Hopfield Model – Continuous Case
The Hopfield model can be generalized using continuous activation functions.
More plausible model.
In this case:
where is a continuous, increasing, non linear function.
Examples
( )
+
=
= ∑
j ij j i
i
i
g u g W V I
V
β βgβ
( ) ]
1,1[
e e
e u e
tanh u u
u
u ∈ −
+
= ββ − −−ββ β
( ) ] [ 0 1
1
1
2
,
u e
g
u∈
= +
− ββ
Funzione di attivazione
ß > 1 ß = 1 ß < 1
-1 +1
( )
x tanh( )
xf = β
Updating Rules
Several possible choices for updating the units :
Asynchronous updating: one unit at a time is selected to have its output set
Synchronous updating: at each time step all units have their output set
Continuous updating: all units continuously and simultaneously change their outputs
Continuous Hopfield Models
Using the continuous updating rule, the network evolves according to the following set of (coupled) differential equations:
where are suitable time constants ( > 0).
Note When the system reaches a fixed point ( / = 0 ) we get
Indeed, we study a very similar dynamics
( )
+
+
−
= +
−
= ∑
j ij j i
i i
i i
i
V g u V g w V I
dt dV
β
τ
βτ
iτ
idV
idt ∀ i
( )
ii g u
V = β
( )
j ij ij
i i
i
u w g u I
dt
du = − + ∑
β+
τ
Modello di Hopfield continuo (energia)
Perché è monotona crescente e .
N.B.
cioè è un punto di equilibrio
( )
( )
02 1 2
1
2
1
≤
′
−
=
−
=
− +
−
=
− +
−
−
=
∑
∑
∑
∑
∑
∑
∑
∑
−dt u du
g
dt du dt dV
I u V
dt T dV
dt I dV dt
V dV dt g
V dV T dt V
T dV dt
dE
i i
i i
i i i
i
i i
j j
ij i
i
i
i i
i i
i j i
ij ij
j i
ij ij
β
β
τ τ
gβ
τ
i> 0
0
0 ⇔ =
= dt
du dt
dE i
u
iThe Energy Function
As the discrete model, the continuous Hopfield network has an “energy” function, provided that W = WT :
Easy to prove that
with equality iff the net reaches a fixed point.
( ) ∑
∑ ∑ + ∑ ∫ −
−
=
−i i i
i j i
V j
i
ij
V V g V dV I V
w
E
0i 12
1
β≤ 0
dt
dE
Modello di Hopfield continuo (relazione con il modello discreto)
Esiste una relazione stretta tra il modello continuo e quello discreto.
Si noti che :
quindi :
Il 2o termine in E diventa :
L’integrale è positivo (0 se Vi=0).
Per il termine diventa trascurabile, quindi la funzione E del modello continuo diventa identica a quello del modello discreto
( )
i( ) ( )
i ii
g u g u g u
V =
β=
1β ≡ β
( )
ii
g V
u = 1
−1β
( ) V dV
g
i V
i
∑ ∫
i −0
1
1β
∞
→ β
Optimization Using Hopfield Network
§ Energy function of Hopfield network
§ The network will evolve into a (locally / globally) minimum energy state
§ Any quadratic cost function can be rewritten as the Hopfield network Energy function. Therefore, it can be minimized using Hopfield network.
§ Classical Traveling Salesperson Problem (TSP)
§ Many other applications
• 2-D, 3-D object recognition
• Image restoration
• Stereo matching
• Computing optical flow
i i
i j
i
i j
ij V V I V
w
E = −
∑ ∑
−∑
2 1