Chestnut Li Gen from Tsinghua 
qubit report | Public account QbitAI

The 2017 Turing Award was awarded to John L. Hennessy and David A. Patterson .

One of them is the former president of Stanford University and the current chairman of Alphabet, and the other is a retired professor at UC Berkeley who is now a scientist on the Google TPU project.

They have been crowned the highest honor in computer science for their pioneering RISC approach, which stands for reduced instruction set computer.

This time, David A. Patterson came to China for the first time after winning the Turing Award.

After an honorary doctorate award from Tsinghua University and a speech titled "Computer Architecture, the Next Golden Era", Prof. Patterson was interviewed for qubits.

The new Turing Award winner, a generation of computer architecture masters, a professor at UC Berkeley for 40 years, and the author of the textbook " Computer Architecture: Quantitative Research Methods " that has influenced generations of engineers...

Share with us more thoughts and feelings outside the lecture.

About the Turing Award, about chips and hardware and software trends, about AI bottlenecks and key transitions, and suggestions for scientific research talents to stay in school or go to enterprises.

Professor Patterson has a lot to say.

△  Professor David A. Patterson

Turing Award?

"Told me 8 years ago that I was nominated"

Professor Patterson said he was nominated for a Turing Award for spoilers eight years ago.

He himself has also participated in the selection of some awards. Generally, he will not inform who he has nominated in advance, so that if he does not win the award in the end, he will not be lost or embarrassed.

But the people on the Turing Award jury are too honest. As early as 8 years ago, I told him that he had been nominated, but year after year, he never kept the clouds open to see the bright moon.

How would you describe this "torture"?

Professor Patterson recalled that before the lottery draws every year, he thought it was his turn this year, but he didn't win.

So this year, he has actually given up his expectations.

Unexpectedly, the 2017 Turing Award was finally awarded to him.

Knowing the moment of winning, there is only one feeling left: Fantastic!

chip

This is the golden age of the combination of soft and hard

In the history of the Turing Award, the number of times awarded to the field of computer hardware is not many.

So does it also mean that the times are changing and the wind direction is changing - a new golden age of hardware is coming?

Professor Patterson does not fully agree.

He believes that there are still many good opportunities in software, and the key is that many talents are involved in it.

However, in the hardware field, there are not many people now, or if the scope is expanded - the combination of software and hardware, the talents gathered are not enough.

However, he is very optimistic about the general trend of the combination of software and hardware, because under the current situation, if you want to seek better performance, the combination of software and hardware is an important direction, and the opportunity is very good.

The environment is also changing.

A few years ago, there were only a handful of startups around hardware and computing, and it was not easy to get investment.

Nowadays, many VCs are focusing more on hardware, or the integration of software and hardware, and many students who originally specialized in machine learning are using the combination of software and hardware to accelerate machine learning, all because demand sees an opportunity.

In short, Professor Patterson's point of view is: the pure software sky has become a combination of software and hardware.

Whether it is scientific research or product development, the combination of software and hardware may be a visible direction—a direction that may have a great impact.

Optimistic about edge computing

But if you're starting a business, Professor Patterson has a little more advice:

Go to the edge to innovate, because the opportunity of cloud can't grab the giant.

Professor Patterson said that even if there is a great chance of combining software and hardware, if you want to compete with giants in the cloud battlefield, 90% of startups may fail.

What kind of company has the power to dictate the cloud market? giant. Companies like Google, Alibaba and Baidu.

Even Intel cannot compete with these giants in the cloud market.

In the battlefield of cloud computing, a bit of history repeats itself.

In the development process of the computer industry, there was a period of vertical integration at first , that is, eating up the entire market from top to bottom. Your hardware is IBM's, the operating system is also IBM's, and you have to buy IBM's for everything.

But then it evolved into horizontal integration , Intel chips, Microsoft operating systems, Oracle software.

Now, the trend is more like vertical integration.

Those giants that already have an advantage in the cloud have a complete software architecture, a complete set of hardware, change the process, and innovate the architecture. The further down the line, the result of vertical integration may surpass that of IBM.

Because of cloud To B, after customers come, cloud service companies can decide what kind of customers to provide corresponding hardware.

Therefore, no matter how great the combination of software and hardware is, the cloud market cannot compete with giants.

But on the side of the edge, it's a completely different world.

Professor Patterson predicts that there will be hundreds of startups, each with its own strengths, and it is uncertain who will laugh for how long and who will have the last laugh.

In short, he is also more optimistic about companies focusing on AI around the edge, whether in the United States or China.

AI bottleneck

The key to the transition lies in the calculation

If the problem is expanded, where will the next opportunity for AI lie?

Still counting, Professor Patterson said.

He explained that although this wave of AI is prosperous due to deep learning, it will not work without computing resources such as GPUs.

So if you talk about the key to the AI ​​transition, he thinks it is still computing.

For example, many machine learning tasks are now limited by computing memory and cache. Applications such as autonomous driving are generating unimaginable data and computing requirements all the time. However, due to computing constraints, many tasks are challenging.

