Speaker: Nicholas Lane, Microsoft Research Asia
Title: Enabling Population-scale Sensing Systems with Dynamic Inference Pipelines and Community Similarity Networks
Sensor-equipped wearables and smartphones are creating new application domains and transforming existing ones - from mobile health to anticipatory computing, mobile sensing is radically changing the way we collect and mine information about people's activities, contexts, and social networks. However, a number of challenges stand in the way of delivering mobile sensing to the masses. For example, how can we develop mobile systems that are capable of accurate and timely inference of high-level human behavior in the face of uncertain and fluctuating energy, computation and cloud availability? Moreover, how can these systems cope with the level of diversity among users (e.g., demographics, lifestyles) and contexts found in large-scale sensing systems?
In this talk, I will present two key ideas that can help scale mobile sensing systems from 100s of people to potentially 100s of millions. The first idea - Dynamic Inference Pipelines (DIP) - enables mobile systems to run powerful chains of sensor inference algorithms that react and reconfigure to changes in system resources (such as, energy and wireless networking conditions). The automatic adaption of pipeline configuration (e.g., sensor data features, cloud computation off-loading), enabled by DIP, leads to significant gains in resource utilization while also balancing the key inference criteria of accuracy and latency. Second, I will discuss Community Similarity Networks (CSN) - an activity recognition framework that exploits crowd-sourced sensor data to personalize classifiers with data contributed from other similar users. CSN incorporates inter-person similarity measurements into the classifier training process resulting in increased robustness to complex environments and heterogeneous user populations. These techniques combine to move mobile sensing forward: nailing it before we scale it.
Nic Lane is a Lead Researcher at Microsoft Research working in the Mobile and Sensing Systems group (MASS). Nic received his Ph.D. from Dartmouth College (2011) where he worked with his co-advisors Andrew Campbell and Tanzeem Choudhury at the intersection of machine learning and mobile sensing. His dissertation helped pioneer community-guided techniques for learning models of human behavior that enable mobile sensing systems to better cope with diverse user populations encountered in the real-world. Nic is an experimental computer scientist who builds novel mobile sensing applications and systems based on well-founded computational models. His work has received a number of awards including best paper awards from UbiComp '12, MobiCASE '12 and PhoneSense '11, and a best paper nomination from UbiComp '11. Nic's recent TPC service includes MobiSys, Ubicomp, HotMobile and SenSys.
For additional information contact: Eyal de Lara