Celebrating Professor Shun-ichi Amari's Kyoto Prize: A Special Issue of the English Journal New Generation Computing and a Report on the JSAI2026 Organized Session

The person who created the origins of modern AI has modern AI draw pictures to tell his autobiography

On the morning of the fourth day of the JSAI2026 Annual Conference (JSAI2026, G Messe Gunma), June 11, 2026. Nearly 100 participants, including those online, gathered in Hall B starting at 9:00 AM.

The organized session, "Commemorating Professor Shun-ichi Amari's Kyoto Prize and Prospects for Neural Network Research," was hosted by the Japanese Society for Artificial Intelligence's English journal, New Generation Computing (NGC). The title of the lecture by Professor Shun-ichi Amari shown on the Zoom screen was "My Lucky Life: Looking Back on 65 Years of Research in Mathematical Engineering".

From the very beginning, my eyes were drawn to one particular slide. The illustrations at the start depicting the "Seven Wonders of the World" and "Post-war Tokyo"—Professor Amari mentioned that these were actually drawn by ChatGPT. The person who created the origins of modern AI has modern AI draw pictures to tell his autobiography. I felt that this single slide condensed the entire essence of this session.

In this article, I would like to report on the proceedings of this session and also share information about the special commemorative issue we are currently compiling, as well as the English journal NGC that serves as its foundation. This is also an invitation for submissions to all researchers.

What is New Generation Computing in the first place?

At the beginning of the session, Editor-in-Chief Ryutaro Ichise (Tokyo Institute of Science) first introduced the journal itself. When he asked the audience, "Has anyone heard of NGC?", only a few hands were raised. To be honest, it cannot be said that it is widely known yet. However, it is a journal I want you to know about.

New Generation Computing is a

• prestigious English journal with a history of about 40 years that became the English journal of the Japanese Society for Artificial Intelligence in 2023.

• Currently published by Springer Nature.

• Impact Factor of 2.8 (2024), which is a relatively high standard in this field.

• Average of 107 days to the first review decision (we are working on further reductions).

• Approximately 99.8K annual downloads. It is a journal that is well-read.

The following article about NGC was published in the May 2022 special feature of the JSAI journal "Artificial Intelligence" titled "Thinking about the Future of Academic Journals," so please take a look if you are interested.

We cover a wide range of fields handled by the Japanese Society for Artificial Intelligence, including learning, data mining, and cognitive computing. And as a benefit that is often overlooked — for society members, via the member page you can view all NGC papers for free. If you haven't accessed it yet, please take a look.

An international special issue commemorating Professor Amari's Kyoto Prize

What we are currently focusing on is a special issue commemorating Professor Shun-ichi Amari's Kyoto Prize. Lead guest editor Hideaki Shimazaki (Kyoto University) explained the aim of the project.

To not keep Professor Amari's achievements within Japan, but to re-disseminate them overseas and connect them to the next generation — this is the primary goal. Therefore, we are proceeding with

• an international editorial structure that invites many authors from overseas and

• includes overseas editors

. With Professor Amari's own support, we have invited contributions from distinguished researchers influenced by him, including Michael A. Arbib, who also appeared in the lecture.

Invited Lecture 1: Professor Shun-ichi Amari, "My Fortunate Life"

Professor Amari's lecture was a story of his research life itself, starting from the scorched earth after the war. I interpreted it as three major waves.

The first wave was the encounter with mathematical engineering, which was considered a "fourth-rate discipline." After being absorbed in the anti-war movement at the University of Tokyo and failing to get the grades for his preferred major, he ended up in the Mathematical Course of the Department of Applied Physics. It was an unpopular department at the time, with "five students for four professors and associate professors." It was an era when it was scorned that "mathematics should be pure, and applications are work for dropouts." However, he says the atmosphere of the professors who were desperately studying for survival—saying "Academic study is free; mathematical engineering is a method, not a subject, so you can do anything"—became Professor Amari's starting point.

The second wave was his research on the perceptron at Kyushu University. Neurons at the time were 0/1 models, and no one knew how to train the hidden layers. So Professor Amari thought, "If you make each neuron analog, the whole becomes continuous, and you can calculate the gradient to update the weights" — this is the origin of what we now call Stochastic Gradient Descent (SGD) and backpropagation. In his 1968 book, he even showed that multi-layer neural circuits could identify patterns that were not linearly separable, even while being ridiculed for engaging in mathematical brain science as "a discipline in the sky."

