This page presents my statement of purpose in computer science as required for the ICS Ph.D. Portfolio. It summarizes my professional interests in research, teaching, service, and/or product development.
There is a large explosion over Washington D.C. Do you send in the troops, the red cross, the firefighters, or the hazmat team? A rolling brownout is traveling across the country, can you predict and mitigate the effects before it hits you? A large tsunami was generated by an earthquake and is heading directly towards Hawaii, do you have enough timely information to make an informed decision on evacuation procedures? Can you obtain this information without compromising the privacy of those you’re trying to protect?
These events are distributed geographically. These events are distributed through time. These events leave clues in waveforms and metadata that can help predict future states of the environment. Smart, internet connected devices are becoming cheaper and easier to build. For the first time ever, these events can be studied in great depth easily using distributed sensor networks. What set of metrics must be present in these types of events that allow us to process the data, maintain users’ privacy, and build communities of devices based off meta-data and data?
When we develop the ability to examine the environment at an extremely detailed manner, we must weigh the costs of privacy introduced by the amount of data received vs the information that is necessary for the particular research being performed. Data fusion along with the increasing amounts of metadata can make it very easy to uncover information that was intended not to be shared. Is it possible to balance information gathering for scientific purposes against the privacy needs of the general public and our users?
Distributed sensor networks often show emergent behavior where the data from distributed sensors can be categorized into communities. I am investigating how distributed sensors can be categorized into communities based off of the meta-data and the actual payload. By adding this concept of communities (both virtual and geographic), I would like to investigate whether private data is not being leaked by performing fusion on these communities.
A smaller sized distributed power quality sensor network without intelligent triggering situated on the island of Hawaii with 100 devices sampling at 4,000 Hz will produce approximately 2.5 Gb of data per hour or about 65 Gb of data per day. A more modest PQ network spread out across the mainland U.S. with 1,000 to 10,000 sensors will produce anywhere from 26 to 268 Gb per hour which translates to upwards of 62 Tb of data per day!
A distributed infrasound sensor network may sample much slower, but added metadata means that each devices sends approximately 16 Kb per minute. At a scale of 10,000 infrasound sensors the network is transmitting 91 Gb per hour or 219 Gb per day.
Since we do not have infinite bandwidth or infinite computational capacity, we need to find a set of metrics that define and constrain sensor networks and allow us to describe sensor network in a mathematical manner based in limitations set by bandwidth and computational constraints.
My research interests are solving these problems in a unified and transportable way. By leveraging distributed and cloud computing, I aim to provide a framework that can meet the demands of this Big Data explosion while maintaining the privacy of users and to advance the fields of distributed sensor networks, distributed sensor network architectures, and data privacy and integrity.
In the next year I plan to introduce algorithms for dealing with the acquisition of temporospatial data in distributed environments. Services such as TempoDB and OpenTSDB claim to offer a large package of analytics for distributed sensor data, however their acquisition relies on simply metrics such as single temperature values. Current services do not scale when trying to collect data with complicated meta-data and or a large vector of fields per measurement.
Within one to two years I hope to implement distributed algorithms for DSP and event detection as part of my PhD work. I hope to design a set of metrics that will quantify the performance of distributed sensor networks. Given a known set of bandwidths and computational capacities, I hope to find how changes to the network affect the performance and data of the overall network?
After receiving my PhD, I hope to continue my work with distributed sensor networks either in industry or academia. The amount of sensors is increasing exponentially and will require new techniques for data processing.
Over the past three years I’ve been building a framework to detect transients in power quality data. I picked up a lot of my research foundation by taking masters classes in software engineering for smart grids, advanced algorithms, advanced operating systems, theory of computation, AI, and web design.
Over the past year I’ve been developing a framework for the collection, analysis, and reporting of temporospatial data. My funded research through the Infrasound Laboratory at the University of Hawaii at Manoa involves detecting, quantifying, and localizing large infrasonic signals by deploying a large number of distributed sensors that continuously stream data.
Through my funding agency I was able to secure a academic cooperation participant (ACP) position with Lawrence Livermore National Labs and have been working with their Big Data scientists to solve issues such as massive distributed data ingestion with type safe persistent queues. My cooperation with LLNL provides a wide-breadth of resources in distributed computing and Big Data.