Page 76 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

64 CHAPTER 3 Third Gen AI as Human Experience Based Expert Systems

simply “Me-Too” copycat, we wish to go beyond the AI, ANN with supervised
learning least mean squares (LMS) cost function and backward error propagation-
algorithm; we consider NI, BNN, unsupervised learning minimum free energy
(MFE), cost function, and backward MFE propagation (Fig. 3.4).
Newtonian learning synaptic weight matrix [W] among neurons under the
isothermal. Its easier to begin to look at the cost function equilibrium at minimum
Helmholtz free energy function after the internal brain energy E subtracted the
unusable thermal noise energy T o S

HY ¼ E T o S[ 0 (3.7)
d½WŠ vH
¼ (3.8)
dt v½WŠ
Control steering wheel Lyaponov convergence of learning of
2

dH vH d½WŠ vH vH vH

¼ ¼ ¼ 0 (3.9)
dt v½WŠ dt v½WŠ v½WŠ v½WŠ
. .
Langevin equation of the car momentum P ¼ mV, with tire-road friction coef-
.
ficient f, car-body aerodynamic fluctuation force FðtÞ

FIGURE 3.4
NI Human Target Recognition must be able to separate binary figure and ground under
the dusk dim light far away. This could be any simple ambiguity figures for computational
simplicity. The idea of NI in BNN for the survival is manifested clearly in “Tigress” &
Ground “Tree.” In contrast, the supervised cost function is LMS AI based on ANN
becomes ambiguous of binary figure and ground Least Mean Squares (LMS) cost function

. . 2 . . 2
F G ¼ G F could not separate to run away for the survival of the

species due to the switch of the algebra sign. However, higher order of moment expansion
of MFE can separate the tiger and tree in the remote dime light for the survival of Homo
sapiens.