Mitsubishi Electric Corp claims to have developed a prototype neural network computer that uses logical as well as intuitive processes to achieve higher accuracy in recognising and interpreting handwriting and spelling. The prototype uses its intuitive function to take stabs at recognising a word, then compares the result with a dictionary stored in the computer. The process is repeated several times until an acceptable answer is found. The company claims that the computer has a 95% accuracy rate recognising English handwriting fed into it by a scanner. Mitsubishi also says that it had developed an optical neural network integrated circuit and notes that this will lead to faster processing on smaller systems when it can be produced in greater numbers – but practical applications for it are likely to be years away.
