stopped training did not indicate unequivocally the stopping point, and thus, the
determination of H and δ. This is an issue for further research.
(i) Residuals of the Modular Product Unit Neural Network Model 1Ωsl (x,w)
80
60
-60
-80
(i1) Absolute Residuals (tj -1 ∩sl (x,w)) ; ordered by the size of tj
(i2) Relative Residuals (tj - 1∩sl(x,w)}/ tj ; ordered by the size of tj
(ii) Residuals of the Two-Stage Neural Network Model Approach ΩLonstr ( x,w )
(ii1) Absolute Residuals (tj - ΩLonstr(χ,w)] ; ordered by the size of tj
(ii2) Relative Residuals (tj - ΩLonstr (χ,w)]/ tj ; ordered by the size of tj
(iii)Residuals of the Origin Constrained Gravity Model τ grav
(iii2) Relative Residuals (tj - τg’av (x,w)]/ tj ; ordered by the size of tj
Figure 4: Residuals of the Modular Product Unit Neural Network 1 Ωsl , the Two-
Stage Neural Network Approach Ω't and the Origin Constrained Gravity
Model Predictions τ igra
ij
(iii1) Absolute Residuals (tj - τjaa (x,w)) ; ordered by the size of tj
30