This example illustrates the operation of the simplified recognition device shown in Figure 3.5, using a 3x3 input retina. The training set has four instances, representing the simplified letters `H' and `L'. The test set consists of two distorted characters (see Figure 3.6).
Figure 3.5: An Example Architecture Using RAM Elements.
Figure 3.6: The Training and Test Sets.
After training the RAM units have the following filling:
| Address = Pattern | ||||||||
| 000 | 001 | 010 | 011 | 100 | 101 | 110 | 111 | |
| RAM11 | 0 | 0 | 0 | 0 | 1 | 1 | 0 | 0 |
| RAM12 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 |
| RAM13 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 |
| RAM21 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 |
| RAM22 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 |
| RAM23 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 1 |
If the test patterns are supplied to the network the system gives the following response: For test pattern 1 the H-discriminator gives a response of `1' and the L-discriminator gives a `2'. For the second test pattern the H-discriminators give a `2' and the L-discriminator gives a `0'. If the threshold is set to two the distorted `L' and `H' are both correctly recognized.