Smart lighting infrastructure and attached sensors promise the ability to perform advanced analyses at unprecedented scale —but only if data processing infrastructure can cope with the complexity and volume. These challenges of complexity and volume are particularly poorly addressed by existing data processing engines. There is little support for complex analytics tasks such as realtime outlier detection over sensor streams in existing Big Data engines, which typically operate on static datasets in a centralized cluster environment; this is a poor fit for dynamic sensor data, which is spread across distributed devices, with complex privacy and administrative constraints on processing.
This project will investigate the co-design of systems and statistical analyses for streaming, real-time detection of outliers and interesting patterns in data collected via lighting infrastructure (i.e., sensors, devices) in large buildings, city infrastructure, and smart homes. The team’s long-term goal is a distributed Big Data engine capable of high-volume, streaming complex analytics tasks over smart lighting infrastructure. During the initial five-month Grand Challenge period, the team will focus on outlier detection—specifically, analyses of deployed lighting and sensor behavior (e.g., is a particular sensor malfunctioning or a particular model of device failing to connect?) and detection of relevant environmental changes (e.g., physical intrusion). Using data from Philips and MIT, the team will prototype a high-scale implementation of several best-of-class outlier detection and explanation techniques and pave the way towards a larger system implementation suitable for potential deployment with Philips.
This project is funded by Philips Lighting.