Source: Internet
Editor: David, Xiao Xianyu
[Xin Zhiyuan Guide] April 29, 2021, Fukushima Kunihiko Fukushima (Kunihiko) won Scientific Achievement Award. He has made outstanding contributions to deep learning, and his most influential work is undoubtedly the "Neocognitron" convolutional neural network architecture.
Recently, the father of LSTM Jürgen Schmidhuber posted a video on his YouTube channel to congratulate Kunihiko Fukushima, the winner of the 2021 Bauer Science Achievement Award.
On April 29 this year, Fukushima gained great influence on deep learning neural network architecture, especially in 2021. Bauer Award for Scientific Achievement.
Actually,Those who are familiar with this Jürgen Schmidhuber know that he has always been concerned that his early original achievements in the field of deep learning have not been widely recognized by the industry.
However, although this video also spent a lot of space to introduce his unrecognized original results, such as ResNet, AlexNet, etc., but also congratulated Fukushima Kunihiko's winning, and He introduced his outstanding achievements.
So, how powerful is this "Bauer Science Achievement Award"? Who is Fukushima Kunihiko?
The oldest science award in the United States, witnessing 200 years of human progress
Speaking of this "Bauer Scientific Achievement Award", it is actually an award set up under the prestigious research institution "Franklin Institute".
Founded in Philadelphia in 1824, the Franklin Institute is a physics education and research institution. Its core goal is to promote the development of science and technology and provide value to the public and the scientific community. Service.
This institution is named after one of the founding fathers of the United States, physicist Benjamin Franklin . Since January 1826, awards for outstanding achievements in science, engineering, and industry have been awarded. It is the oldest comprehensive science and technology award program in the United States.
To date, the Franklin Institute Awards have rewarded more than 2,000 innovators, pioneers, engineers, and scientists worldwide.
It can be said that the history of this award is the history of human discovery and progress in the past two centuries. Many great inventions that have completely subverted our way of life today, from diesel engines to Mars rover , from typewriters to mobile phones, from X-ray technology to genome editing, the discoverers have all won this award.
The list of the winners of the Franklin Society Award is a list of human "all-stars":
0p Third from left in the front row) Won the Franklin Engineering Award
Nikola Tesla, Thomas Edison, Mrs. Curie_span_span_span_span, Marx Planck, Henry Ford, Enrico Fermi, Edwin Hubble, Stephen Hawking, Gordon Moore, , Gates ......
So far,121 Franklin Society Prize winners have won the Nobel Prize. Moreover, most Nobel Prize winners have won the Franklin Institute Prize before winning the Nobel Prize.
In 1988, the Philadelphia chemical manufacturer Henry Bower donated US$7.5 million to the Franklin Institute, and the Bauer Scientific Achievement Award was established. The Bauer Prize for Scientific Achievement is awarded on a different theme every year, and the annual prize is no less than 250,000 US dollars.
The deadline for nomination for this award each year is the end of April, and nominees must be alive.
Fukushima Kunihiko: Who is it? The name Kunihiko Fukushima doesn't sound like a legend, but his work is definitely an indelible legend in the history of the development of neural networks.
In 1979, Dr. Fukushima developed a neural network model for pattern recognition in STRL: Neocognitron.
Very strange, right? But this Neocognitron, in today’s words,Called Convolutional Neural Network (CNN), it is one of the greatest inventions of the basic structure of deep neural networks, and it is also the core technology of current artificial intelligence.
What? Wasn't the convolutional neural network invented by a big man named Yann LeCun? Why did it change to Kunihiko Fukushima again?
Strictly speaking, LeCun is not the first convolutional neural network (CNN) to use error backpropagation to train the first architecture. The man who invented this structure. The Neocognitron introduced by Dr. Fukushima is the first neural network that uses convolution and downsampling, and it is also the prototype of a convolutional neural network.
Kunihiko Fukushima designed an artificial multilayer neural network with learning capabilities that can imitate the visual network of the brain. This "insight" has become the basis of modern artificial intelligence technology. Dr. Fukushima's work has brought a series of practical applications, from self-driving cars to facial recognition, from cancer detection to flood prediction, and there will be more and more applications.
