Embedded Files
Research Overview: I am interested in solving a wide range of problems with a flavor of complexity in them. I like to combine science and engineering that uses machine learning, data science, and intelligent adaptive systems to model and analyze non-linear, system dynamics. For example, I aim to understand how global ecological changes going to effect the timing and abundance of the eco-services and the landscape use of Alaska fisheries; how reliable are the brain-wave patters when you think (about an apple for example) to be used for user authentication; or measuring how prepared our institutional and government structures are address the global environmental change by mapping the relationships between the actors.
Coupled socio-ecological systems research:
I work to understand of the adaptive capacity of Arctic communities and climate change. I worked on a variety of projects that address the community level adaptive capacity and the rapidly changing ecological systems in Alaska. My projects and collaborations that contribute to understanding how society interacts with the environment include NSF EPSCoR funded research (award 1208927): Alaska Adapting to Changing Environments, which studied changing Arctic ecology and its implications on society; A Municipality of Anchorage study of the solid waste composition at the municipal transfer station and the landfill dynamics using unmanned aerial vehicles with computer vision post-processing; IRD’s automatic anomaly detection for the irrigation canals of Afghanistan
I built verified, predictive, adaptable agent-based models of coupled socio-ecological systems of Kenai River fisheries and the subsistence hunting land-use of an Arctic community that were used as decision support tools to test the impacts of proposed policy alterations, the effects of changing ecological drivers, or the impact of a proposed oil development site. I also conducted multiple Social Network Analysis and produced models of stakeholder, institutions and non-government organizations that contribute to understanding of the local, state and federal governance network connectivities and are used to analyze and plan climate mitigation
Sensor networks:
My basic research interests are on the architecture side of designing distributed, decentralized, locally connected, asynchronous sensor networks. The project was originally funded by the Department of Homeland Security and investigates novel sensor network architecture composed of simple, locally connected, asynchronous components with low communication bandwidth to detect event based network disturbances. The deployed network self-assembles and each component reads signals (acoustic signatures, antennae activations, or changes in the network topology ) from the environment. A neuromorphic network design uses the in-place components to validate if the received signals belong to a desired event or not. The network architecture can be viewed as a hardware implementation of a direction-less artificial neural network for event sensing. From the data science perspective, this project redefines the meaning of information. Since each network component reads-in information from the environment, communicates the signals to locally adjacent components, and verifies if the signals belong to an event of interest, the information is not the state of individual component, but rather the spatio-temporal pattern of component activations.
Theory:
Finally, my theoretical work on the agent based models focused on development of a statistically based machine learning toolkit to assess the changes in the agents behavior to understand model sensitivity, adaptive capacity and model emergent behavior. It aims to answer the question: "How do the collectives of agents compute or solve problems?" Common assessment of an agent based model crudely simplifies the effects of multiple drivers of agent behavior into a few plots of measured outputs disregarding how agents achieved the goal. The Geometry of Behavioral Spaces (GOBS) framework uses agent trajectories as an input and creates a Markovian state-space transition network where the nodes represent the generalized agent behaviors (behavioral primitives) with the edges being the likelihood of an agent transitioning from one stable behavior to another. Two model executions using two sets of model parameters are then computationally compared using the resulting GOBS networks to assess how model behavior changed in-terms of agent behaviors. The broader impact of this work includes the first non-reductionistic, behavior-based assessment of agent-based models; automating the models parameter search for interesting value configurations; and the future applications of this approach as a general pattern finding framework in the fields of computer security, material science, and biological genotype vs. phenotype bindings.
ADAC: Distributed sensor networks
Science for Alaska: Distributed sensor networks
How do we know what we know in complex systems
Current Research Projects and Topics:
Current Research Projects and Topics:
Understanding the interaction Dynamics between the Social and Environmental Systems
Alaskans living in rural communities rely on the subsistence harvest of salmon, moose, caribou and whale to survive. This collection of projects studies how much work would it take for the subsistence communities of the circumpolar North to adapt to the changes of the subsistence resources caused by the global environmental change. This work involves building agent based models of coupled socio-ecolgical systems, building plausible future scenarios and analyzing their impact on intra-system couplings, and analyzing social networks of actors.
Kenai Peninsula is known for its fishing. This project is centered around understanding the interactions between the social and environmental systems there. The project goals are to answer questions about adaptive capacity of the systems, determine the best use policies, protect fragile salmon populations, and develop decision support tools for educational and management purposes along the way.