The third wave was the founding of information geometry. Professor Amari, who had "grown tired" of neural cell model research, thought: "When you look at information broadly, there is Shannon's probability, testing theory analysis, and computer algebra, but only geometry is not conveyed. So shouldn't we call it 'information geometry'?". This idea of treating a collection of probability distributions as a manifold moved David Cox in the UK, involved world-class statisticians like Efron and Rao, and eventually grew into a major field for which Springer launched a specialized journal.

And now, to modern AI. Professor Amari said, "There are singular structures everywhere in the parameter space of deep neural networks. Isn't this the essence?", and spurred the trend of relying only on scaling laws, saying, "There should be more essential theories. If we can do that, even Japan, which is inferior in financial power, can stand up to the world."

The end was a quiet, yet sharp warning to future society.

Looking back on his research life, what Professor Amari entrusted to the younger generation was the words, "I want you to create a civilization that maximizes the potential of the self, based on playfulness and the joy of working".

Invited Lecture 2: Ryo Karakida, "The Present of Deep Learning Connected from Statistical Neurodynamics"

The next speaker, Ryo Karakida (National Institute of Advanced Industrial Science and Technology), is a next-generation theorist who follows in Professor Amari's research lineage. He began with an anecdote about how he met Professor Amari at a poster session while analyzing his synchronous firing model during his master's studies, and later received a handwritten "Amari Memo."

Mr. Karakida's lecture vividly demonstrated that the "statistical neurodynamics" pioneered by Professor Amari around 1970 is directly connected to the forefront of modern deep learning.

• Reducing large networks to a small number of macroscopic variables: Statistical Neurodynamics

• → Signal propagation theory by Stanford and Google BrainSignal Propagation Theory

• → Joint research with Professor Amari: Hessian analysis of the Fisher information matrix

• → NTK (Neural Tangent Kernel) regime, and μP which preserves feature learning

• → Natural Gradient Descent proposed by Professor Amari, and the second-order optimization Muon officially implemented in PyTorch

• → From associative memory devised by Professor Amari, to modern Hopfield networks, Transformer attention mechanisms, diffusion models, and continual learning

Many of the technologies supporting modern AI can be traced back to Amari's theory. It was impressive to see the audience nodding deeply at Mr. Karakida's words: "When you think about various problems, the principles always lie in Professor Amari's mathematics of neural networks."

Dialogue: "Why hasn't Professor Amari won a Nobel Prize?"

The second half of the session was a dialogue featuring Professor Amari. Now that AI has become a subject of the Nobel Prize (2024 Physics Prize), the discussion has even reached international circles: "Backpropagation and Hopfield-type networks were done by Amari long ago. Why hasn't he won a Nobel Prize?" Such topics were brought up. This is precisely why I believe a special issue that re-disseminates these achievements to the world is meaningful.

When the moderator, Mr. Shimazaki, asked, "If Professor Amari were a graduate student today, would you work on AI or the brain?"

When asked about the "goal" of neural network research, he replied, "Memory, the mechanism of inference, and the mechanism of consciousness. These must be considered within a broader framework that transcends brain science." His gaze was still looking toward the uncharted territory ahead.

Mr. Karakida's lecture concluded with these words: "Once again, Professor Amari, congratulations on receiving the Kyoto Prize." The feelings of all us organizers are exactly the same.

Conclusion: Boosting Japanese AI Research through an English Journal

In his closing remarks, co-organizer Shin-ichi Shirakawa (Yokohama National University) stated, "I have realized once again that Professor Amari's influence continues to this day. From the perspective of an English journal, we want to continue planning initiatives that boost Japanese AI research."

New Generation Computing is our important stage for delivering research from Japan to the world. Just as Professor Amari carved out a global field from what was once called a "fourth-rate discipline," your research, too, may become the source of an unexpected future.

We sincerely look forward to your submissions to New Generation Computing.

(Text by Yoji Kiyota / NGC Deputy Editor-in-Chief, Co-organizer of this session)