Sriraam Natarajan
Post-Doctoral Research Associate
Department of Computer Science
6720 Medical Sciences Center,
1300, University Avenue,
University of Wisconsin Madison
Email: natarasr@biostat.wisc.edu
Post-Doctoral Research Associate
Department of Computer Science
6720 Medical Sciences Center,
1300, University Avenue,
University of Wisconsin Madison
Email: natarasr@biostat.wisc.edu
I am currently a Post-Doc at the Department of Computer Science in the University of Wisconsin Madison, working with Jude Shavlik and David Page. My research interests lie in the field of Artificial Intelligence and Machine Learning. More specifically, I am interested in the areas of Statistical Relational Learning, Reinforcement Learning, Bayesian Networks and Planning.
I completed my PhD in fall 2007 under Dr.Prasad Tadepalli in the School of EECS at Oregon State University.
Developing inference methods for accelerating inference in Statistical Relational Learning models. Working on inference methods based on counting and bounds propagation. The goal is to make inference feasible in SRL models. This will have a serious impact on the applicability of SRL models such as MLNs in several real-world problems. The lifted inference methods are developed for dynamic setting and hence can be used for activity recognition, longitudinal data modeling, Filtering, Monitoring etc.
Currently working on developing an "intelligent student". The basic idea in this project is to build a student that learns from the teacher using several different machine learning methods. My goal in this project is to use statistical relational learning models to specify prior knowledge to the agent (student) that employs the use of reinforcement learning to learn the optimal policy. We developed a preprocessing algorithm for performing faster inference in MLNs. We also worked on automating runs across different domains and worked on a layered approach to setting domain-independent parameters.
Currently, exploring the use of statistical relational learning models for Bio-Medical applications such as gene-fold prediction and health care monitoring of nursing home patients. These are longitudinal data and hence we model them as logical dynamic models. We are currently working on developing 2 different directions - first is to extend Tree-Augmented Naive Bayes Networks (TANs) to dynamic setting and learning predictive ILP rules using DTANs as features. The second approach is to extend Markov Logic Networks (MLNs) to dynamic setting and learn discriminative DMLN rules for prediction. We plan to evaluate the two models on datasets from nursing homes in Wisconsin.
Developing a model that combines ideas from Statistical Relational Learning and Hierarchical Plan recognition to build a Logical Hierarchical Hidden Markov Model (LoHiHMM) that is obtained from a model of the user’s work flow. It can then be unrolled to perform inference over the user’s goals based on observations.
Currently working on developing an intelligent assistant that can learn to assist a user in a rich environment. Formulated the problem of learning to assist as a POMDP with the hidden states being user’s goal structure and developed approximate techniques for solving the POMDP. The model uses a relational hierarchical structure which is unrolled into a Dynamic Bayesian Network (DBN) to track the user’s progress towards the goal. Evaluated the assistant on two game-like domains and demonstrated that the assistant can exploit prior knowledge effectively. Presently, working on formalizing the model and performing more experiments.
Designed a First-Order Conditional Influence language to enable the domain expert to write rules about the domain and developed learning algorithms for learning the parameters of these rules. Evaluated the algorithms on two domains: A synthetic domain to determine if the algorithms correctly capture the probability distributions and a folder prediction task to predict the folder of a document given its attributes. Currently working on automating the learning of these rules by using some prior knowledge.
Considered the problem of learning in the presence of time-varying preferences among multiple objectives. Proposed a method that stores a finite number of policies, chooses an appropriate policy for any weight vector and improves upon it. Verified the hypotheses empirically in two domains: a modified version of the Buridan's ass problem and a network routing Domain.
Sriraam Natarajan,Intelligent Assistants - A Decision-Theoretic Model: Effective Decision-Theoretic Assistance Through Relational Hierarchical Models. VDM-Verlag 2009
Sriraam Natarajan,Prasad Tadepalli, Thomas G. Dietterich and Alan Fern. Learning First-Order Probabilistic Models with Combining Rules . Annals of Mathematics and AI, Special Issue on Probabilistic Relational Learning 2009
Neville Mehta, Sriraam Natarajan, Prasad Tadepalli and Alan Fern. Transfer in Variable Reward Hierarchical Reinforcement Learning. Invited contribution to Inductive transfer in Machine Learning, Springer Verlag publications, 2008.
Alan Fern, Sriraam Natarajan, Kshitij Judah and Prasad Tadepalli, A Decision theoretic model of Assistance, Journal Of Artificial Intelligence Research (JAIR) – Submitted.
Sriraam Natarajan, Prasad Tadepalli and Alan Fern, A Relational Hierarchical Model of Decision-Theoretic Assistance ACM Transactions on Autonomous and Adaptive Systems– Under Review.
