Page 79 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

2. Third Gen AI 67

networking, stress management (relaxing, walking, no alcohol); which are six
different initial boundary conditions that might be applied with WSEP. The
classifier in the associative memory (AM) in the hippocampus is called learning.
The adjective “Deep Learning” can refer to large databases and multiple layer
structured hierarchical learning to achieve a higher-level abstraction of how to
structure the multiple layers. It is possible sometime to overfit in a subtle way, so
the ANN becomes “brittle” outside the training set, for example, Ohlson [10].
This architectures development can be viewed as emulation of the brains ar-
chitecture which has been molded by evolution for the survival of the species.
5. Big Data Analyses (BDA): BDA has been formed by the Internet Industrial
Consortium. With the Cloud, it provides a great experimental resource to
explore algorithms that could not be studied using conventional computer re-
sources. Google’s cofounder Sergey Brin sponsored the algorithm AI AlphaGo.
He was surprised by the intuition, the beauty, and skills displayed by AlphaGo.
As a matter of fact, Google Brain AlphaGo Avatar beat Korea grandmaster Lee
Se-Dol in Chinese Go game, 4:1, as millions watched in real time on Sunday
March 13, 2016 on World Wide Web. This accomplishment has demonstrated
Alan Turing’s definition of AI that it cannot tell if the other end is human or
machine. Now six decades later Alpha technology can beat humans at most
games using self-training. Likewise, FaceBook has trained 3D color blocks
image recognition which will eventually provide an age and emotional-
independent face recognition up to 97 years old. Andrew Ng at Baidu used
YouTube as a data source and discovered to his surprise that the favorite pet of
mankind to be cat, not dog! Speech pattern recognition capability of BDA by
Baidu has utilized the MPD computing based on the classical ANN with su-
pervised deep learning (SDL). A program called Deep Speech outperforms
HMMs, using Big Data that can share all possible initial and boundary condi-
tions. It can be used to generate user information creating more data for business
in a positive feedback loop.

2.3 FUZZY MEMBERSHIP FUNCTION (FMF AND DATA BASIS)

Zadeh has spent much of his career teaching us that it is possible to implement open
set logic in engineering applications where membership is not binary [1,0], but
instead a distribution of values on (0,1). Likewise, UC Berkeley colleague Walter
Freeman computed in brain neurophysiology deterministic chaos dynamics to sup-
port open set possibility solutions. Such applications can occur in engineering appli-
cations which are formulated in terms of positive feedback loops. The spaces are
possibility spaces rather than probability spaces. When trying to determine a
move in chess or Go, we explore a possibility space of potential moves rather
than making an assessment of the most probable next move. Possibility is also an
useful alternative to explore large collections of data that encapsulate human
behavior. This possible membership concept is important to exploration of large
data as which often dont have definitive membership relations when partial analysis