Roni Khardon

Roni Khardon
Professor
Department of Computer Science
Luddy School
Indiana University, Bloomington

Research Interests and Projects: My interests are in the areas of Machine Learning and Data Mining, Artificial Intelligence, and Efficient Algorithms and my research explores theoretical questions, empirical questions, and applications. Most recently my work has focused on problems in two broad areas: Theory, Algorithms and Applications of Graphical models and Stochastic Planning and Reinforcement Learning and their Relation to Probabilistic Inference.

Please beware of phishing attempts: Multiple phishing attempts using my name have been circulating over the last few months. Typical examples inlcude fake job offers that ask you to communicate with me by text/phone or through an alternate non-IU email address - these are phishing attempts. Sophisticated variations and schemes are evolving. If you are reading this web page after seeing an odd offer it is likely a phishing attempt. Please help others by reporting these. Information about reporting phishing at IU is at: https://phishing.iu.edu/report/index.html

Recent News

6/2022: RSS 2024 paper: develops an efficient constant-time online update method for sparse GP which is effective for non-stationary data, with applications for Robotic Information Gathering.
Read the paper on arxiv.
5/2024: Congratuations! to Weizhe Chen who has successfully defended his PhD thesis.
3/2024: Our group's collaborative work, with Chanh Kieu from Earth and Astmospheric Sciences, is being presented at the American Metrological Society, Conference on Hurricanes and Tropical Meteorology, May, 2024: Link to talk abstract
11/2023: Explainable Models via Compression of Tree Ensembles, published in the Machine Learning Journal. The ideas developed in the paper have two outcomes: extracting explainable models from ensemble methods and a flexible tool for manipulating relational data structures that can be used in other algorithms.
7/2023: Our team is the Winner of the Probabilistic Track of the International Planning Competition held as part of ICAPS 2023.
See organizers' slides summarizing the results
We used a variant of the DiSProD planner (IJCAI 2023 paper).
Congratulations to Palash Chatterjee and Ashutosh Chapagain.
6/2023: IJRR paper : Adaptive Robotic Information Gathering via Non-Stationary Gaussian Processes. The paper develops a new kernel for modelling non-stationary phenomena, shows how this improves uncertainty quantification when used with Gaussian Processes, and how it imporves performance in Robotic Information Gathering.
Read the paper on arxiv.
6/2023: New NSF Grant will support research on planning/control, uncertainty quantification in machine learning and reinforcement learning, with some applications in robotics.
NSF award #2246261: RI: Small: Approximate Inference for Planning and Reinforcement Learning
5/2023: Upcoming IJCAI 2023 paper: develops DiSProD, a new planner for continuous and hybrid environments, with applications to controlling Robotic Systems. The planner works through Differentiable Symbolic Propagation of Distributions, and is a generalization of our top-performing SOGBOFA planner (that only handled discrete models).
Read the paper on arxiv.
Watch videos of field experiments showing DiSProD controlling a ground vehicle (Jackal) and Surface Vessell (Heron) on the DiSProD project page
5/2023: New manuscript: develops VIFO, a new approach for uncertainty quantification in deep neural networks. VIFO is based on a new variational approximation which combines the simplicity of non-Bayesian approaches with uncertainty quantification given by Bayesian approaches.
Read the paper on arxiv.