I'm working with 2D torroidal or rectangular SOMs on the order of 50x50 and larger. Obviously, this creates 2500 comparisons in order to find the node with weights closest to a training example. Considering I have 400,000+ training examples, this becomes computationally demanding. Does anyone have any experience with optimizing this process? I've considered BSP trees, but I don't see a way to ensure that I'll find the closest node without searching all of them. -Kirk