24 28th February, Wednesday

I've started gathering experimental results today for varying the number of hidden units per layer, I used [32, 48, 64, 96, 128, 192, 256] units to start with, but will be reducing these in my next test.

I will also be looking to rerun some of the more promising experiments on BC, and want to try another technique referenced in my thesis, whereby you do the same method of learning but apply it to a fixed size network and vary the connections, dropping each of those each time. I have some reasons why I think this could be an exciting experiment and will be outlining them in writing later in the week.