We are presenting a demo titled "Building IoT Applications with Accessors in CapeCode" at ICCPS during CPSWeek 2016 in Vienna, Austria.
I recently graduated from UC Berkeley EECS with a PhD and I currently work at Periscope Data as a software engineer.
I was previously part of the Ptolemy research group, advised by Edward A. Lee. My Ph.D. work addresses challenges in developing data-driven cyber-physical systems. Many sensor network applications rely on the heterogeneous integration of hardware and software. In the presence of dynamic component interactions and the perpetually changing environment, system behavior often needs to be dynamic, furthermore, it needs to be learned. In this setting, design, test and integration become more integrated, requiring novel approaches for composing smart systems.
I am working on a framework named PILOT (Ptolemy Inference, Learning, and Optimization Toolkit), which aims at enabling dynamic development of IoT applications based on on-line machine learning and optimization. PILOT is a Java-based toolkit that is developed as part of the Ptolemy II framework. The main challenges we address with PILOT are utilizing streaming data in developing data-driven systems and aspect-oriented modeling for modular design of complex cyber-physical systems.
The rapid growth of networked smart sensors today offers unprecedented volumes of continually streaming data. This renders most classical control and optimization techniques that are based on monolithic approaches ineffective for cloud-based large scale application design for the IoT. We present PILOT (Ptolemy Inference, Learning, and Optimization Toolkit), an actor-oriented machine learning and optimization toolkit that is designed for developing data-intensive distributed applications for sensor networks. PILOT presents an actor interface that enables developing complex learning and optimization tasks for large scale sensor networks in a scalable and state-space aware fashion. We demonstrate key capabilities of the toolkit with a cloud-based cooperative mobile robot target tracking scenario, and study how the framework achieves design and implementation of control policies by including higher-level abstractions of learning and optimization tasks as part of the system design.