Denning Sparse Distributed Memory 1989
<p>Dec 21, 2010 10:37 PM
 
from the American Scientist 77  1989.  Peter Denning
commenting on Pentti Kanerva's model of sparse memory.
 
SDM  =  sparse distrib memory.
 
I understand this.
see example on page  334
 
the task is to ASSOCIATIVELY STORE a particular CUE pattern and a DATA INPUT PATTERN,
so that when a similar cue appears in the future, the system will OUTPUT the data input pattern.
 
all patterns are 1000 bits  and the entire memory stores  1 million words each 1000 bits long.
 
the CUE pattern determines where the input pattern is stored (ie the addresses which will
be used to stored the input pattern)
 
Note that the total space is 21000 possible 1000 bit words.
 
first initialize the memory by randomly assigning a 1000 bit random word to each position in memory.
 
Each of the one million words has an address decoder that simultaneously determines whether that particular word
in memory is a match to the CUE word.  (It matches if the the cue word is within a fixed Hamming distance....
it must match inputs on at least  451 bits eg;  it so happens that 1 /1000 of total words in the space will match.
So,   1 thousandth of the words in the address register will be activated...  that tells the computer
 
in which 1000 words of the million words to store the input pattern.
 
then  each bit of the data input vector is compared to the contents of the memory at the bit position in the designated
storage locations  (the thousand retrieved by the cue vector).
Each bit there is incremented if it matches and decremented if it does not match. (the incrementing and decrmenting
is done to a "counter array" which stores these numbers.
 
 
 
 
 
 
 </p>