METHOD AND DEVICE FOR PATTERN RECOGNITION BY NEURAL NETWORK

    公开(公告)号:JPH0765165A

    公开(公告)日:1995-03-10

    申请号:JP21138693

    申请日:1993-08-26

    Applicant: IBM JAPAN

    Abstract: PURPOSE:To improve the recognition rate of the pattern recognizing device which uses the neural network. CONSTITUTION:The section (0,1) of output of the neural network is equally divided into M (M: integer larger than 2) and the number or frequencies of correct answer and wrong answer pattern data included in an (i)th section [(i-1)/M, i/M] are denoted as mu1i and mu0i (where i=1 to M). At this time, when this netowrk grants an output included in the (i)th section for unknown pattern data, the likelihood P1i that this pattern is in this category is stored as an equation P1i=(mu1i+1)/(mu1i+mu01+2) and a likelihood conversion table to be outputted. Then a likelihood converter inputs a value included in the (i)th section [(i-1)/M, i/M] when it is outputted from the neural network and outputs P1i as what is called normalized likelihood.

Patent Agency Ranking