The legendary life of the father of deep learning

widgets Article
The legendary life of the father of deep learning

Geoffrey Hinton, a professor in the computer department of the University of Toronto, is the originator of Deep Learning. Let's talk about his story.

View Image

He has a legendary aunt

But let ’s talk about his aunt first. His aunt Joan Hinton is a legendary figure related to China. The Chinese name is Han Chun, a scientist who participated in the Manhattan Project, but later raised dozens in China. Year of the Ox.

View Image

In 1945, 23-year-old Hinton had a female scientist in the Manhattan Project, involved in the manufacture of human history 's first atomic bomb. The Red Revolution in China also attracted her deeply, so she gave up pure scientific research and followed her husband, an American young man with common ideals, who came to Yanan, China early, and held it in the cave in Yanan, a revolutionary shrine in 1949. The wedding in the flames.

After that, the two worked together to help the Chinese people develop animal husbandry, contributed their lives to China's dairy industry, and designed and built China's first mechanized cattle farm in Changping, Beijing. When Yang died early in 2003, the ashes were buried in the cattle farm.

His aunt is a legend, but Geoffrey Hinton is no less legendary than her.

Google bought a company for tens of millions of dollars for him

In 2013, Google acquired a startup company DNNResearch from the University of Toronto. In fact, the company has only three members, Geoffrey Hinton and his two newly graduated students-Alex Krizhevsky and Ilya Sutskever.

View Image

Or we can understand that Google bought this company specifically for Geoffrey Hinton to work for them. Some people ridiculed that Google spent tens of millions of dollars to buy several papers.

Of course, before this, Google also gave his research team $ 600,000 in research funding.

Grand Learning Grandpa

Geoffrey Hinton himself is the ancestor of Kaishan ancestor of the Deep Learning school.

What is Deep Learning? To put it simply, it is to build and simulate an artificial neural network that simulates the human brain for analysis and learning.

View Image

For example, a widely circulated example is that Google built the world's largest electronic analog neural network with 16,000 computer processors and showed it 10 million randomly selected videos from YouTube. Under the spontaneous condition without external instructions, the artificial neural network learned to recognize the cat's face autonomously.

Since its introduction in 2006, Deep Learning has greatly promoted progress in speech recognition, vision, and natural language processing.

Explore how the brain works

While studying psychology at Cambridge University, Hinton discovered that the human brain has billions of nerve cells, and they interact with each other through synapses to form extremely complex interconnections. However, scientists cannot explain these specific effects and connections. How does the nerve learn and calculate? This is the question he always wanted to study.

View Image

This issue has made some progress with his efforts. He and his partners built layers of interconnected artificial neuron models that mimic the behavior of the brain and deal with complex problems such as vision and language. He is also committed to studying methods that can make neural networks better, to better simulate certain aspects of the human brain.

In the early 1980s, the performance of computers was far from being able to handle the huge data sets required by artificial neural networks (ANN). The research of neural networks fell into a low tide after a short period of intense heat.

The bench has been cold for "thirty years"

In the following twenty years, although some researchers still insisted on the research of artificial neural networks, the research on artificial neural networks in the entire academic community has basically fallen into stagnation, and researchers have not received relevant research funds There are very few published high-quality papers related to neural networks. Even the well-known academic conference NIPS (Advances in Neural Information Processing Systems) has become a conference that has nothing to do with neural networks.

Geoffrey Hinton's academic career has been ups and downs like artificial neural networks. Fortunately, he has never given up the research of artificial neural networks. In order to realize their ideas, they gathered regularly for seminars, built more powerful deep learning algorithms, and operated larger data sets.

View Image

However, the academic circle is still not interested in the research of neural networks, and their research is difficult to apply for research funding. In fact, the lack of academic circles is not without reason. Many of the results of artificial neural networks are difficult to explain or prove with mathematics. Everyone is just constantly adjusting parameters and improving algorithms to get better results.

The turning point of the matter appeared around 2006, Geoffrey Hilton and his students invented the engineering method of using GPU to optimize deep neural network, and published a paper in Science and related journals, and for the first time proposed "deep belief network" the concept of. He gave a new term to the learning methods related to multi-layer neural networks-"deep learning".

View Image

Subsequent research on deep learning was brilliant and widely used in the field of image processing and speech recognition. For example, Geoffrey Hilton students won the 2012 ImageNet with deep learning algorithms.

Internet giants have begun to notice them, and this field has only started to get hot. Four deep learning (can) kings headed by Hinton and Wu Enda have emerged in the field of deep learning.

The spring of deep learning and artificial intelligence

The IT Internet companies closest to artificial intelligence are keenly aware of this opportunity. Since 2011, the maturity of deep learning algorithms has led to a leap in artificial intelligence technology. Companies including Microsoft, Apple, Google, Facebook, and domestic BAT have begun to lay out artificial intelligence and deep learning in depth, trying to grasp the wind and become the next industry. The giant of change.

View Image

Domestic long-term adherence to the deployment of artificial intelligence is also a veteran electrical appliance manufacturer-Changhong

Since 2012, Changhong has entered the basic research work of artificial intelligence and deep learning. It has successively researched and developed speech recognition chips, research and development and application of artificial intelligence technology based on big data, and research and development and application of machine vision (face recognition) technology. At present, Changhong has accumulated advanced and mature computer vision algorithm achievements in face recognition, target detection and target classification.

They recently released the world's first CHiQ artificial intelligence TV. Changhong TV for Artificial Intelligence awareness, decision making, feedback three large capacity.

View Image

At the "cognitive" level, based on the Ciri + voice platform independently developed by Changhong, CHiQ achieves efficient voice interaction collaboration and semantic recognition and understanding. Among them, the voice recognition rate reaches 97%, and the overall voice coverage rate is 50%, so that the open natural voice interaction distance between the person and the TV can reach 30 meters, that is, the user can use voice to interact with the TV in any corner of the home.

View Image

At the "decision-making" level, Changhong CHiQ artificial intelligence TV can make business logic judgments based on neural network learning algorithms. At the "feedback level", its built-in personalized recommendation platform truly realizes self-learning and growth of recommendations through continuous recommendation effect feedback, model optimization and product iteration.

Changhong Xiaobai is a robot elf who "lives" on this TV. She accompanies, learns and serves users, and provides users with personalized services for watching TV, listening to music, playing games and other black electricity services.

View Image

(The image of Changhong Xiaobai)

In layman's terms, you can talk to Changhong Xiaobai:

Xiaobai, tell me, what programs are good? Hi, I want to see the high points Movie ~

It can give you a timely response to these questions (search and start broadcast within 3 seconds). It will also learn your viewing preferences and history, and give you personalized recommendations.

It is said that in the future Xiaobai will also add physical form, load projection technology, visual image technology, automatic charging and other black technologies.

It can be said that this product marks the first time Changhong TV has entered the era of artificial intelligence.

View Image

Long press the QR code to follow, and sneak into Silicon Valley with the secret agent

Weibo & Facebook: Secret Agent in Silicon Valley


widgets Related Articles

widgets Contribution

This article is contributed by Anonymous and text available under CC-SA-4.0