The communities on the Arctic Sea are traditional caribou hunters, but in recent years moose is also available as a harvest resource. The warming climate will also make some fish species more abundant. On the other side of the equation is the oil industry developing new extraction sites that might adversely effect both the hunters and substance resources. This project studies the adaptive capacity of the Northern communities with respect to the development of the natural resource extraction sites, permafrost degradation, and the availability of the subsistence resources.
Funding: Alaska NSF EPSCoR
Collaborators and Research Assistants: 20+ collaborators University of Alaska system and Pacific Northwest, Alaska Department of Fish and Game, University of Idaho, 9 UG research assistants, 1 MS Student. September 2014 – present.
Characterizing Anchorage’s Solid Waste with Techniques of Today and Tomorrow
This project uses Unmanned Areal Vehicles for aerial surveillance and machine learning for image processing to understand the solid waste composition at the transfer station and the landfill behavior. With a civil engineering collaborator, we explore computer vision to characterize the solid waste composition at the Anchorage Transfer Station. The LIDAR UAV missions collect information about the rate of fill and compaction. The project’s goal is to understand the landfill behavior and forecast its lifespan.
Funding: Municipality of Anchorage, Solid Waste Services.
Collaborators: Dr. Aaron Dotson, Maxwell Franklin, Masa Hu. 2017
Complex Networks
Looking at the world around us as a network of interconnected components can reveal unknown patterns of relationships. Using network view allows us to look at the system wide interaction patterns among the system components versus the traditional, reductionistic approach of one (discrete) data point at a time ( using term frequency metrics for example). The network tools incorporate the relationships between otherwise discrete pieces of information and often reveal structural relationships that are unseen by analyzing data through a keyhole of simple statistics.
This research topic combines a collection of projects that include:
- Analyzing the anthropology citation network to reveal paradigmatic clusters
- What religion drives Nigerian economy project analyzed the networks of publicly traded Nigerian companies and the relationships between their board members
- Analyzing a corpus of written text by constructing semantic concepts flows through the corpus of text
- Climate adaptation networks mapped the government and institutional relationships to understand the climate change mitigation in Alaska.
Web Yup'ik-ization.
This topic combines several smaller projects that includes: successful digitization of the Yup'ik dictionary, design and implementation of the basic agglutinative Yupik spell-checker. The future projects include using artificial intelligence to infer the language's grammar rules from the written corpus, redesign the static grammar rule set into a dynamic set with intelligent lookup, and a voice-to-text translator.
Collaborators and Research Assistants: Eric Somerville. January 2014 – current.
Inactive or temporarily suspended projects:
Inactive or temporarily suspended projects:
Massively parallel, asynchronous, decentralized, locally connected networks of simple power-aware sensor networks for environmental sensing in the Arctic regions.
This project studies the fundamental properties, engineering and deployment of massively parallel, decentralized, asynchronous, locally connected networks of simple components that can solve problems by collectively processing detected information . Examples of such networks are wide ranging from social networks, swarm robotics, sensor networks, nano-architectures, central processing units (CPUs) to human brain and gene expression networks. I strive to better understand what makes these networks "computationally capable". Note that both terms "computationally" and "capable" have a very loose meaning, since each term conveys different meaning in respect to which network context it is used. In particular, I am interested in how to design-and-build, control-and-program, and study-and-analyze these networks in order to solve a given problem.
In broader context, I strive to understand if these networks share similar underlying characteristics such as minimum connectivity of nodes, timing requirements, resilience to communication and component failures etc. Once we have better understanding of the network architectures, how can we exploit the network design to engineer networks to solve a given problem?
Funding: Arctic Domain Awareness Center, Department of Homeland Security
Collaborators and Research Assistants: Drs. Kenrick Mock, Aaron Dotson. United States Coast Guard, 5 UG research assistants August 2015 – 2017.
Brain Waves Based Authentication in Noisy and Distractive Environments
User authentication using brain waves in a controlled laboratory setting has been experimentally verified to have up to 99% accuracy. This is a Brain Computer Interface (BCI) usability study to determine the accuracy of user authentication and identification using brain activation patterns recorded from users who performed various mental tasks in noisy and distractive environments.
Collaborators and Research Assistants: Dr. Yasuhiro Ozuru, Frazer Tee, Cody McWilliams, Masa Hu. 2013 – 2017.
Neuromorphic Implementation of Human Inspired Computer Vision System for Object Recognition in Surveillance Video.
