Algodoo

(somewhere something is specifically setting scene.my.i5 to 0.2, and despite my best efforts to find it, I could not. Instead, I replaced all instances if i5 with i50, which is only a half fix, because now there's a weird varible name. But it functions as it should have been now.)

This synthetically weighted neural network recognizes eight patterns in a 3x3 scalar grid. To change the value of one of the scalars in the grid, click it. The eight recognized patterns in the middle of the black box, the brightest green cell represents its most confident guess, while the brightest blue or darkest red represents its least confident. All the weights in the net are one of two values, described as such (0 actually represents -1 to make each value the same size in text):

100100100 left vertical
010010010 middle vertical
001001001 right vertical
111000000 top horiontal
000111000 middle horizontal
000000111 bottom horizontal
001010100 topright diagonal
100010001 topleft diagonal

These 8 convolutions are then fed into an image generator which attepts to recreate the image based just on the values the 8 convolutions returned.