196
13 Epilogue
of artificial intelligence at that time, there were expert systems and IBM’s Deep
Blue. 2 While doing various searches using the Google webpage (which was already
there at that time), I learned that neural networks were terribly interesting, but gave
up because I couldn’t understand the backpropagation method. On the other hand,
after entering high school, I began to find physics interesting, and then in 2015 I got
a PhD in particle physics and became a postdoc researcher. Then big news came in
2016 that Google’s artificial intelligence had defeated humans at Go. 3 In response
to the news, I began study and research while exchanging messages with one of the
authors of this book, Akinori Tanaka. When I studied machine learning for the first
time, I realized that mathematical tools were similar to physics, so I could easily
understand ideas and applied them. As I studied further, I understood that some
ideas were imported from physics. 4 The research went well and was formed into
several papers. In this book, physics researchers explain deep learning and related
fields from the basics to the application from that point of view, and in fact, this is
the path I followed.
My specialty is lattice QCD, and it was no surprise to discover that a person
in the field is one of the developers of Jupyter Notebook. 5 This is because if you
are doing physics using a computer, you have to visualize a large amount of data.
The WWW, the core technology of today’s Internet, was developed at the European
Nuclear Research Organization (CERN). 6 Also, the preprint service called arXiv 7
was created by a particle physicist. 8 At the meta level, physics supports the next
generation of technology. I am glad if the readers who have studied physics and deep
learning through this book will theoretically solve the mystery of generalization of
deep learning and clarify the relationship between physics and machine learning.
At the same time, I would be happy as well if readers will be responsible for the
next generation of technological innovation that gave us the WWW and Jupyter
Notebook.
2 Deep Blue was a supercomputer that defeated the world chess champion Garli Kasparov in 1997.
3 Deep Mind’s Alpha Go defeated Lee Sedol.
4 This is about the Boltzmann machines that triggered the huge progress in deep learning.
5 This person is Fernando Pérez from University of Colorado at Boulder. He obtained a PhD in
Lattice QCD, and was a student of Anna Hasenfratz who is famous in the field.
6 The actual development was done by Tim Berners-Lee, an information scientist who was there,
but I don’t think it would have been done without the accelerator, which is a machine that generates
a lot of data by itself, or without a strong motivation to understand nature. It’s also not hard to see
how the hacker spirit of the physicist community (in the original sense) could make this system
free to use.
7 The arXiv, https://arxiv.org is a web archive service for researchers to upload research preprints
on physics and mathematics.
8 Paul Ginsparg, a prominent physicist whose name is known in the important Gisparg–Wilson
relation in lattice QCD, first started the server for the arXiv. The preceding service was founded by
a physicist, Joanne Cohn. (I thank Shinichi Nojiri, Professor of Nagoya University, for providing
this information.)
13 Epilogue
of artificial intelligence at that time, there were expert systems and IBM’s Deep
Blue. 2 While doing various searches using the Google webpage (which was already
there at that time), I learned that neural networks were terribly interesting, but gave
up because I couldn’t understand the backpropagation method. On the other hand,
after entering high school, I began to find physics interesting, and then in 2015 I got
a PhD in particle physics and became a postdoc researcher. Then big news came in
2016 that Google’s artificial intelligence had defeated humans at Go. 3 In response
to the news, I began study and research while exchanging messages with one of the
authors of this book, Akinori Tanaka. When I studied machine learning for the first
time, I realized that mathematical tools were similar to physics, so I could easily
understand ideas and applied them. As I studied further, I understood that some
ideas were imported from physics. 4 The research went well and was formed into
several papers. In this book, physics researchers explain deep learning and related
fields from the basics to the application from that point of view, and in fact, this is
the path I followed.
My specialty is lattice QCD, and it was no surprise to discover that a person
in the field is one of the developers of Jupyter Notebook. 5 This is because if you
are doing physics using a computer, you have to visualize a large amount of data.
The WWW, the core technology of today’s Internet, was developed at the European
Nuclear Research Organization (CERN). 6 Also, the preprint service called arXiv 7
was created by a particle physicist. 8 At the meta level, physics supports the next
generation of technology. I am glad if the readers who have studied physics and deep
learning through this book will theoretically solve the mystery of generalization of
deep learning and clarify the relationship between physics and machine learning.
At the same time, I would be happy as well if readers will be responsible for the
next generation of technological innovation that gave us the WWW and Jupyter
Notebook.
2 Deep Blue was a supercomputer that defeated the world chess champion Garli Kasparov in 1997.
3 Deep Mind’s Alpha Go defeated Lee Sedol.
4 This is about the Boltzmann machines that triggered the huge progress in deep learning.
5 This person is Fernando Pérez from University of Colorado at Boulder. He obtained a PhD in
Lattice QCD, and was a student of Anna Hasenfratz who is famous in the field.
6 The actual development was done by Tim Berners-Lee, an information scientist who was there,
but I don’t think it would have been done without the accelerator, which is a machine that generates
a lot of data by itself, or without a strong motivation to understand nature. It’s also not hard to see
how the hacker spirit of the physicist community (in the original sense) could make this system
free to use.
7 The arXiv, https://arxiv.org is a web archive service for researchers to upload research preprints
on physics and mathematics.
8 Paul Ginsparg, a prominent physicist whose name is known in the important Gisparg–Wilson
relation in lattice QCD, first started the server for the arXiv. The preceding service was founded by
a physicist, Joanne Cohn. (I thank Shinichi Nojiri, Professor of Nagoya University, for providing
this information.)