Extension to Delta Rule already studied before?

[pmr@di.fc.ul.pt]

Dear connectionists,

while working on my Ph.D. thesis, I've been using neural nets that apply 
the 'Delta Rule'. Nevertheless, for some instances of the problems I've been 
addressing, the 'Delta Rule' does not seem appropriate: it promotes weights 
that do not differ substantially. While trying to overcome this property, I've 
came across the following rule:

w_ij = w_ij + a_i * (t_j - a_j) * LR * w_ij

where...
   w_ij - weight of the connection from input neuron i to output neuron j
   a_i  - activity of input neuron i
   t_j  - target activity for output unit j
   a_j  - activity of output neuron j
   LR   - learning rate

Has this learning rule been studied before? Are there references ?


Many thanks for your attention,
kind regards,
pedro