This project aims to design and implement a distributed version of a human inspired model of a computer vision system in a low-cost, power-aware hardware architecture with a near real-time performance to recognize various objects in the video. Intended use of this system is for environmental monitoring applications, intelligent unmanned vehicles, and driver assistance systems.
Collaborators and Research Assistants: Dr. Sam Sieward, John Harriss. September 2013 – 2015.
Past Research Projects and Topics (completed):
Past Research Projects and Topics (completed):
Robust anomaly detection with control feedback for irrigation canals for reconstruction efforts in Afghanistan.
I proposed, implemented and tested a machine learning based anomaly detection in an implementation of a sensor network for irrigation canal monitoring. The algorithm detected anomaly in the operation of individual sensors as well as across multiple sensors in the geo-spatial region. The proof of concept showed how an operational component can communicate these findings to the regional controller with instructions to correct the anomaly by opening or closing the flood gates or checking the sensors' functionality.
Funding: Afghanistan Reconstruction Trust Fund, World Bank Group. International Relief and Development (IRD).
Collaborators and Research Assistants: Dr. Aaron Dotson, Dr. Caixia Wang, Dr. Kenrick Mock, Matthew Devins. February 2014 – June 2015.
Information Processing in Two-Dimensional Cellular Automata.
I proposed, implemented and tested a novel information processing abstraction of collective computation in two-dimensional cellular automata. I designed statistically based filters to identify spatio-temporal sites with information significance and proposed a dynamic model to abstract the information transfer patterns in the lattices.
More information here: Cellular Automata
Funding: MARCO/FENA.
Advisor: Dr. Melanie Mitchell. Department of Computer Science. January, 2007 – July 2011.
Applied Robotics – Remotely Controlled Observatory.
I developed multi-process, multi-threaded, multi-hardware platform architecture to remotely control an astronomy observatory. The platform is designed for Internet-based remote sensing, asynchronous communication with hardware, and basic image processing. I validated the software design using formal methods applied on the deterministic finite automata based software design.
Advisor: Dr. Erik Bodegom. Department of Physics. 2000-2002.
Multiple Objective Optimization.
I combined Data Envelopment Analysis and evolutionary computations to design a strategy for the evaluation of Multiple Objective Optimization searches. I applied a newly developed approach to plan integrated circuit layout using a multi-objective optimization search.
Funding: Intel/NSF
Advisor: Dr. Garrison Greenwood. Department of Electrical and Computer Engineering. 2002-2003.
Evolutionary Computations Using Reconstructability Analysis.
I implemented an evolutionary algorithm with heuristic ordering of free variables to solve a multimodal function. I applied the Reconstructability Analysis on the objective multimodal function and assessed the information gain for the model's optimal ordering of variables.
Funding: NSF
Advisor: Dr. Martin Zwick. Department of Systems Science. January, 2004 – December, 2005.
Current and Formal Student Thesis:
- Undergraduate Research Thesis Advisor: Masa Hu. Shape Recognition Using Computer Vision: Classification of Growth Regimes in Chemical Gardens. University of Alaska Anchorage. Honors College. BS Computer Science. Awarded Spring 2017.
- MS Thesis Advisor: Maxwell Franklin. Scenario-Based Studies Using an Agent-Based Model to Investigate Impacts of Socio-Ecological Changes at the Kenai River: Implications for Fisheries and Salmon Processors. University of Alaska Anchorage. Department of Civil Engineering. MS Civil Engineering. Awarded Summer 2017.
- MS Thesis Committee: Matthew Ahlrichs. Sensor Network Lifecycle Assessment. University of Alaska Anchorage. Department of Civil Engineering. MS Civil and Environmental Engineering. In process – planned Summer 2017.
- MS Thesis Committee: Jason Kintner. Planning of optical networks. University of Alaska Anchorage. Engineering and Project Management Department. MS Project Management. Spring 2016.
- Undergraduate Research Thesis Advisor: Blake Romero. Reflections in the Data: Citation Network Analysis as a Method for Investigating the Development of Anthropological Theoretical Frameworks and the Composition of Social Theory. University of Alaska Anchorage. Honors College. BS Anthropology. Spring 2015.
- MS Thesis Advisor: Qandeel Fatima. Text Summarization. University of Northern British Columbia. Department of Computer Science. MS Computer Science. In process – planned Spring 2018.
- External Reviewer. Sherry Wolf. Homeless Young Adults Caring For Their Health. Advisor: Dr. Sharyl Toscano. MS Nursing Thesis Defense. April 8, 2014. Spring 2015
Google Sites
Report abuse