Conclusions | The Future |
---|---|
* Interacting with a real-world environment,
some variables (ie opponent) are not under experimental control
* Rating method found that uses all available information to obtain a maximum likelihood estimation * Good and bad found amongst both humans & robots * Tron Learning, as a system, shown * Human Learning shown in individual humans * Coevolving agents and humans is possible (also see TD-Leaf, Baxter et.al. 98) |
* Use Statistical Rating as Fitness Function
But: * Newton-Raphson involves inverting NxN matrix, a few times. So: * Glickman (1999) (Incremental) Parameter Estimation w/ Error Also: * Multi-Dimensional models? * Pareto? |