Title: Building Better Models Faster: A Charming Tool for Versioning Data and Experiments
Abstract: Machine Learning is experimental in nature. We try various features, algorithms, modeling techniques, and parameter configurations in order to find the best solution for the problem as quickly as possible. The challenge is keeping track of what worked and what didn’t, as well as preserving reproducibility while maximizing code reusability. This is where MLOps comes to play an essential role in Machine Learning systems development. This tutorial equips ML researchers and practitioners, who want to automate and operate their ML products, with a lightweight and interoperable MLOps tool to spend less time manually tracking results, version datasets and experiments with reproducibility, and tune hyperparameters automatically.
Bio: Tung Dao is a Senior ML Platform Engineer at Shopee (Singapore) where he is focusing on developing a distributed training framework and orchestrating model deployment at the production scale. He is also an MLOps Lead at AICOVIDVN, a community project collaborated with Vietnam Government Agencies, where he architects AWS-based ML systems for a cough-based COVID-19 detection application. Previously, he was a Machine Learning Engineer at Dell Technologies where he focused on building Deep Learning models for cloud-based Computer Vision applications. Tung graduated in Computer Engineering from Hanoi University of Science and Technology and received a Research Scholarship to research about Computer Vision at Nanyang Technological University (Singapore). He has experience as a consultant in designing ML systems for different private companies in Singapore and Vietnam. He is also the Founder of an AI blog (AIengineer.net) and the Admin of the MLOps Vietnam community.