Utkarsh Soni

I am a Computer Science Ph.D. student at the School of Computing and Augmented Intelligence, Arizona State University. I work in the Yochan research group directed by Prof Subbarao Kambhampati. My current research focuses on enabling preference specification, preference incorporation, and explanation generation in human-in-the-loop reinforcement learning systems. I have also worked on generating personalized explanations for robot's actions in multi-user settings; and using data visualization as a modality for effective human-AI communication.

Prior to my PhD, I completed my M.S. in Computer Science at ASU, where I worked in the Vader Lab under Prof Ross Maciejewski on graphical layout perception. I also worked on an urban employment tool meant for government decision makers to help them analyze multiple aspects of employment data in their region.

Email  /  CV  /  Google Scholar  /  Twitter  /  Linkedin  /  Github

profile photo
Publications
Towards customizable reinforcement learning agents: Enabling preference specification through online vocabulary expansion
Utkarsh Soni, Sarath Sreedharan, Mudit Verma, Lin Guan, Matthew Marquez, Subbarao Kambhampati
NeurIPS workshop on Human in the Loop Learning (2022)

We propose PRESCA (PREference Specification through Concept Acquisition), a system that allows users to specify their preferences in terms of concepts that they understand to a RL agent.

Bridging The Gap: Providing Post-Hoc Symbolic Explanations for sequential decision making problems with inscrutable representations
Sarath Sreedharan, Utkarsh Soni, Mudit Verma, Siddharth Srivastava, Subbarao Kambhampati
ICLR, 2022

A method for generating contrastive explanations, in terms of user-specified concepts, for an agent in a sequential decision making setting.

Not all users are the same: Providing personalized explanations for sequential decision making problems
Utkarsh Soni, Sarath Sreedharan, Subbarao Kambhampati
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2021)

[video]

End-to-end adaptive explanation generation system that learns different types of users, interacts with them and adjusts it's explanation on the fly.

Why? Why not? When? Visual Explanations of Agent Behaviour in Reinforcement Learning
Aditi Mishra, Utkarsh Soni, Jinbin Huang, Chris Bryan
In Proceedings of IEEE Pacific Visualization 2022

Providing a visual analytics interface to question and thus gain trust in an autonomous agent's decision.

Integrating Planning, Execution and Monitoring in the presence of Open World Novelties: Case Study of an Open World Monopoly Solver
Sriram Gopalakrishnan*, Utkarsh Soni*, Tung Thai, Panagiotis Lymperopoulos, Matthias Scheutz, Subbarao Kambhampati
* indicates equal contribution
IntEX workshop at ICAPS, 2021

An adaptive agent that plans online to adapt to novelty introduced during stochastic multi-agent games

Feature-directed Active Learning for Learning User Preferences
Sriram Gopalakrishnan, Utkarsh Soni
XAIP workshop at ICAPS 2019

An active learning based method to learn user's preferences in sequential decision making tasks.

The Perception of Graph Properties in Graph Layouts
Utkarsh Soni, Yafeng Lu, Brett Hansen, Helen Purchase, Stephen Kobourov, Ross Maciejewski
Eurographics Conference on Visualization (EuroVis) 2018

Modeling human's perception of graph properties in different graph layouts via a large scale user study

Same stats, different graphs: Exploring the space of graphs in terms of graph properties
Hang Chen, Utkarsh Soni, Yafeng Lu, Vahan Huroyan, Ross Maciejewski, Stephen Kobourov
IEEE Transactions On Visualization And Computer Graphics, 2021

We show how summary statistic of graphs can be misleading and can be the same across widely varying graphs by a visual analytic tool


Website code from Jon Barron