On the pattern recognition of noisy subsequence trees

In this paper, we consider the problem of recognizing ordered labeled trees by processing their noisy subsequence-trees which are "patched-up" noisy p ortions of their fragments. We assume that we are given H, a finite diction ary of ordered labeled trees. X* is an unknown element of H, and U is any a rbitrary subsequence-tree of X*. We consider the problem of estimating X* b y processing Y, which is a noisy version of U. The solution which we presen t is, to our knowledge, the first reported solution to the problem. We solv e the problem by sequentially comparing Y with every element X of H, the ba sis of comparison being a new dissimilarity measure between two trees, whic h implicitly captures the properties of the corrupting mechanism ("channel" ) which noisily garbles U into Y. The algorithm which incorporates this con straint has been used to test our pattern recognition system yielding a rem arkable accuracy. Experimental results which involve manually constructed t rees of sizes between 25 and 35 nodes, and which contain an average of 21.8 errors per tree demonstrate that the scheme has about 92.8 percent accurac y. Similar experiments for randomly generated trees yielded an accuracy of 86.4 percent.