|
Brian J. Taylor
Graduate Student |
[Research] [Publications] [Recent CV]
|
About Me |
I am in my second incarnation as a graduate student. I graduated in 1999 from West Virginia University (WVU) with an M.S. in electrical engineering, but my focus was on software engineering. I investigated the safety-critical aspects of adaptive systems like neural networks, and how we could apply software engineering techniques like Verification and Validation to them. After the MSEE I went to work at the Institute for Scientific Research, Inc. (ISR) (now a part of the West Virginia High Technology Consortium Foundation), a small WV research company. For ISR I was involved with several projects on intelligent flight control for the NASA Dryden Flight Research Center, NASA Ames Research Center, and the NASA IV&V Facility. I continued studying engineering practices on adaptive systems, leading research teams, and finally developing two guides to help other engineers and scientists working in this area. I decided that I wanted to be a better researcher and scientist so I went back to grad school to pursue a PhD in computer science. I am now a student at the University of Massachusetts Amherst working in the Knowledge Discovery Laboratory under David Jensen. My interests have evolved and I conduct research in the area of machine learning, specifically relational learning and causal knowledge discovery. As an engineer, I still have a love for systems development and much of my work tries to bring engineering and science together. |
|
|
|
|
Current Research Interests |
Causal Knowledge Discovery I am part of a KDL team working to create a system that enables causal knowledge discovery. Our first approach was the identification of quasi-experimental designs (QEDs) from static, observational data. QEDs are traditionally found through manual, time-consuming means. Our automated system is able to find large numbers of possible QEDs and then reduce that number through analysis of dependencies in the data, generating a list of highly likely candidates for good QEDs. Our current work is a larger system that will identify a larger set of potential experimental designs including QEDs. The software can then automatically apply these designs and build a causal model over the data set through an iterative process. Peer Production/Collaborative Systems I have developed a small research platform to investigate peer production and collaborative sensing systems called Photobase. These are systems where people come together to generate or gather content that is then used and shared by the community. Wikipedia is an example. Photobase allows us to experiment while participants use the system so we can evaluate what influences and effects their behavior and levels of participation. The Photobase design and experimental controls enable strong causal inference. For example I have already found that when participants view the collaboration as a competition, they actually participate less frequently than those who do not. The study also suggests that for any participatory system that includes areas infrequently traveled will either need to rely on a very large and carefully selected participant list or the use of coordination to guarantee coverage. Relational Learning The first area I began to research at UMass was the area of machine learning, specifically relational learning algorithms and graphical models. I've been investigating the relational probability tree (RPT) and the relational dependency network (RDN) and looking at how different characteristics of a data set can bias the learning algorithms. Biases like relational autocorrelation and degree dependency can influence relational learning algorithms and cause them to believe there is knowledge within the data that is not present. So far I have worked on:
|
|
|
|
|
Past Research Interests |
Verification and Validation of Adaptive Systems I've investigated the techniques necessary to use adaptive systems, primarily neural networks, in highly critical applications such as commercial jetliners, space exploration, and power generation. Neural networks are an approach that no longer receive as much attention, but very little has been done on them to make them practical. Would you trust a neural network in your car that augmented your braking ability? Do you know for sure that as the network learns or when it compensates that it is safe and won't cause you to suddenly veer out of control? What are the verification and validation techniques we need to develop and use to make sure neural network technology is safe so that it can be used in systems that directly impact human lives or expensive equipment? Techniques that I have explored include formal proofs of neural network safety, mathematical and statistical approaches to study neural network safe adaptation, and development of engineering practices like testing techniques. |
|
|
|
|
Publications |
Thesis Photobase - A Research Platform to Investigate Peer Production and Collaborative Sensing Systems. Regressive Model Approach to the Generation of Test Trajectories. Books Methods and Procedures for the Verification and Validation of Artificial Neural Networks. Guidance for the Verification and Validation of Neural Networks. Journal Articles Independent Verification and Validation of Neural Networks Developing Practitioner Assistance. Automated Generation of Test Trajectories for Embedded Flight Control Systems. Characterization of Poly(phenylsilsesquioxane) Thin film Planar Optical Waveguides. Conference Papers Building Better Participatory Sensing Systems. Automatic Identification of Quasi-experimental Designs for Discovering Causal Knowledge. Relational Data Pre-processing Techniques for Improved Securities Fraud Detection. Rule Extraction as a Formal Method for the Verification and Validation of Neural Networks. A Geometric Rule Extraction Approach used for Verification and Validation of a Safety Critical Application. Weaving it All Together - A Methodology for the Verification and Validation of Adaptive Neural Networks. Rule Extraction From Dynamic Cell Structure Neural Network Used in a Safety Critical Application. Verification and Validation of Neural Networks: A Sampling of Research in Progress. Evaluation of Regressive Methods for Automated Generation of Test Trajectories. Regressive Model Approach to the Generation of Test Trajectories. Polymer Waveguide Cointegration with Microelectromechanical Systems (MEMS) for Integrated Optical Metrology. Characterization of Poly(phenylsilsesquioxane) (PPSQ) for Planar Integrated Optical Waveguide Applications. Polymer Guided Wave Integrated Optics: An Enabling Technology for Micro Opto Electro Mechanical Systems. Tech Reports Methods and Procedures for the Independent Verification and Validation of Neural Networks. Introduction to the Development of Methodologies for the Independent Verification and Validation of Neural Networks. Toward Reliable Neural Network Software for the Development of Methodologies for the Independent Verification and Validation of Neural Networks. |