Part I
Physical View of Deep Learning
Let us get into Part I of this book, in which we study and understand machine
learning from a physics perspective. In this art, we will use physics language to
see how neural networks “emerge.” In general textbooks, neural networks used
in machine learning and deep learning are introduced as imitating the functions
of brain neural networks and neurons, and various neural network structures have
been improved for the purpose of learning and application. No particular physics
perspective is used there. As a result, many readers may suspect that, even if neural
networks have a physics perspective, they can only be interpreted as retrospective
interpretations of individual concepts. However, in Part I, from a physics point
of view, we will focus on how neural networks and their accompanying concepts
emerge naturally, and mainly look at the “derivation” of neural networks.
Incidentally, although we cannot say for sure, we imagine that the “derivation”
here seems to be something that was actually done implicitly in the minds of
researchers. This is our feeling after having read the papers written at the dawn
of deep learning research when the research subject was shifting from Boltzmann
machines to neural nets.
Of course, not all machine learning concepts are physically derived. We will
explain basic concepts such as what is machine learning, in physics language. If
you have learned a bit of physics, you can naturally feel the answer to the question
of how to perform and optimize the machine learning.
In addition, the concepts described below do not cover all of the concepts in
machine learning in an exhaustive manner, but focus only on the basic concepts
necessary to understand what machine learning is. However, understanding these
in the language of physics will help to physically understand machine learning that
is currently applied in various ways, and will also be useful for research related to
physics and machine learning described in Part II. Now, we shall give each chapter
a brief introduction as a destination sign for readers.
Chapter 2: Introduction to Machine Learning First, we learn the general theory
of machine learning. We shall take a look at examples of what learning is, what
is the meaning of “machines learned,” and what relative entropy is. We will learn
Physical View of Deep Learning
Let us get into Part I of this book, in which we study and understand machine
learning from a physics perspective. In this art, we will use physics language to
see how neural networks “emerge.” In general textbooks, neural networks used
in machine learning and deep learning are introduced as imitating the functions
of brain neural networks and neurons, and various neural network structures have
been improved for the purpose of learning and application. No particular physics
perspective is used there. As a result, many readers may suspect that, even if neural
networks have a physics perspective, they can only be interpreted as retrospective
interpretations of individual concepts. However, in Part I, from a physics point
of view, we will focus on how neural networks and their accompanying concepts
emerge naturally, and mainly look at the “derivation” of neural networks.
Incidentally, although we cannot say for sure, we imagine that the “derivation”
here seems to be something that was actually done implicitly in the minds of
researchers. This is our feeling after having read the papers written at the dawn
of deep learning research when the research subject was shifting from Boltzmann
machines to neural nets.
Of course, not all machine learning concepts are physically derived. We will
explain basic concepts such as what is machine learning, in physics language. If
you have learned a bit of physics, you can naturally feel the answer to the question
of how to perform and optimize the machine learning.
In addition, the concepts described below do not cover all of the concepts in
machine learning in an exhaustive manner, but focus only on the basic concepts
necessary to understand what machine learning is. However, understanding these
in the language of physics will help to physically understand machine learning that
is currently applied in various ways, and will also be useful for research related to
physics and machine learning described in Part II. Now, we shall give each chapter
a brief introduction as a destination sign for readers.
Chapter 2: Introduction to Machine Learning First, we learn the general theory
of machine learning. We shall take a look at examples of what learning is, what
is the meaning of “machines learned,” and what relative entropy is. We will learn