Shiwali Mohan

About

I am a doctoral candidate in Computer Science and Engineering at The University of Michigan Ann Arbor. I have been working with Professor John Laird on the Soar Cognitive Architecture in the Michigan AI Lab since 2009.

I am interested in situated language for intelligent agents and the relationship of language processing (comprehension, acquisiton, generation and interaction) with other cognitive capabilities such as perception, procedural and declarative knowledge, and learning.

My past work has focussed on cognitive agents that can learn to navigate new, complex environments. We have looked at agents that start with minimal background knowledge about an environment and build relevant, relational structures using reinforcement learning in the Infinite Mario RL domain.

I am from India, where I recieved an undergraduate degree in Instrumentation and Control Engineering from University of Delhi. After four years of struggling with control equations and diffrential calculus, I ended up loving programming and other geeky, Computer Science-y things. I worked with Yahoo! India as a software engineer on their Strategic Data Solutions team for time, before deciding to denounce the real world, and the money that comes with it and became a graduate student.

I am available at shiwali.at.umich.dot.edu and shiwali.dot.mohan.at.gmail.dot.com. My curriculum vitae dated 2012-05-17 is here.

Research

Grounded Natural Language

I am involved with the grounded language acquisition project at Michigan. Apart from the Soar group at University of Michigan, the team also includes APRIL at Michigan and SoarTech.

I am currently studying comprehension of sentences by exploiting non-linguistic context such as perceptual information, procedural knowledge and semantic information.

I have designed and implemented an interaction model for agents instantiated in Soar cognitive architecture that allows limited mixed-initiative interaction with an instructor. The agents can then derive generally applicable procedural knowledge from a history of interactions (available in episodic/semantic memory of the agent) using situated explanation.

Related Software: SBOLT, Soar Agents, Soar

Reinforcement Learning

In the past, I was investigating the capabilities of Soar-RL in complex environments. I designed, implemented and analyzed reinforcement learning agents for Infinite Mario. I also implemented modular reinforcement learning for Soar cognitive architecture that allows the agent to learn multiple MDPs at the same time.

Related Software: MarioSoar

Talks
Modular Reinforcement Learning In Soar, 31st Soar Workshop, June 2011, Ann Arbor Michigan
Towards an Architecture for Learning with Instruction, 31st Soar Workshop, June 2011, Ann Arbor Michigan
Reinforcement Learning in Infinite Mario, 30th Soar Workshop, June 2010, Ann Arbor Michigan
Learning Background Knowledge thorugh Instruction, 30th Soar Workshop, June 2010, Ann Arbor Michigan
Learning to Play Mario, 29th Soar Workshop, June 2009, Ann Arbor, Michigan

Miscellaneous
On Generating Grounded Language in Cognitive Architecture, EECS 590, Natural Language Processing, April 2011, University of Michigan
Relational Reinforcement Learning in Infinite Mario, Preliminary Examination, September 2010, University of Michigan
Classification of Executed and Imagined Motor Movement EEG Signals, EECS 545, Machine Learning, December 2009, University of Michigan
Towards a Resource Aware Scheduler in Hadoop, EECS 589, Advanced Computer Networks, December 2009, University of Michigan

References

John Laird, Keegan Kinkade, Shiwali Mohan, and Joseph Xu. Cognitive Robotics Using the Soar Cognitive Architecture. In Proceedings of the 8th International Cognitive Robotics Workshop, 2012.
[ bib | .pdf ]

Shiwali Mohan and John Laird. Exploring Mixed-Initiative Interaction for Learning with Situated Instruction in Cognitive Agents. In Proceedings of the 26th AAAI Conference on Artificial Intelligence, 2012. (Extended Abstract).
[ bib | .pdf ]

Shiwali Mohan and John Laird. Learning Actions and Action Verbs from Human-Agent Interaction. In Proceedings of the 26th AAAI Conference on Artificial Intelligence, 2012. (Extended Abstract).
[ bib | .pdf ]

Shiwali Mohan and John Laird. Situated Comprehension of Imperative Sentences in Embodied, Cognitive Agents. In The AAAI 2012 Workshop on Grounding Language for Physical Systems, 2012.
[ bib | .pdf ]

Shiwali Mohan and John Laird. An Object-Oriented Approach to Reinforcement Learning in an Action Game. In Proceedings of 7th the Artificial Intelligence for Interactive Digital Entertainment Conference, AIIDE, 2011.
[ bib | http | .pdf | Abstract ]

Shiwali Mohan and John Laird. Towards Situated, Interactive, Instructable Agents in a Cognitive Architecture. In Papers from the 2011 AAAI Fall Symposium Series, 2011.
[ bib | http | .pdf | Abstract ]

Shiwali Mohan and John Laird. Relational Reinforcement Learning in Infinite Mario. In Proceedings of the 24th AAAI Conference on Artificial Intelligence, AAAI, 2010. (Extended Abstract).
[ bib | http | .pdf | Abstract ]

Shiwali Mohan and John E. Laird. Learning to Play Mario. Technical Report CCA-TR-2009-03, Center for Cognitive Architecture, University of Michigan, Ann Arbor, Michigan, 2009.
[ bib | http | Abstract ]

Niladri Chatterjee and Shiwali Mohan. Discovering Word Senses from Text using Random Indexing. In Proceedings of the 9th International Conference on Computational linguistics and Intelligent Text Processing, CICLing, 2008. Best Paper Award.
[ bib | http | .pdf | Abstract ]

Niladri Chatterjee and Shiwali Mohan. Extraction-Based Single-Document Summarization Using Random Indexing. In Proceeding of the 19th IEEE International Conference on Tools with Artificial Intelligence, ICTAI, 2007.
[ bib | http | .pdf | Abstract ]