Keeping AI Models Relevant in the Presence of Drift

How does an evolving environment affect the predictions of a Machine Learning (ML) system? Often, the characteristics of real-world environments change as time passes. When we build an AI system that looks for behavioral markers for a certain condition, what do we do when the behavioral markers change? They are no longer a good predictor, and we need to retrain our predictor system. As an environment changes and evolves, its evolutionary path, sometimes called 'drift,' can affect how a system in that environment functions. Environmental drift may adversely affect how well an ML system makes predictions. Robert Vanderwall will discuss a few types of prediction models and their characteristics. He will provide a clear definition of drift, examine the contexts in which it applies, and give plenty of useful examples. He will investigate various types of environmental drifts, their causes, and mathematical descriptions of the evolutionary path. Robert will discuss our understanding of environmental evolution and mathematical machinery to gain insight into its effects on model predictions. He will dig deeper into the implications of evolution on predictions of ML models. Robert will provide you with the tools and approaches available to identify drift, mitigate that drift, and balance the increase in complexity with the improvements.