Krleza

Hi, I’m Matthew. I’m currently a graduate student in computer science at McGill University located in Montreal, Quebec, Canada. Broadly speaking, my ongoing research interests are at the intersection of deep learning, computer vision, and medical imaging. More specifically, I’m interested in supervised-learning disease classification and detection in medical images (X-Rays, and Computed Tomography scans). Professionally, I’m a software developer with a broad range of experience in machine learning, data engineering, web development, and software testing and quality assurance.

Publications

Coming soon…

Work Experience

Software Development Engineer Intern at Amazon Robotics LLC.
(May 2019 – September 2019)
I worked on a prototype computer vision product aimed to help automate inventory control, quality assurance, and manual item scanning for Amazon’s fulfillment centres (potential of saving money in the order of millions yearly). My team and I developed the prototype by using RGB camera hardware, Python, OpenCV, and Deep Learning (PyTorch).

Solution Associate at Deloitte Digital
(May 2018 – September 2018)
I worked on a customer relationship management software solution (Salesforce) for a multi-billion-dollar private client aimed to digitize their sales department and automate accounting procedures (help save money in the order of 1-5 million yearly). I developed front-end and back-end features in Apex (Java-like OOP language), JavaScript, and HTML,

Software Developer at Ericsson Canada
(May 2017 – September 2017)
I worked on an internationally sold media delivery network. I worked on quality assurance and testing by writing code analysis scripts in Bash and by improving test code coverage in Java.

Research Projects

Compositional Networks for Tumor Localization
(May 2020 – August 2021)
I evaluated the use of Compositional Networks (a machine learning system using deep learning and compositional models) for extremely data-efficient tumor localization in medical imaging by using Python, PyTorch, and Comet.ml. My research thesis covered this topic, the thesis is currently being finished.

Naive Neural Network Transfer Reinforcement Learning
(February 2020 – May 2020)
I evaluated the use of transfer learning of deep reinforcement learning state-value-based and policy-gradient based systems such as Deep Q-Networks, REINFORCE, and Deep Deterministic Policy Gradient in 2-dimensional environments using Python, PyTorch, Google Cloud AI, and Open AI’s Gym package. My work showed that policy-gradient models transfer more successfully than state-valued methods within my tested environments. My paper can be read here.

Natural Language Processing for Fake News Classification
(October 2019 – December 2019)
My team and I evaluated classical (Naive Bayes) and deep learning machine learning NLP models (Bi-LSTM, BERT) for fake news classification by using Python, NLTK, and PyTorch. Our work showed that BERT outperformed all other techniques. Our paper can be read here.

Movie Recommendation as a Use Case for Bipartite Link Prediction
(October 2019 – December 2019)
I created a bipartite graph from the MovieLens-1M (ML-1M) dataset and applied two existing link prediction algorithms for movie recommendation and rating prediction on the constructed graph. My method demonstrates results that are just under the performance of the state-of-the-art. The paper can be read here.

Hackathon Projects

CodeJam 2020 (McGill University Hackathon)
(November 2020)
My team and I developed developed an AI driven online clothing shopping solution to help users “digitally try” clothes on before buying them. Out of over 30 teams, we won 1st place and the “Smart Online Shopping Solution” subcategory for a total prize of $4000.00. I worked on the front-end and back-end components in JavaScript and React.JS.

MAIS Hacks 2020 (McGill University Hackathon)
(October 2020)
My team and I developed developed an AI-driven solution for skin lesion detection through their personal smartphones. My team and I won 2nd place and the “Best Social Good” subcategory prize out of over 50 teams.

ImplementAI 2019 (McGill University Hackathon)
(October 2019)
My team and I developed developed an AI-driven solution online parental censorship. I trained the deep learning models for image censorship. My team and I won 2nd place out of over 30 teams. Learn more about it here: https://devpost.com/software/soteria-iemtas