Professor Tim Timothy Oates
Professor · Tenured
Department of Computer Science and Electrical Engineering
College of Engineering and Information Technology
He/Him/His/Himself
About
Dr. Tim Oates is an Oros Family Professor of Computing at the
University of Maryland Baltimore County. He received B.S. degrees in
Computer Science and Electrical Engineering from North Carolina State
University in 1989, and M.S. and Ph.D. degrees from the University of
Massachusetts Amherst in 1997 and 2000, respectively. Prior to coming
to UMBC in the Fall of 2001, he spent a year as a postdoc in the
Artificial Intelligence Lab at the Massachusetts Institute of
Technology. In 2004 Dr. Oates won a prestigious NSF CAREER award.
Research interests
My general research is in the areas of machine learning and artificial intelligence, with a focus on discovering latent structure in data. My early work, which I continue to this day, looked at learning grammatical structure of formal languages. Today that has branched into learning hierarchical models of time series. My work on time series has taken me into physiological data, with work on detecting seizures and predicting the need for blood transfusions for patients with brain injuries.
Another interest is statistical natural language processing, with a focus on extracting knowledge from text. Most recently that has produced algorithms for finding causal explanations that link newswire stories, and for characterizing the certainty of knowledge extracted by systems viewed as black boxes (something that is crucial for downstream consumers of that knowledge).
Another significant thread is metacognition, where the focus is on developing methods that allow learners to determine, on their own, when learned knowledge is no longer effective, hat the problem might be, and how to address it. This can lead to much more robust intelligent systems.
More recently, my work has expanded into reinforcement learning, a field that considers how to choose actions so as to maximize a scalar reward through time. I've explored ways of using humans to provide feedback to RL systems so that they can learn more quickly, and of allowing multi-agent teams to coordinate more effectively using relational (graph-based) representations of states and relational reinforcement learning.
Due to recent advances in large language models, my students and I have explored a number of topics in that space. One is to combine my past work on metacognition to imbue LLMs with the ability to reflect and automatically refine their problem solving approaches. I'm also exploring the role of schemas in expertise, helping LLMs use declarative stores of knowledge to become more effective experts.
Teaching interests
I very much enjoy teaching a wide variety of courses, though recently my focus has been on teaching the machine learning course that I developed some years ago due to the demand among the students, both graduate and undergraduate. My approach to the classroom is highly interactive, with a preference for chalk as opposed to slides. I hope to continue expanding the scope of the courses I have experience teaching, which currently spans topics as diverse as compilers, discrete math, artificial intelligence, robotics, and data structures.
I've recently become involved in teaching data science at both the graduate and undergraduate levels. The data science course had been taught a few times by different people in very different ways. I applied my extensive data science consulting experience to build a course that uses modern tools to attack common problems. The course is now a good blend of theoretical and applied content.
Education
-
Ph D, Computer Science
— University of Massachusetts Amherst (2001) Grounding Knowledge in Sensors: Unsupervised Learning for Language and Planning
- MS, Computer Science — University of Massachusetts Amherst (1997)
- BS, Computer Science — North Carolina State University (1989)
- BS, Electrical Engineering — North Carolina State University (1989)
Publications
- Controlling Swarms by Visual Demonstration 2016
- Encoding time series as images for visual inspection and classification using tiled convolutional neural networks 2015
- On-line signature verification using symbolic aggregate approximation (SAX) and sequential minimal optimization (SMO) 2014
- Time warping symbolic aggregation approximation with a bag-of-patterns representation for time series classification 2014
- Finding Story Chain in Newswire Articles Using Random Walks 2013
- Imaging time-series to improve classification and imputation
- Pooling SAX-BoP Approaches with Boosting to Classify Multivariate Synchronous Physiological Time Series Data
Grants and Contracts
- Human-in-the-Loop Model Implementation for Efficient Deep Reinforcement Learning 2018
- Radio Frequency Signal Classification Using Deep Neural Networks 2017