Ph.D., Computer Science, Northwestern University, 2019
M.S., Computer Science, Northwestern University, 2011
B.S., Computer Engineering, Penn State University, 2009
Dr. Matthew (Matt) McLure is interested in spatial representations and processing, especially in the geospatial and sketching realms, and in machine learning techniques that can learn from few examples. He is also interested in cognitive architectures, knowledge representation, and multimodal HCI, where natural language works in concert with a shared workspace like a white board or a map. Ultimately, he wants to create personalized AI collaborators that can help us ideate, design, and solve problems - keeping up without getting in the way.
Since he joined SIFT in 2019, Dr. McLure has worked on the Clic project, as part of DARPA’s Communicating with Computers program, and on Project7 for the intelligence community. In these two projects, he has primarily developed user interface components that support deictic reference in a dialog system, gene set analyses, named-entity resolution, comparative analysis, and map-based interaction. He has also been involved in the OpenMIND project for DARPA’s SAIL-ON program, where he has worked on a hypothesis/experimentation framework, an ensemble of visual classifiers, and visualization/analysis tools to help us see what OpenMIND sees.
Dr. McLure did his graduate work in the Qualitative Reasoning Group at Northwestern University, where he wanted to create an AI that could learn to draw and draw to learn. His research, funded by AFOSR, focused on qualitative spatial representations, and on machine learning techniques built on analogical matching (approximate labeled graph matching). His approaches targeted learning properties that are suitable for low-shot, rich-data problems like the recognition of sketched objects or geographical concepts. He also came up with and implemented a novel constraint satisfaction technique for drawing novel prototypical examples of the structured symbolic concepts that the system had learned, or novel variations of them - a step toward drawing to actively learn.
While at Northwestern, Matt worked on the open-domain sketch understanding system CogSketch and on the Companions cognitive architecture, where he enabled large-scale, long-lived automated experimentation and multimodal interaction.
Before becoming enamored with AI, Matt worked at Intel Corporation, helping design inter-core communication logic and radiation-hardened components.