EMA: Efficient Model Adaptation for Learning-based Systems

This paper introduces EMA, the first model adaptation system designed to help learning-based systems adapt to evolving environments with minimal operational overhead. It addresses costly model training and extensive data collection challenges by reducing expensive model training.