Legendary master He Kaiming was revealed to have returned to academia. Netizens: He is going to become the person with the highest citations at MIT.

Mengchen Alex from Aofei Temple
Qubit | Public account QbitAI

AI guru He Yuming has the latest trend and is returning to academia .

MIT CSAIL Laboratory announced that He Kaiming will give an academic lecture at MIT next Monday, March 13th.

This matter has attracted widespread attention in the AI ​​circle. In a related Zhihu question, MIT doctor @Charles pointed out that the seminar He Kaiming participated in was all about Job Talks .

Chen Yiran, a professor at Duke University, also said that he has always heard that He Kaiming is looking for a teaching position, and now it is finally confirmed.

Job-seeking speeches are a major tradition in academia. Applicants for doctoral degrees or teaching positions come to school to give academic reports on their own research results. They are usually held in March and April every year.

And entering this stage means that He Yuming has at least passed the resume screening, and his research results and abilities have been recognized by MIT.

Some netizens predicted that given that Li Feifei’s lecture at MIT was packed with people, He Kaiming’s lecture may also be very popular and become a large-scale star-chasing scene for deep learning .

However, some netizens added in the comment area that this does not mean that He Kaiming will eventually choose to go to MIT, and may also go to other schools, but in short, he is basically certain to find a teaching position. ‍ ‍

Or become the person with the highest citations in MIT

Although He Kaiming's final choice is uncertain, sharp-eyed netizens have discovered that if he really goes to MIT, he will have the highest number of citations in the school .

Currently, the most cited person at MIT is Robert Langer, a heavyweight professor in the Department of Chemical and Biomedical Engineering, with more than 380,000 citations.

He Kaiming has been cited more than 400,000 times .

Although it is not possible to directly compare different disciplines, the number of citations by He Kaiming is really eye-catching.

Some netizens said that although the number of citations may not necessarily explain too many problems, "the person with the astronomical number of citations must be a great person."

When it comes to He Kaiming's most cited research, it must be ResNet . It broke through the 100,000 mark at the end of 2021 and has now risen to 150,000.

Although ResNet itself is researched in the field of computer vision, its core idea, residual connection, has crossed over and become a basic component of modern deep learning models .

AlphaGo Zero, which started the last AI boom, was completed by combining ResNet + reinforcement learning + Monte Carlo search.

In ChatGPT , which started the latest AI craze , the "T" in it means that residual connections are also used in the Transformer network.

In computer vision, He Kaiming's main contributions also include a series of studies such as Faster R-CNN and subsequent Mask R-CNN, which have been the mainstream method of target detection for many years.

His main recent research interest is unsupervised learning . MAE, proposed at the end of 21, uses the mask pre-training method of language models on visual models, paving the way for large-scale unsupervised pre-training of large visual models.

Recently, he also introduced the masking method into CLIP, the basic model of many AI painting applications, increasing the training speed by 3.7 times.

In the upcoming lecture at MIT, the theme prepared by He Kaiming is also "visual intelligence".

Including his results in ResNet, object detection, and the impact of visual unsupervised learning on future research.

As for MIT CSAIL (MIT Computer Science and Artificial Intelligence Laboratory) , which He Kaiming may join , it is the largest laboratory at MIT and a world-renowned computing science and AI laboratory.

Many well-known figures in the fields of computer science and AI gather here, including Tim Berners-Lee, the inventor of the World Wide Web, distributed computing and concurrent algorithm experts, Turing Award winners Leslie Lamport and Barbara Liskov, etc.

When Meta AI has product pressure

He Kaiming received his undergraduate degree from Tsinghua University and his PhD from the Chinese University of Hong Kong. In 2011, he joined Microsoft Research Asia.

Since 2016, he has joined Facebook AI Research, also known as FAIR , and continues to study computer vision.

But just in the past year, the department has undergone two major adjustments, both of which enhanced applied research and reduced basic research .

In June last year, due to the impact of the company's transformation into the Metaverse, FAIR was downgraded to a subordinate organization of Reality Labs, the Metaverse department .

At the same time, the algorithm teams that support its major APPs are transferred to the product engineering team. The new architecture gives the product team greater authority and accelerates the application of technology.

At the end of February this year, Zuckerberg made another adjustment in the face of the ChatGPT craze and announced the establishment of a top AI product team.

Gather everyone in the company who is engaged in AIGC.

Many years ago, when LeCun stepped down from FAIR's management and became chief AI scientist, researchers such as Tian Yuandong and Wu Yuxin who were still at FAIR said that there was no product pressure in this department .

However, due to a series of recent adjustments, Meta has lost a lot of research talents, including at least four well-known scholars such as Edward Grefenstette, who led reinforcement learning research, leaving. Even the AI ​​laboratory in London has lost most of its top researchers.

Judging from He Kaiming's return to academia, he is still more inclined to engage in basic research.

One More Thing

During this period, the comparison between academia and industry once again became a hot topic in the AI ​​circle.

On the one hand, some netizens pointed out that life is difficult for large industrial companies. There are more engineering and application-oriented positions and fewer and fewer research-oriented positions.

On the other hand, large models have taken over, and the cost of AI scientific research has increased exponentially . Even the top journal Science published an article in the main issue stating: The computing power controlled by academia and industry is no longer in the same order of magnitude.

Under the attack of the two, AI giants returning to academia and leaving big companies to start businesses have become two simultaneous trends.

In addition to He Kaiming, representatives who have returned to academia include Pete Warden, a founding member of Google Tensorflow. Domestically, Qi Yuan, the former vice president of Ant Financial, has joined Fudan University, and Dai Jifeng, the former executive research director of SenseTime, has joined Tsinghua University.

Representative figures who have left major companies to start their own businesses include the Google Transformer authors who left to form Adept AI and Character.AI, as well as Amazon’s Li Mu and Alex Smola, who were just revealed two days ago to start a business together.

Under the Zhihu hot topic "What do you think of Kaiming interviewing for a teaching position at MIT?", a highly praised answer is:

Reference links:

[1] https://www.csail.mit.edu/event/eecs-special-seminar-kaiming-he-pursuit-visual-intelligence
[2] https://weibo.com/2199733231/4877346375735190
[3] https ://www.zhihu.com/question/588205714

click here