Research

Smart homes are distributed systems that should provide reliable home automation services. Distributed controllers have been used to improve the reliability of smart homes. However, despite the redundancy in controllers, the reliability of smart homes is hindered by the unreliable network infrastructure comprising home Wi-Fi and low-power wireless networks. As a result, smart homes cannot be trusted with much more than casual automation to date. This dissertation identifies the limitations of traditional fault-tolerant approaches that require inter-controller coordination over an unreliable network infrastructure. A new class of decentralized control paradigm named Banyan, with fully autonomous controllers, is introduced by this dissertation. The dissertation also presents a new coordination-less universal idempotent scheme that prevents duplicate actuation. Under a realistic system model for smart homes, it is proved that reliable, consistent, timely control decisions are made in a decentralized unison. This paradigm's first test bed implementation is presented in the smart home context. Multiple controllers may contribute to running automation scripts at opportune points during their run and control flows among the controllers to enhance reliability and availability while maintaining consistency. This dissertation concludes with simulation and empirical results that demonstrate the capability of Banyan to support reliable automation in the presence of unreliable networks in the smart home.

​

Mentor: Dr. Chenyang Lu 

Human beings, most life forms, and cameras use the visible and unpolarized part of the electromagnetic spectrum to see or construct images. In this research rotation, I empirically verify whether sensors that can capture the polarization of light (these sensors were designed and fabricated in the lab) can form better images under difficult light conditions.

​

Mentor: Dr. Viktor Gruev 

Geographically distributed organizations require their data objects to remain coherent. However, challenges arise when multiple sites update the same object, physical items are shipped in all directions, and communication is sporadic, low bandwidth and unreliable. I designed and researched transaction representation languages and communication automation patterns to solve these challenges. The research provided clear visibility and accurate data for multi-site decision-making and operation.

​

The results of this research are key building blocks of Visual Office (VO), an ERP system I developed on top of this distributed data underpinning. VO solved the problem of distributed heterogeneous data stores, some of which were databases and a diverse set of other data sources.

​

VO was implemented in large organizations. The company benefitted from particular success with distributed businesses, such as franchise management and chain stores. VO offered modules such as inventory control, order management, invoicing, contact information, and other business forms.

Wireless sensor networks can improve patient care by collecting continuous vital signs and providing clinical decision support. Such a system has the potential to save lives with real-time alerts sent between nurse bedside visits or physician rounds (in non-critical care hospital units, these periods may be as long as 8 hours). With a reliable system, telemetry can continue during treatment in other hospital departments or at home, encouraging patients to be ambulatory, which supports faster recovery.

​

We, a multidisciplinary team of Physicians, Machine Learning (ML), and System researchers from Washington University School of Medicine in St. Louis and Washington University Engineering School, developed a system that enables early diagnosis of impending clinical deterioration. The systems research focuses on reliably delivering vital signs in hospital environments to provide real-time alerts. The predictions of the Early Warning System (EWS) we developed are highly associated with ICU transfers, and the algorithms alerted personnel hours before the onset of conditions.

​

Mentors:

• Dr. Chenyang Lu 

• Dr. Thomas Bailey 

• Dr. Marin Kollef 

A primary challenge in allowing computers to operate as cash registered was the multifarious ways data and time could be manipulated by users, which conflicts with IRS regulations. This research explored how a hardware chip and system software could be integrated to enable memory and real-world clock that could not be tampered.

​

The results of this research became part of the successful Thinking Cashier product.

​

Mentors:

• My father, Lior Dor 

I was part of a multidisciplinary team from Washington University School of Medicine and Washington University Engineering School, both in St. Louis. The requirements were guided by an advisory committee of older adults. This research aims to develop and test a novel fall detection system for living at home. Our research is novel in its approach and data sources. We are not trying to devise a fall detection algorithm per se, nor testing our ideas on students. Instead, we characterize older adult gait and movement patterns using machine learning enabled by volunteers from the target population. The research goals are to build a non-intrusive, omnipresent fall detection system with high efficacy.

​

Mentors:

• Dr. Chenyang Lu 

• Dr. Susan Stark 

Early IBM PCs had no real-world clock, were limited in ISA slots, and occasionally
halted. My father, Lior Dor, invented; and I built and researched how to enable the original CPU to perform all of its functions while, in parallel, injecting a real-world clock into the OS. Subsequent research added a correctly (timely) injected RESET button and signal.

The result of the research culminated in a product named Up2Date that was a wild commercial success for many years until the Intel 80286 CPU became prevalent.

​

Mentors:

• My father, Lior Dor 

Researched the level of parallelism and accelerating scale of various bioinformatics algorithms by exploiting the inherent parallelism of DNA and RNA representation (the four nucleotide monomers letters). The algorithms were implemented on high-performance embedded computing architectures, comparing and mixing FPGA, GPU, and multicore chips along the processing pipeline.

​

Mentors:

• Dr. Jeremy Buhler 

• Dr. Roger Chamberlain

• Dr. Mark Franklin

Before massively parallel server farms made full-text search plausible, I researched various ways to index and store a large corpus of textual data for fast retrieval. At the culmination of the research, the technology returned results in sub-second times when running on a single 80286-based computer. With this technology, we built Probe, a technician management system with a fully searchable knowledge base.