 | Hello,
I'm studying the notes from my AI lectures. One thing is very confusing.
The professors tries to explain how would back-propagation work with hardware memory.
- So we have 2 by 2 layer structure
- 4 inputs A,B,C,D
- and the following diagram: http://i44.tinypic.com/351bxw2.png
u - denotes undefined.
The blocks with 4 segments represent 4 bit memory.
The algorithm states that its initial value is arbitrary set.
Then the first set of input is applied to the system. If output is correct the addressed undefined fields are marked as defined.
...I'm lost at this point
How is this diagram working? What is stored in the memory ? - Is it the weight vector ? If it is a weight vector how are these 4 bits combined with the only two bits of A,B?
Please any help will be appreciated. If you have seen similar diagram somewhere I could use the info to relate it to my problem.
| |
