Sequential Decision Making Under Uncertainty (MDP, POMDP)
Markov Decision Processes (MDPs) and Partially Observable Markov Decision Processes (POMDPs) provide a flexible mathematical framework for modeling decision making when uncertainty is present in the system. My research primarily focuses on solving MDPs and POMDPs with continuous observation and action spaces using sampling-based techniques, such that the proposed algorithms not only have provable theoretical convergence guarantees but also are computationally efficient and can be deployed in real autonomous systems.
- Sparse tree search optimality guarantees in POMDPs with continuous observation spaces.
Lim, M. H., Tomlin, C. J., & Sunberg, Z. N. (2020). IJCAI-PRICAI 2020 (To appear). [arXiv]
- MDP & POMDP algorithms for continuous observation and action spaces.
Algorithms to solve MDPs and POMDPs with continuous observation and action spaces.
Learning-Based Perception with Model-Based Control and Planning
In many realistic scenarios, simple and well-understood dynamics models and planning techniques are sufficient for control, such as navigation for model cars or quadcopters. Rather, the burden is on the vision and perception components that require learning. My research focuses on efficiently combining deep-learning based perception with traditional control and planning techniques, such that the agents can navigate in a priori unknown environments.
- Quadcopter 3D Visual Waypoint Navigation.
In this line of work, we focus on implementing and extending the Visual Waypoint Navigation to quadcopters using deep-learning based perception with robust optimal control techniques, such that the agent is able to navigate in unknown 3D environments.
Robotics & Artificial Intelligence
nuro.ai (Jun. 2020 - Sep. 2020)
Incoming summer intern, working with the control and planning team under SWE/research role for self-driving car.
IMC Financial Markets (Jan. 2016 - Aug. 2016)
As a summer intern, I modeled inter-index and inter-expiry implied volatility correlations through on-line updates with weighted linear regression. With the model, I proposed a correction to the pricing model to factor in implied vol. correlations to improve estimated trading gains up to $1,000/day.
As a winter intern, I learned theoretical and technical concepts of quantitative trading, such as valuation and quote allocation, through mock trading and mini programming projects using Python's pandas library.
- Creating analogs of thermal distributions from diabatic excitations in ion-trap-based quantum simulation.
Lim, M. H., Yoshimura, B. T., & Freericks, J. K. (2016). New J. Phys. 18 043026. [Journal] [arXiv]
- On the Chromatic Number of R4.
Exoo, G., Ismailescu, D., & Lim, M. (2014). Disc. Comput. Geom. 52(2): 416. [Journal]
Stochastic MCMC Simulation
Harvard University - Advisor: Prof. Samuel Kou (2016 - 2018)
Computational Condensed Matter
Georgetown University Physics REU - Advisor: Prof. James Freericks (2015)
Harvard University - Advisor: Prof. Amir Yacoby (2015)
Geometric Graph Theory
Indiana University, Hofstra University - Advisor: Prof. Geoffrey Exoo, Prof. Dan Ismailescu (2012 - 2014)