explore more-complex scenarios to include large volumes of data from which it is difficult to completely
model the entire world, and hence, statistical methods
are used, such as DL.
The challenge for most complex scenarios is
that the a priori information is unknown, the data
collected is unknown, and the objective is unknown.
Although many techniques are exploratory, the
discovery of some attributes infers relationships.
For example, evolutionary approaches mimic an
understanding of genetic diversity that allows for
mutations and adaptations. If a reasonable solution is
found from clustering, then some unknown classes are
revealed, leaving a set of data still unclassified. In other
cases, partial learning results explain some details.
Another approach is more propositional, leveraging
the symbolic and probabilistic combinations such as a
Gaussian mixture model based on known classes.
Figure 2. Three Waves of AI.
Name Device Approach to Data Analysis
Symbolic Logical statements First order logic, truth tables
Probabilistic Graphical models Bayesian statistics, conditional probabilities
Connectionist Artificial neural networks Computational model of linked statistical error
gradient minimization Multilayer artificial neural networks (DL)
Analogistic Support vector machines Pattern recognition via distance computations
in feature hyperspace Kernel methods
Evolutionary Genetic algorithms Competitive random variations for discovery
of survival adaptations Genetic programming
Possibilistic Fuzzy inference systems Expansion of classic logic to accommodate
ambiguous partial truths Evidential reasoning
Table 5. Types of ML.
Waves of Artificial Intelligence