Research interests
- Reinforcement Learning
- Machine Learning
- Probability Theory.
Dr. Kohar's research focuses on the development of a novel framework for sequential decision-making under uncertainty, known as Bayes-adaptive semi-Markov decision processes (BA-SMDP). This framework emphasizes the critical role of timing in decision making—a factor well-recognized in intelligent beings such as animals and humans but often overlooked in the machine learning community.
A central challenge in this research is the exploration-exploitation trade-off: whether an agent should explore the environment to discover larger rewards or exploit its current knowledge about the rewards' locations. Dr. Kohar's model addresses this trade-off by enabling agents to learn about potential unknown rewards, the dynamics of the environment, and the optimal timing of actions while maximizing total rewards.
The BA-SMDP framework can be recast as a partially observable semi-Markov decision process (POSMDP), facilitating the development of a novel algorithm called ChronosPerseus. This algorithm leverages importance sampling and recent advancements in GPU computing to provide rapid and efficient solutions.
Dr. Kohar's innovative approach to decision-making in complex, uncertain environments, which incorporates timing considerations and manages the exploration-exploitation trade-off, has significant implications for both theoretical research and real-world applications. This work provides a robust foundation for further study and practical implementation.
Short biography
Dr. Kohar is an Assistant Professor in the Mathematics and Computer Science Department. He received his BSc (Hons) degree in Mathematics and Statistics, with a focus on Applied Mathematics, from McMaster University. In 2014, he completed his MSc in Mathematics at the Royal Military College of Canada (RMC) in Kingston, under the supervision of Dr. W. J. Hurley and Dr. J. Brimberg. Dr. Kohar then pursued and earned his PhD in Mathematics at RMC, focusing on developing the novel Bayes-Adaptive Semi-Markov Decision Process, under the guidance of Dr. F. Rivest and Dr. A. Gosselin.
Since 2013, Dr. Kohar has been teaching undergraduate courses in discrete mathematics, probability theory, statistics, and calculus at RMC. In 2016, his first textbook, "Basic Discrete Mathematics: Logic, Sets, and Probability," was published by World Scientific and has since been used to teach discrete mathematics to students in the Faculty of Social Sciences and Humanities.
Selected list of publications
- Kohar, R. (2016) Basic Discrete Mathematics: Logic, Set Theory, and Probability, World Scientific. (732 pages.) DOI: 10.1142/9851
- Rivest, F., and Kohar, R. (2020) A New Timing Error Cost Function for Binary Time Series Prediction. IEEE Transactions on Neural Networks and Learning Systems. 31:1, pp. 174-185. DOI: 10.1109/TNNLS.2019.2900046
- Hurley, W. J., Brimberg, J., Kohar, R. (2014) Examining a Gambler’s Claims: Probabilistic Fact Checking and Don Johnson’s Extraordinary Blackjack Winning Streak. Chance. 21:1, pp. 31-37 DOI: 10.1080/09332480.2014.890867
- Rivest, F., Kohar, R., Amadou Boukary, N. (2014) Learning to Predict Events On-line: A Semi-Markov Model for Reinforcement Learning NIPS 2014 Workshop: Autonomously Learning Robots, 12 December 2014.
- Shoucri, R. M., Kohar, R. ESVPR and Heart Failure. World Congress on Medical Physics and Biomedical Engineering, IFMBE Proceedings, 39:180-183, 2012. DOI: 10.1007/978-3-642-29305-4_49
- Shoucri, R. M., Kohar, R. Criteria for Study of Heart Failure derived from ESPVR. 34th Annual International Conference of the IEEE EMBS, 5586-5589, 2012. DOI: 10.1109/EMBC.2012.6347260