Context in Computing. A Cross-Disciplinary Approach for Modeling the Real World
This chapter focuses on multi-view learning, incremental learning and meta-learning, highly relevant yet understudied machine learning principles that enable life-long learning in context-aware mobile applications.We present a Bayesian framework to realize them in modern ubiquitous computing environments that are characterized by dynamic and ever-evolving contexts inferred from heterogeneous sensors with varying churn rates. These techniques enable life-long learning in the context-aware applications to meta-learn their learning principles and continuously adapt the context models in-tune with their environments. We study the benefits of the proposed techniques and demonstrate their advantages for context-aware mobile applications.