Memory-Aware Active Learning in Mobile Sensing Systems

We propose a novel active learning framework for activity recognition, which takes limitations of the oracle into account when selecting wearable sensor data for annotation. It is inspired by human-beings' limited capacity to respond to prompts on their mobile device, introducing the notion of mindful active learning and proposing a computational framework, called EMMA, to maximize the active learning performance taking informativeness of sensor data, query budget, and human memory into account. We formulate this optimization problem, propose an approach to model memory retention, discuss the complexity of the problem, and propose a greedy heuristic to solve it. Additionally, we design an approach to perform mindful active learning in batch where multiple sensor observations are selected simultaneously for querying and design two instantiations of EMMA to perform active learning in batch mode. We demonstrate the effectiveness of our approach using three datasets and by simulating oracles with various memory strengths. Our results indicate that EMMA achieves an accuracy of, on average, 13.5% higher than the case when only informativeness of samples is considered. We observe that mindful active learning is most beneficial when the query budget is small and/or the oracle's memory is weak.