The Community for Technology Leaders
Green Image
<p><b>Abstract</b>—The existence of principal interconnections useful in solving the proliferation problem in higher order Hebbian-type associative memories is introduced. Among all legal interconnections, we prove there exists a subset <it>T</it><sub><it>pr</it></sub> that carries more information than the others. Regardless of the network order <it>p</it>, the elements in <it>T</it><sub><it>pr</it></sub> are shown to be those interconnections <it>T</it> that fall within the range of</p><tf>$$\sqrt {m_s} \le \left| T \right| \le 2 \sqrt {m_s},$$</tf><p>where <it>m</it><sub><it>s</it></sub> equals the number of stored codewords. Memories that use only <it>T</it><sub><it>pr</it></sub> can maintain original generalization performance, using less than 50 percent of the total number of interconnections.</p>
Associative memory, redundant interconnections, Hebbian rule, convergence probability, synchronous updating, confidence interval.

J. Wang, "Principal Interconnections in Higher Order Hebbian-Type Associative Memories," in IEEE Transactions on Knowledge & Data Engineering, vol. 10, no. , pp. 342-345, 1998.
83 ms
(Ver 3.3 (11022016))