Table 1. Comparison of the resulting analytic interaction matrices for an autoassociative task with the two representations.

	{-1,+1} representation	{0,+1} representation
learning rule	$H_{ij}={\displaystyle \sum _q{a}_i^{(q)}{a}_j^{(q)}}(\mathrm{Hebb}‐\mathrm{rule})$	$M_{ij}={\displaystyle \sum _q(2a_i^{(q)}-1)a_j^{(q)}}$
symmetry	Hij = H	non-symmetric
neutralities in weight matrix	no neutral elements	can be neutral elements (“opposite to” zero vector elements)
main diagonal	always positive (if allowed)	can be negative or positive
structure	has a unique structure	many different realizations