Igor V. Karpov


Igor V. Karpov ()
1616 Guadalupe, Ste. 2.408
Austin, TX 78701 USA
Tel.: +1-(512)-553-6131

Igor V. Karpov

I am a Ph.D. student in the Department of Computer Science at the University of Texas at Austin. I started at UT in 2004 and switched to the PhD program after completing my MSCS in 2006. Before that, I graduated from Rice University with a BS in Computer Science in 2003. I am originally from Pushchino, Moscow Region, Russia and I have lived in the United States since 1994.


My research centers around agents that can learn from humans. In paritcular I am studying how machine learning methods, such as neuroevolution, can be extended to learn effectively from human training modalities such as advice, examples, and task shaping via environment and reinforcement.

I am a member of the Neural Networks Research Group, and I am variously involved with the following projects:

The Maze mod of OpenNERO OpenNERO - OpenNERO is an open source software platform for research and education in Artificial Intelligence. It includes a growing number of environments and algorithms designed to demonstrate and compare different approaches. OpenNERO is based on previous NERO game project and uses the NEAT/rtNEAT library for real-time neuroevolution. [Karpov'08].
Human game traces collected in an Unreal Tournament 2004 level Human-Like Behavior in Games - The 2K BotPrize competition challenges the participants to design an human-like artificial player for a first-person action game (Unreal Tournament). The task of designing a bot capable of passing a Turing-like (though non-linguistic) test while interacting with human players has inspired several new research directions. As part of the UT^2 bot, I am interested in learning from collected traces of human game behavior.[Karpov'12].
Human-Assisted Neuroevolution Human Assisted Machine Learning We build and experimentally compare different ways to leverage the complimentary strengths of human input and machine learning techniques in order to design complex behaviors for games and other sequential decision tasks. [Karpov'11].
  GIVE Challenge - the goal of the GIVE challenge is to measure and improve the state of the art in Natural Language Generation systems. The systems in the challenge face the task of guiding human players through an interactive 3-D environment to solve a puzzle. They are judged on both on the objective and subjective quality of the instructions they generate.


Older and off-topic projects

Seminars and Reading Groups


I have been the Research Educator for several iterations of the Computational Intelligence in Game Design stream of the Freshman Research Initiative. The goal of the two-semester course sequence is to introduce freshmen and sophomores interested in computer science research to the area of artificial intelligence in games.





Professional Experience



Author: Igor V. Karpov This page is Valid XHTML 1.0 Strict and Valid CSS Last modified: Saturday, 19-Jul-2014 12:13:31 CDT