Table 1.

Overview over all learning rules discussed in this paper

Class		Two Inputs
Type	Name	Output: υ	Rule: dw1/dt	Convergence	Comment
CL	S&B	w 0 x 0 + w 1 x 1	u1 υ′	divergent	stimulus substitution
	ISO	w 0 u 0 + w 1 u 1*	u 1 v′	x 0 = 0, unstable!	symmetric, diff. Hebb
	ICO	w 0 u 0 + w 1 u 1*	u′ 0 u 1	x 0 = 0	diff. Heterosyn.
	ISO3	w 0 u 0 + w 1 u 1*		x 0 = 0, (R = 0)	3-factor diff. Hebb
RL	TD	w 1 x 1		〈δ〉 = 0	critic only, no actions
	TD-r	w 0 u 0 + w 1 u 1*		〈δ〉 = 0, x 0 = 0	mixed Hebb + diff. Hebb
	TD-a			〈δ〉 = 0	recursive rule

Class		Filter Bank		Summary
Type	Name	Output: υ	Convergence	General Comment	Use
CL	S&B	not applicable	not applicable	only of historical relevance	−
	ISO		x 0 = 0 for certain	unstable! As optimal h i are unknown,	−
			unknown h i	convergence cannot be guaranteed, input control
	ICO		x 0 = 0	robust, heterosynaptic, input control	+
	ISO3		x 0 = 0, (R = 0)	robust, input control	+
RL	TD	serial compound rep.	〈δ〉 = 0	robust, output control	+
	TD-r	not tested so far	〈δ〉 = 0, x 0 = 0	undesirable mix of Hebb & diff. Hebb	−
	TD-a	not tested so far	〈δ〉 = 0	not tested so far	?

The asterisk * depicts identical equations within one column