Page 77 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

2. Third Gen AI 65

.
dP . .
¼ fP þ FðtÞ (3.10a)
dt

. .
FðtÞ$Fðt0Þ ¼ 2k B f dðt t0Þ (3.10b)
Human vision system has multiple layer deep feature extraction and can detect
single photon at warm body temperature against the quantum mechanics uncertainty
principle and Shannon statistical information theory.
Definition: the probability is defined by Komogorov triplet: [closed set, a
metric, and a measure].
Wide Sense Analyticity: There is a differentiable energy function of the scalar
energy landscape. Given the set of initial and boundary conditions, each corresponds
to a gradient descent result.
Wide Sense Causality: The feedforward neural network (NN) can take the
gradient descent from the initial labeled boundary conditions to reach an unlabeled
local minimum.
DL in multiple layers (about 10e100) that propagate nodal excitation data
through the connection weight matrix [W j,i ] between j-th & i-th neuron processor el-
ements (about millions per layer). By creating multilayer neural networks, we can
accommodate multicut in the classifier domain for a broader class of machine
learning that can reduce the false alarm rate (FAR). The reason why it is necessary
is not due to the nuisance false positive rate (FPR); but rather the detrimental false
negative rate (FNR) that can delay the early convergence solution, or not sick for
“seeing physician” (cf. R. Lipmann, Intro NN, IEEE ASP Magazine (1986)).
Corollary: a rule-based decision making is inefficient, and not yet smart. Proof:
Since it takes another rule to break this rule, and so on, so forth, the chain of rules
becomes an open set that like bifurcation cascade cannot be normalized into the
probability; but an open set of possibilities called FMF by Lotfi Zadeh, and when
using the bifurcation dynamics to chaos by Walter Freeman.
Theorem: Experience-based expert system (EBES) is wiser gaining from the
other experience. For example, we have equipped with identical full collision sensor
suite FMF, NewtoneLangevin inertial motion-road friction coefficient FMF, and
time (seconds) and location (10th feet) GPS FMF. We consider the scenario of iden-
tical AVs facing the cardinal rules: “stopping at a red light.” “red light can only right
turn” “pedestrian first (including wearing uniform police who is not walking).”
Result: M&S identical AV have learned to glide over slowly in the midnight at
the Gobi desert.
1. Cloud Big Databases, for example, billion smartphones, in positive enhance-
ment loops. Let the machine statistically generate all possible FMFs with
different gliding distance in triangle shape (with a mean and a variance). It is
associated with different brake-stopping FMF distances for the 1000 cars to