Carlos "Greg" Diuk Wasser Department of Computer Science Rutgers University

Contact Me:

• Email: cdiuk AT cs.rutgers.edu
• Office Phone:  732-445-6185
• Office:  Hill 427 (RL^3 lab)
• Mailing Address:
  Hill Center - Busch Campus
  110 Frelinghuysen Road
  Piscataway, New Jersey 08855

Research:

I'm working with Michael Littman at the Rutgers Laboratory for Real-Life Reinforcement Learning (RL)^3.

I'm currently involved in a number of projects:

• Transfer learning
  • How can an agent "transfer" knowledge obtained while performing a previous task into a new and -somewhat- similar task?
• Efficient reinforcement learning
  • Exploring how factored state spaces, hierarchies and model-based frameworks can be combined to achieve efficient learning.
• Matching
  • Creating agents that can model and profit from the matching phenomenon observed in animals.
• Dynamic Bayes Net exploration
  • The objective is to speed up learning by identifying and exploiting independence relations among features that constitute a state space

RL publications:

2007. “Efficient Structure Learning in Factored-state MDPs”, Alexander L. Strehl, Carlos Diuk and Michael L. Littman. AAAI 2007.pdf

2007. “An adaptive anomaly detector for worm detection ”, John Mark Agosta, Carlos Diuk, Jaideep Chandrashekar and Carl Livadas. Second Workshop on Tackling Computer Systems Problems with Machine Learning Techniques (sysML-07). pdf

2006. “A Hierarchical Approach to Efficient Reinforcement Learning in Deterministic Domains”, Carlos Diuk, Alexander L. Strehl and Michael L. Littman. AAMAS’06. pdf

Workshop presentations:

2006. Invited Speaker at AAMAS Hierarchical Autonomous Agents and Multi-Agent Systems: “A Hierarchical Approach to Efficient Reinforcement Learning”.

2006. “Using Classifiers to Transfer Knowledge ”, Thomas J. Walsh, Carlos Diuk and Michael Littman. Presented at the New York Academy of Science Machine Learning Symposium.

2006. “Efficient exploration and learning of structure in factored-state MDPs ”, Carlos Diuk, Michael L. Littman, Alexander L. Strehl. Presented at NIPS Workshop “Towards a New Reinforcement Learning?”.

2005. “A Hierarchical Approach to Efficient Reinforcement Learning in Factored State Spaces”, Carlos Diuk, Michael L. Littman, and Alexander L. Strehl. Presented at the the 22nd International Conference on Machine Learning (ICML 2005), Workshop on Rich Representations for Reinforcement Learning, Bonn, Germany, 2005.

Previous publications:

2002. “Computer tools for reconstructing a genealogy”, Carlos Diuk and Enrique Tándeter. International Journal of History and Computing. Edinburgh University Press.

2001. “Computer tools for reconstructing a genealogy”, Carlos Diuk and Enrique Tándeter. XVth International Conference of the Association for History and Computing, Poznan, Poland.

Teaching:

Download full CV

Links:

Friends

• Rohan Fernandes.
• Miguel Angel Mosteiro.