Jude Shavlik, Sriraam Natarajan Speeding up Inference in Markov Logic Networks By Preprocessing to Reduce the Size of the Resulting Grounded Network , IJCAI 2009. (Acceptance Rate 25.7%)
Sriraam Natarajan, Prasad Tadepalli, Gautam Kunapuli, Jude Shavlik Learning Parameters for Relational Probabilistic Models with Noisy-Or Combining Rule, ICML-A 2009
Kristian Kersting, Babak Ahmadi, Sriraam Natarajan. Counting Lifted Belief Propagation , UAI 2009
Sriraam Natarajan, Gautam Kunapuli, Ciaran O' Reilly, Rich Maclin, Trevor Walker, David Page, and Jude Shavlik ILP for Bootstrapped Learning: A Layered Approach to Automating the ILP Setup Problem , International Conference on Inductive Logic Programming 2009.
Sriraam Natarajan, Hung H.Bui, Prasad Tadepalli, Kristian Kersting, Weng-Keen Wong. Logical Hierarchical Hidden Markov Models for User Activity Recognition , International Conference on Inductive Logic Programming 2008. (Acceptance Rate 27%)
Sriraam Natarajan, Prasad Tadepalli and Alan Fern. A Relational Hierarchical Model of Decision-Theoretic Assistance. Proceedings of the International Conference on Inductive Logic Programming, (ILP 2007), Corvallis USA. (Acceptance Rate 25%)
Alan Fern, Sriraam Natarajan, Kshitij Judah and Prasad Tadepalli, A Decision theoretic model of Assistance, Proceedings of The International Joint Conference in Artificial Intelligence (IJCAI 2007) Hyderabad India. (Acceptance Rate 18%)
Sriraam Natarajan, Prasad Tadepalli, Eric Altendorf, Thomas G. Dietterich, Alan Fern and Angelo Restificar. Learning First-Order Probabilistic Models with Combining Rules. Proceedings of The 22nd International Conference on Machine Learning (ICML 2005) Bonn, Germany. (Acceptance Rate 27%)
Sriraam Natarajan and Prasad Tadepalli. Dynamic Preferences in Multi-Criteria Reinforcement Learning.Proceedings of The 22nd International Conference on Machine Learning (ICML 2005) Bonn, Germany. (Acceptance Rate 27%)
Sriraam Natarajan, Prasad Tadepalli Gautam Kunapuli and Jude Shavlik. Knowledge Intensive Learning: Directed vs. Undirected SRL Models. International Workshop in SRL 2009.
Rodrigo De Salvo Braz, Sriraam Natarajan, Hung Bui, Jude Shavlik, and Stuart Russell. Anytime Lifted Belief Propagation. International Workshop in SRL 2009.
Sriraam Natarajan, Irene Ong, David Haight, David Page, Vitor Santos Costa. Modeling Temporal Biomedical Data by SRL, ECML workshop on Bio-Medical Applications using SRL, 2008.
Sriraam Natarajan, Kshitij Judah, Prasad Tadepalli and Alan Fern. A Decision-Theoretic Model of Assistance - Evaluation, Extensions and Open Problems, AAAI 2007 Spring Symposium on Interaction Challenges for Intelligent Assistants, Stanford University, USA.
Sriraam Natarajan, Prasad Tadepalli and Alan Fern. Exploiting prior knowledge in Intelligent Assistants - Combining relational models with hierarchies, Extended Abstract in the Proceedings of the Dagstuhl Seminar on Probabilistic, Logical and Relational Learning - A Further Synthesis, (2007).
Sriraam Natarajan and Eric Altendorf. First Order Conditional Influence Language. Technical Report CS05-30-01 September 2005.
Alan Fern, Sriraam Natarajan, KshitijJudah and Prasad Tadepalli. A Decision theoretic model of Assistance, Modeling Others from Observations workshop in AAAI 2006.
Sriraam Natarajan, Weng-Keen Wong and Prasad Tadepalli, Structure Refinement in First Order Conditional Influence Language, Open Problems in Statistical Relational Learning, ICML 2006.
Neville Mehta, Sriraam Natarajan, Prasad Tadepalli and Alan Fern. Transfer in Variable Reward Hierarchical Reinforcement Learning. Inductive Transfer NIPS workshop 2005.
Lisa Torrey, Jude Shavlik, Sriraam Natarajan, Pavan Kuppilli and Trever Walker. Transfer in Reinforcement Learning via Markov Logic Networks. AAAI workshop on Transfer Learning for Complex Tasks 2008.
Hung Bui, Fedrico Cesari, Daniel Elenius, David Morley, Sriraam Natarajan, Shahin Saadati, Eric Yeh, and Neil Yorke-Smith. A Context-Aware Personal Desktop Assistant. Demonstrations track of Autonomous Agent and MultiAgent Systems, 2008.