So he re-emphasized that computing, or edge computing further, could bring surprises to the next step in AI development.

Of course, Professor Patterson also opened up a little "off-topic", because everyone was asking about new technologies, but not paying attention to new challenges.

Due to the rapid advancement of AI, the scale of data and computing is unprecedented, so the security issues behind it also deserve more attention.

If there is demand, there will naturally be opportunities, so Professor Patterson also said that there are also great opportunities in the field of security.

△  Students are asking for autographs with textbooks

Where to go for scientific research talents

I retire from UC Berkeley after my body couldn't take it anymore

Finally, we also asked Professor Patterson about the talent challenges that industry-university-research institutes are facing.

Because after the revival of AI, many scientific research talents in colleges and universities who shoulder the responsibility of education have been poached by companies with a lot of money. It is not impossible to create a talent gap for the next generation.

The most typical case, such as Uber almost took away CMU's robotics department.

But Professor Patterson is not pessimistic. He believes that the phenomena of Uber and CMU do exist, but they are not normal.

Because for many talents, the most hopeful state is to "take care of both sides", half of the time serving the enterprise, and the other half of the time continuing to participate in the research of the school's cutting-edge topics, such "part-time" cases are becoming more and more common.

But Professor Patterson denied that he joined Google after retiring from UC Berkeley, not for the above reasons.

It's because his body is a little overwhelmed.

Professor Patterson said that before UC Berkeley retired, he was 68 years old, and if he had to work 60 hours a week, he would not be able to stick to it.

And he retired after 40 years of teaching, which is quite appropriate, because if 45 retires, it sounds similar to 40, but if he sticks to 50, his confidence is limited.

So he decided to retire, but still wanted to do something.

I just joined Google during the academic vacation in 2013. At that time, I participated in a project called Pixel visual core, mainly on the Pixel 2 mobile phone hardware, to accelerate image processing and other matters to achieve high-definition and realistic photos.

At that time, he felt very happy. He didn't need to listen to people's reports at all times. It was similar to the state when he was studying for a Ph.D.

Now, he is doing TPU research and development at Google. Professor Patterson said this is a brand new field and he is still looking for the best way to work, but it is also very interesting because it can stimulate human potential and intelligence (Intellectually Stimulating).

In school or in business?

If you compare the difference between doing research in schools and doing research in companies, Professor Patterson would answer: different perspectives, different results.

In school, you can be imaginative, keep imagining the future, and dare to think and do.

But in the enterprise, you can't be too futuristic, and when it comes to more futuristic plans, you must be more pragmatic.

In addition, you can publish papers in school without reservations. However, in the company, it is much more cautious. It will consider how much information can be disclosed and whether it has enough competitive advantages. The main thing is to let the outside world know that the research is very powerful, but how to achieve it cannot be too detailed.

However, for Professor Patterson himself, he feels that the past is a prologue, and he enjoys his past, present and future.

For example, if he hadn't stayed at the school, he might not have written textbooks with John Hennessy, who won the Turing Award together, and he would still have some regrets.

In short, follow your heart and enjoy the moment.

Three more things

1. Professor David A. Patterson was awarded an honorary doctorate degree by Tsinghua University this time, and his keynote speech "Computer Architecture, the Next Golden Age" will be shared with the public in the follow-up.

2. At the meeting, Professor Patterson also sprinkled a handful of dog food, saying that he and his wife have known each other for 51 years and are very happy (retiring from UC Berkeley is also to better accompany his family).

3. Professor Patterson has another fate with Tsinghua University. His only 2 Chinese Ph.D. disciples are all from Tsinghua University. One is Tan Zhangxi, the founder of the chip entrepreneurship project OURS, and the other is Xu Wei, an assistant professor at the Tsinghua Institute of Interdisciplinary Information Research. Of course, Xu Wei's current "Boss", Professor Yao Qizhi, is also a Turing Award winner ^_^

△  Two Chinese Ph.D. disciples of Prof. Patterson:

Xu Wei on the left, Tan Zhangxi on the right

-End- _ _

Annual selection application

join the community

The qubit AI community has started recruiting. Students who are interested in AI are welcome to reply to the keyword "communication group" on the QbitAI public account (QbitAI) dialogue interface to obtain the method of joining the group;

In addition, qubit professional sub-groups (autonomous driving, CV, NLP, machine learning, etc.) are recruiting for engineers and researchers who are engaged in related fields.

To enter the professional group, please reply to the keyword "professional group" on the QbitAI official account (QbitAI) dialogue interface to obtain the method of joining the group. (Professional group review is stricter, please understand)

Event planning and recruitment

Qubit is recruiting event planners, and will be responsible for the planning and execution of online and offline related activities in different fields and dimensions. Smart and reliable partners are welcome to join, and I hope you can have some relevant experience in event planning or operation. For relevant details, please reply to the word "recruitment" on the QbitAI dialogue interface.

Qubit  QbitAI · Toutiao signed author

վ'ᴗ' ի Track new developments in AI technology and products