Growth experience
Dr. Fukushima was in a poor family when he was a child, and had little extra money to buy toys. An uncle gave Fukushima an extra transformer and a disassembled electric motor to repair. Curiosity made him passionate about wires and circuits, and he also used scrap materials to make electric trains and radios.
's love for electronic technology keeps him followingIn 1958, after obtaining a bachelor's degree, he continued to study electronics, and in 1966 he obtained a doctorate degree in electrical engineering from Kyoto University .
After graduating from Kyoto University, Fukushima joined the research department of the Japan Broadcasting Corporation NHK , where his research on efficient coding of TV signals became the direction of his doctoral thesis.
The legendary story began in 1965, when NHK established a basic research institution called Fukushima Broadcasting Science Laboratory there. A research group for visual and auditory information processing. This group includes not only engineers, but also neurophysiologists and psychologists. Its purpose is to study the biological brain, and it is also the ultimate destination of the signals transmitted by television and radio.
The birth of "Neocognitron"
If you see something moving, it is actually your brain using a series of hierarchical classifications to finally judge it.
First, your brain may process the size of the object, and then your brain will narrow the recognition range by defining features such as color and shape, while gathering other related memories and past experiences in Together.
It is this layered model processing and our personal life experience,We can immediately distinguish cats from dogs. When we were toddlers, we might mistake foxes as dogs or hamsters as mice, but our brains will soon learn the correct classification again.
Fukushima's plan is to build and train an electronic version of the human brain recognition system.
_span1 _span1 _span1 _pbr0sp _pbr5p _span1 _span
_pbr5p , Assemble artificial neural network . His first artificial network "Cognitron" demonstrated an ability to recognize patterns, but it was powerless to move, rotate, or partially blur.
Based on the understanding of neurophysiology , Fukushima felt that a larger, more layered and organized network should be needed to achieve advanced pattern recognition capabilities.
1979,"Neocognitron" came out, It was inspired by two kinds of nerve cells known to exist in the primary visual cortex, one is a simple "S" cell and the other is a complex "C" cell. They follow The cascade sequence arrangement can be used for pattern recognition tasks.
Fukushima's "Neocognitron" stacks layers of artificial "S" cells and "C" cells, and combines learning rules to modify the interaction between cells and these internal hidden layers. This first deep learning neural network successfully recognized handwritten digits from 0 to 9 and can adapt to changes in writing.
Although today, handwritten number recognition is only a toy case for beginners in machine learning, but if time goes back to 1979, the ability of "Neocognitron" is definitely an amazing feat.
A lifetime of honor and honor. After moving to Osaka University in 1989, Dr. Fukushima expanded his research on neural networks and machine learning, not only for visual pattern recognition, but also imitating many other functions of the biological brain Make a neural network model.
Today, Dr. Fukushima mainly works from his home in Tokyo, developing new training methods and architectures for neural networks, and identifying patterns of deformation and displacement more robustly with smaller computational requirements. Although the vast majority of people have retired long ago at the age of 83, Dr. Fukushima still published his latest paper in November 2019.
Dr. Fukushima's lifetime honors,In addition to the Bauer Scientific Achievement Award in April this year, there are also the Japan Science and Technology Agency Award (1985), the Institute of Electrical and Electronics Engineers Neural Network Pioneer Award (2003), the Asia-Pacific Neural Network Assembly Outstanding Achievement Award (2005), International Society for Neural Networks Helmholtz Award (2012), Electronics, Information and Communications Engineer Society Outstanding Achievement Award (2017), Japanese Society for Neural Networks Academic Award (2017), and Takayanagi Kenjiro Award (2020) And a series of awards.
master style
In a recent interview, Dr. Fukushima was asked about his research methods and said:
17 It’s about using your own creativity instead of simply following current popular research. In order to get new ideas, it’s important to pay attention to your subject and your own work. You must stay focused on your goals, and Not let your research direction wander aimlessly.”
Reference link:
. https://www.fi.edu/laureates/kunihiko-fukushima
https://twitter.com/schmidhuberai/status/1448168005469413376?s=21 _pan1span p0 span nhk.or.jp/strl/english/publica/giken_dayori/194/1.html
.-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-