Recurrent neural networks for learning mixed kth-order Markov chains

Many approaches for Markov chain construction determine only the parameters for the first order Markov chain. In this paper, we present a method for constructing mixed k(th) order Markov Chains by using recurrent neural networks. In terms of input length n, our method needs O(n) operations. We apply our method for classification on the Splice-junction Gene Sequences Database. Experimental results show that our method has less error rate than the traditional Markov Models.