S. Lawrence, A. Tsoi, and C. Giles, "Noisy times series prediction using symbolic representation and recurrent neural network grammatical inference," tech. rep., University of Maryland, 1997.  Home/Search   Document Details and Download   Summary   Related Articles

....a graph is very deep. Because the recursive learning is derived from dynamical system in which state varies with nodes. Just as shown in Algorithm 2. 2, state transition from leaves to root is undertaken a long path, which is proven to be difficult in the case of recurrent neural network (see [55] [56]) Although the quantization of depth suitable for recursive learning is problem specific, one principal is to avoid generating very deep graphs. Quad tree can usually generate a tree with a certain depth, possibly, it can create very deep tree representation for an image. Instead, contour tree or ....

S. Lawrence, A. Tsoi, and C. Giles, "Noisy times series prediction using symbolic representation and recurrent neural network grammatical inference," tech. rep., University of Maryland, 1997.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC