[comp.ai.neural-nets] BAM vs. Hopfield net

shawnd@coho.ee.ubc.ca (Shawn Day) (07/13/90)

I have recently been reading about Kosko's Bidirectional Associative
Memories (BAM's) [1]. The author claims that the Hopfield network
is a special case of the BAM with the two sets of representational
units compressed into one. It seems to me, however that the
Hopfield network is the most general case, allowing complete
interconnectivity between units. The BAM appears to be a
special case of the Hopfield network with restricted connectivity.

If this is indeed the case, then I fail to see the distinction
between the two network architectures, and the BAM presents nothing
new. Has this topic been previously discussed, or does anyone
have an opinion on the subject? Or am I missing something obvious?

Thanks in advance for any feedback.

[1] Kosko, Bart. "Bidirectional Associative Memories", IEEE Transactions
on Systems, Man, and Cybernetics, Vol. 18, No. 1, 1988.

+---------------------------------------------------------------------------+
|  Shawn Day                UBC VLSI Salmon Net              (604) 264-0024 |
|  6609 Lime Street                                        shawnd@ee.ubc.ca |
|  Vancouver, British Columbia              ...!ubc-vision!ee.ubc.ca!shawnd |
|  Canada V6P 5V7                  shawnd@ee.ubc.cdn%ean.ubc.ca@RELAY.CSNET |
+---------------------------------------------------------------------------+