Data abundance along with scarcity of machine learning experts necessitates progressive automation of end-to-end machine learning workflows. To this end, Automated Machine Learning (AutoML) has emerged as a prominent research area. Real world data often arrives as streams or batches, and data distributions evolve over time causing concept drift. Models need to handle non-i.i.d. data and transfer knowledge across time through continuous self-evaluation and adaptation adhering to resource constraints. Creating self-maintaining models which automatically adapt to concept drift to operate in a lifelong learning setting is a challenging problem. In this talk, we introduce the key concepts in this area, and present our model entitled AutoGBT which attempts to address this problem using an adaptive self-optimized machine learning pipeline based on gradient boosting trees with automatic hyper-parameter tuning using Sequential Model-Based Optimization (SMBO). Our model secured the first position in the recent NeurIPS 2018 AutoML challenge.
Jobin Wilson is a Principal Data Scientist - R&D at Flytxt, and leads Flytxt's AI algorithms R&D group. He is an active researcher and innovator, and is responsible for driving Flytxt's R&D and intellectual property generation efforts in AI algorithms. He has published several research papers and patents in pattern recognition and machine learning. He holds an MS(Research) in Electrical Engineering from IIT Delhi. He is presently pursuing his PhD in Electrical Engineering at the same institution. His research interests include Recommender systems, Artificial Intelligence and online non-stationary learning.