MACHINE LEARNING TECHNIQUES - LASA
MACHINE LEARNING TECHNIQUES - LASA
MACHINE LEARNING TECHNIQUES - LASA
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
153<br />
To simplify the description of the information retrieval, one, often, uses:<br />
1<br />
η = (6.66)<br />
K<br />
6.9.3 Retrieval Phase<br />
Each neuron updates its state as if it were a single neuron with the threshold activation function<br />
⎛⎛ ⎞⎞ ⎧⎧1 if z ≥ 0<br />
xj = θ⎜⎜<br />
wijxi<br />
⎟⎟, θ( z)<br />
=⎨⎨<br />
⎝⎝ i ⎠⎠ ⎩⎩-1 if z < 0<br />
∑ (6.67)<br />
x ∈−<br />
n<br />
j<br />
{ 1,1}<br />
The activity of the neurons can be updated either synchronously or asynchronously. The retrieval<br />
procedure is iterative and consists of the following:<br />
Initialization:<br />
r<br />
At time t=0, inject a pattern to the net x µ<br />
random pattern), i.e. set the state of all the network’s neuron to<br />
Update:<br />
Synchronous Update:<br />
(possibly a noisy version of a stored pattern or a<br />
( 0 ) ,..., ( 0)<br />
x = x x = x<br />
µ µ<br />
1 1 K K<br />
⎛⎛<br />
xj( t) = θ aj( t− ) ∀ aj t− = ⎜⎜ wijxi<br />
t−<br />
⎝⎝ i<br />
( 1 ) j ( 1) ( 1)<br />
⎞⎞<br />
⎟⎟<br />
⎠⎠<br />
∑ (6.68)<br />
All neurons compute their activation and update their state simultaneously following:<br />
Asynchronous Update:<br />
One neuron at a time computes its activations and updates its state. The sequence of<br />
selected neuron may be fixed or random:<br />
⎛⎛<br />
xj( t) = θ aj( t− ) aj t− = ⎜⎜ wijxi<br />
t−<br />
⎝⎝ i<br />
( 1 ) ( 1) ( 1)<br />
( ) x ( t )<br />
x t = −1 ∀ i≠<br />
j<br />
i<br />
i<br />
⎞⎞<br />
⎟⎟<br />
⎠⎠<br />
∑ (6.69)<br />
© A.G.Billard 2004 – Last Update March 2011