I'll give a short talk about my research into effective methods for modelling and optimizing policies in large spatial domains where parallel actions are needed for every point in space. The example domain I am focusing on is forestry planning with the complicating influence of  stochastic, spatial disturbance processes such as the Mountain Pine Beetle infestation in BC. 

I'll go over some approaches we have tried so far using policy gradient reinforcement learning.  I will describe my current, ongoing attempts to find useful, easy to sample policy descriptions using graphical models.  And I will describe a number of important pieces of the puzzle we do not yet know how to do and elicit discussion and advice on possible solutions.