Thomas Petersen
- Account Type:
- member
- Location:
- , Denmark
- Member since:
- March 9, 2007
- Last Login:
- April 27, 2009
- Last Update:
- January 4, 2009
- Bio:
Skills
- Primary Skill:
- Advanced Programmer
- Secondary Skill:
- Intermediate Designer
- Tertiary Skill:
- Beginner 3D Artist
- Summary:
- Speciality is developing and training AI using Neural Networks that is trained using reinforced learning - trained by playing aginst itself with no human expert knowledge needed.
The core of the AI itself is of a number of weights used in the neural network. The strength of the AI depends on how well the AI is trained to play a particular game.
This method ensures that AI can be produced to play at many different strength-levels from beginner (almost no training of the weights) to expert (requires intensive training). The training is done offline, and runs very fast once trained.
The method is inspired by the work of Tesauro (www.research.ibm.com/massive/tdl.html) that has produced very stong AI for the game of Backgammon.
We have developed the method further to suit games that do not use a random generator (dice), e.g. board-games like Kalaha and Zanzibar Bao (Mancala variants).
Apart from this I have coded C/C++ for 20 years now and have some skills in developing assets using Blender
(www.blender.org/)
Recent Activity
No recent activity