We restrict our attention to some specific members of the above class. First, we assume
the hidden transfer function φh (∙) to be the identity function. Thus, the output of the
h-th hidden product unit of the j-th network module, denoted by jzh, reads as
2j
jzh = Φh(jneth) = ∏ χβn j=1,∙∙∙,J;h = 1,∙∙∙,H
(19)
n = 2 j-1
Second, we assume that each output unit, Ωsl (x, w)j =: yj ( j = 1,..., J), uses a non-
linear normalised transfer function ψj (∙) :
net j
ωsl (x, w ) j = Ψ j ( netj ) = b∖1j ) ------- j = 1,∙∙∙, J (-0)
Σ net j '
f=1
that resembles the Bradley-Terry-Luce model augmented by a bias unit b. j ^. net j is the
value of the weighted sum for the j-th output node given by
H 2j
(21)
netj = ∑Yh ∏ xβhn j = I,-,J
h-1 n-2 j -1
The choice of the output transfer function (20) is motivated by the goal to ensure that
the network outputs satisfy the conservation principle (Ledent 1985) that is enforced
from the viewpoint of origins if b^jj^ = bii.^ [origin constrained case] or from the
viewpoint of destinations if b.^ = t>^j^ [destination constrained case]. Ωsl (x, w)j may be
interpreted as probability of spatial interactions, conditioned on the output ∑H1 jzh of
the j-th network module.
With the two above specifications, our modular product unit neural network to model
origin constrained spatial interactions reads in a compact way as
∑ r h ι2i √■
(22)
⅛ (x, w )j≈ ⅛l J-H n------
ΣΣrh ∏ xfhh
j l-1 h l-1 n-2 j ,-1
13
More intriguing information
1. ‘Goodwill is not enough’2. Does adult education at upper secondary level influence annual wage earnings?
3. Peer Reviewed, Open Access, Free
4. The name is absent
5. Draft of paper published in:
6. On the Integration of Digital Technologies into Mathematics Classrooms
7. Implementation of Rule Based Algorithm for Sandhi-Vicheda Of Compound Hindi Words
8. The name is absent
9. The name is absent
10. Discourse Patterns in First Language Use at Hcme and Second Language Learning at School: an Ethnographic Approach