Meeting 10-07-2012

Discussion

• Thesis:
• Structure stays as it is (maybe some sections will still be moved later-on)
• Add research question about computational complexity of both algorithms
• Transposition tree: recreate split tests from stored samples if perfect recall is used (but otherwise do not copy them)
• Plots: 
• Include data of up to 10^3 roll-outs
• Regret: change y-axis to log scale
• Thesis: 
• Separate plots for the variations of each algorithm
• Plots to compare best variants
• Plots for variations with similar curves
• Transposition tree: related work is function approximation / generalization with decision / regression trees in reinforcement learning
• Transposition tree for light-dark domain:
• Changes to light-dark domain:
• Continuous state space with a squared region for the goal
• Discrete actions, corrupted by some Gaussian noise
• Belief space represented as a multivariate Gaussian centered around the agent's actual location (x,y) with three parameters: mean(x), mean(y), standard deviation

Planning

• Hand-in of chapters 2 and 3 this Friday July 13
• Next meeting: Monday July 15