Machine Learning and Big Data Healthcare
Nursing care is one of the most expensive components in the US healthcare systems; hospitalized patients spend most of their times under the nursing care. In collaborations with researchers from the University of Florida, this project deals with the discovery of best nursing care practices through statistical analysis and machine learning techniques using a large collection of standardized high dimensional and sparse nursing care plans data. In particular, we are developing models with accurate prediction of patient outcomes relevant to palliative care at the shift level and the episode level using three sets of predictors (patient characteristics, nursing care, and nurse characteristics). Some of the patients outcome of our interest include chronic pain, cardiac pump effectiveness, death anxiety, anticipatory grieving, etc.Also we are developing visualization techniques and decision support systems for immediate translation of findings into useful and meaningful decision support at the point of care.
Sponsor: National Institute of Health
Context Aware Wireless Sensors and IoT Networks
According to a recent study by Goldman and Sachs, over 12 billion devices coupled with heterogeneous sensors are already connected to the Internet of Things (IoT) and by 2020 that number should become 20 billion. Recent works have already attempted to capitalize on the results from social networks research to generalize the concept to social networks of devices in the IoT realm. One of the core motivations for the proposed project is based on the observation that current practices related to the design of IoT systems are still rather compartmentalized, and the dynamics-aware fusion of data collection, analytics processing and actuation across a variety of heterogeneous IoTs has limited support. This project addresses the evolving coupling of heterogeneous connected smart objects (things) and human end users that dynamically engage in various interactions for better functioning of the objects and enhanced user experience. We have developed the concept of Heterogeneity And Context Aware Dynamic Avatars (HACADAs) that allow inter-operability, and facilitate efficient collection and processing of data for analytics. Such abstractions also allow provision of services by and to the smart objects that may originate from diverse environments such as homes, cities, plants, healthcare, etc. By establishing an Adaptable Internet of Things (AIoT) structure, our research will enable automatic and virtual resource optimization and intelligence for various instances of inter-connectivity among such objects.
Sponsor: National Science Foundation
High Performance and Reconfigurable Computing
The project explores the design of algorithmic mapping of data and compute intensive applications on reconfigurable platforms such as FPGAs, with the aim to improve resource utilization, including amount of compute and memory logic, data I/O, throughput, and power consumption. We use hardware description language as well as high-level synthesis (HLS)-based OpenCL framework to implement our techniques. This study also allows us to investigate the tradeoffs between computer and memory operations, as well as in memory computations. Applications of interest include embedded systems for medical imaging, proteomics, synthetic aperture radar imaging, and network intrusion detection.
Device to Device Multimedia Communication using Network Coding
With the prevalence of smart mobile and handheld devices, collaborative communication among these devices is becoming essential to realize efficient communication systems. Instantly decodable network coding (IDNC) considers the same problem of collaborative communication, but focuses on instant decodability. In particular, a network-coded packet should be decodable by at least one of the devices in a cooperating group. This characteristic of IDNC makes it feasible for real-time multimedia applications in which packets are passed to the application layer immediately after they are decoded. However, for applications such as video streaming, not all packets have the same importance and not all devices are interested in the same quality of content. This problem becomes more interesting and challenging when additional, but realistic constraints, such as strict deadline, bandwidth, or limited energy are added in the problem formulation.In this project we are investigating the use of IDNC to design D2D multimedia communication solutions for sharing broadcast multimedia contents.
Sponsor: National Science Foundation