More OCR work, and the start of some ‘pure’ AI

Another busy month getting some OCR work up to production standard – there’s a big difference between ‘basically working’ and ‘industrial grade’.  It’s taken a lot of work, but is there now.

I have written a test harness that allows bulk amounts of test data to be tested and retested, in as little as 1 second each time, and in an automated way after each round of training.  A bit of ‘infrastructure’ and formal testing goes as long way at this stage.

Success rates in the test suite are now up to 100%, and the real prodution environment beckons.

In other news, I have some work to do on a ‘pure’ AI project – some software to take one player’s turn in a ‘full information’ two-player game, and associated research on scoring algorithms.  The idea is to see what level of ‘intelligent behaviour’ emerges from just the raw computation, given enough time and CPU cores.  It sounds like my idea of fun 🙂