Natural Computation & Coadaptive Systems Lab Advanced
Research Computing Center
Phone: (407) 882-0313
R. Paul Wiegand is an Research Associate Professor at the
University of Central Florida's Institute for Simulation and Training. He conducts research in machine learning and
optimization, teaches for the Modeling & Simulation graduate program, and
directs the Advances Research Computing Center.
He also holds a secondary joint appointment with the Department of
Dr. Wiegand runs the Natural Computation & Coadaptive
Systems lab. His research interests primarily focus on methods of natural computation,
theory of coadaptive and coevolutionary computation, application of coadaptive
methods for multiagent learning, and high performance / high throughput
computing. More generally, he is interested in designing and applying effective
learning algorithms and representations for generating and modeling robust
heterogeneous, multiagent team behaviors.
Biologically Inspired Approaches for Reasoning on Complex Functional Networks
The NCCS lab has worked with geographers at the
University of Wyoming on grants from the National Geospatial Intelligence
Agency to investigate efficient methods to learn patterns of behaviors of
entities operating within a network that is geospatially embedded (e.g., people
walking along streets in a town). This includes investigation of robust
graph-based, probabilistic methods for matching behaviors of entities operating
on such networks under different types of uncertainty, which allows one to
recognize when two routes within a network are similar, even when those routes
share little explicit path information as well as a way to detect anomalous
activity patterns on the network in near real-time.
Coevolutionary Peer Instruction
The NCCS lab works with educational specialists and
instructors of introductory computer science courses at the Citadel, University
of South Florida, and Ramapo on NSF-funded work to examine machine learning
methods for extracting comparison geometries from student-problem interactions
for both categorization of student challenges, as well as development of
improved problem sets. These include
leveraging ideas from the theory of coevolutionary computation to to identify
programming concepts that are informatively easy or hard for students to
master. Our approach is different from
more traditional analyses such as item response analysis in the sense that we
consider Pareto dominance relationships within the multidimensional structure
of student-problem performance data rather than average performance measures.
This method permits us to uncover not just the problems on which students are
struggling, but the variety of difficulties different students face. The
results of our analysis not only have implications for how to scale up and
improve adap- tive tutoring software, but also have the promise of contributing
to the construction of a concept inventory for introductory programming.
RICHES Mosaic Interface
The NCCS lab consults with the Regional Initiative for
Collecting Histories, Experiences and Stories (RICHES) project from the UCF
Department of History by providing machine learning expertise and advice in
guiding the Connections feature of an archive of digitized historical
documents, imagery, and audio, the RICHES Mosaic Interface (MI). Connections uses a multi-criteria connections
algorithm to make item selection recommendations and browse through the MI. The
Connections algorithm allows researchers to examine a selected artifact and
nearest related items in the archive based on multiple criteria from the metadata
contained in the artifact of interest.
Nature-Inspired Coordinated Control Systems for Multi-Agent/Robot Systems
The NCCS lab has investigated the use of particle and
swarm inspired systems for multiagent control.
In particular, we have examined how complex, heterogeneous
multi-agent/robot systems can be constructed such that the growth of the
parameter space scales reasonably with the different types and specializations
within the team. This is useful for
robust, cooptimized learning methods.
Moreover, we have examined how the dichotomy of the nature of such
models to both predict simulated natural behavior and provide control
instructions to the underlying agents can complicate representation and machine
learning methods for behaviors. These
complications are even more sophisticated when team members include human
agents that are not under the direct control of the nature-based system.
High Performance Benchmarking for Practical, Distributed Applications
The NCCS lab has worked with a number of researchers both
to establish robust benchmarking methods for various practical, distributed
applications, as well as to implement computationally efficient practical
solutions to key engineering tasks.
These include benchmarks for distributed constructive simulation for the
DoD, as well as implementation of solvers for large-scale, sparse FEM/FVM
solvers for GPU, and hybrid GPU/distributed systems on applications relevant to
NASA and JPL.