Korean AI team plagiarizes shock academia! One tutor with 51 students, or plagiarism recidivist

This year, CVPR, It's really a bit magical ！

First, Seoul University, South Korea AI Team Oral, The former 4% My excellent thesis , Was exposed to be suspected of plagiarism 10 Papers . In some places , even Not a word has changed .

The incident made headlines in South Korea on the same day , Netizens at home and abroad were shocked ： Get it CVPR That's great .

More people broke the news , The team has plagiarized more than once , The basic operation belongs to .

As a result, this wave is not over yet , Another wave has begun .

IBM Published in CVPR Papers on TableFormer Accused of plagiarizing domestic 2021 An article published in .

Broke the news that ,IBM The paper plagiarized their methods , Including pretreatment 、 visualization 、 Reasoning 、 System solutions, etc , But they did not quote their papers .

……

Plagiarism occurs frequently , So many netizens sigh ： This session CVPR It's magic .

Some even listed such a problem directly on Zhihu ：

Let's expose it .CVPR 2022 Which papers are plagiarized ？

The Korean team was exposed to plagiarism more than once

First, let's take a look at the Korean team plagiarism incident that has attracted public attention .

In the latest public response of correspondents yinshenglao and yizuo Kim , All indicate that this plagiarism belongs to yizuo Personal behavior , It has nothing to do with the rest of the team .

Yinsheng explained ：

Other co authors of the team sent their respective parts to a work , But he didn't use it in the end , But plagiarized other papers .

He added , Now the other two papers that I am reviewing are also Be withdrawn from 了 .

however , Netizens found that , The team seems to have plagiarized more than one paper .

Such as this one ：《Towards Fast and Accurate Object Detection in Bio-Inspired Spiking Neural Networks Through Bayesian Optimization》.

The detection software displays , It has 40% Suspected of plagiarism , There are 27 The expression is highly similar to previous studies .

There's another one 《Energy-aware Placement for SRAM-NVM Hybrid FPGAs》, Similar problems have been found .

Correspondents are all old Yinsheng .

meanwhile , It is also pointed out that Professor Yin's laboratory is very large .

Doctoral students have 37 position , Plus master's and postdoctoral , The laboratory has 51 Students are instructed by yinshenglao .

This is also worrying , Whether a professor can guide so many students at the same time , Whether the quality of their experiments and papers will be affected .

And from the beginning of the incident —— From the content of the tubing exposure video , Among the papers suspected of plagiarism , A lot of places are Wholly intact Ground handling is the expression of previous research .

The parts involved are Introduction、Preliminaries、Method etc. .

There are even some places where there is no word left copy.

There are as many plagiarized paragraphs in the full text 25 It's about , Exposure of video display problems took 7 minute .

Many of the plagiarized papers were NeureIPS、AAAI Wait for the top to receive , Among them, several of the first works are Korean .

No wonder the people who uploaded the exposed video , Will give the video such a “ Dazzling ” The title of the ：

E2V-SDE or: How I Learned to Stop Worrying and Love Plagiarism.

E2V-SDE also called ： How did I not worry and love plagiarism .

there E2V-SDE, From the old team of Yin Sheng who was CVPR 2022 Received paper title .

Now? , As the events continue to ferment , The old team of Yin Sheng also aroused the group anger of Korean netizens .

Need to know , Yin Sheng is always an authoritative expert in the field of artificial intelligence in Korea , From the Institute of artificial intelligence, Seoul University .

This year, 6 month , He has just been selected by the Ministry of science and technology of South Korea as “ Lead researcher ”, Enjoy about... Per year 8 Billion South Korean won （ Renminbi equivalent 416 ten thousand ） Scientific research funds .

But now , His team was caught up in a plagiarism scandal .

Someone left a message saying , I hope relevant units can seriously deal with this matter , Let the relevant researchers sign out academic circle .

In the past 30 In the year , Many people study it day and night , So that South Korea can CV More papers will be published at the summit , And their behavior is so excessive ！

actually , South Korea in CVPR 2022 The achievements in the course of , Seoul University alone 25 Papers were selected .

Someone said directly , This is a disgrace to Korea .

The deeper reasons include , There are voices in South Korea worried about , This event will be “ The Huangyuxi incident ” For the second time .

2005 year , He was the chief professor of the Veterinary College of Seoul University 、 Hwang Woo Suk, once regarded as the national hero of Korea , He was exposed to have forged a number of research results .

Its presence 《Science》 The stem cell studies published in the are all nonsense .

2009 year , Hwang Woo Suk was sentenced to 2 Years in prison , Probation 3 year .

IBM Accused of plagiarism

The Korean plagiarism is still fermenting ,IBM It was also revealed that they were selected CVPR The paper is suspected of plagiarism .

A researcher from Ping An technology , Lists nine pieces of evidence , Involving methodology 、 Preprocessing 、 Post processing 、 Reasoning 、 Text line detection and recognition , It has been emailed to CVPR 2022 program chairs.

First of all, from the perspective of core methodology , Both are table content recognition tasks , The name is similar , One is TableMASTER, One is TableFormer.

among , The image input of both is 448*448. The former （TableMASTER） The training length is 500, While the latter （TableFormer） It was changed to 512.

The informant said , Many traces can be seen inside ,IBM Is running on their open source pre training model , Just changed some details .

And the most obscure is the direct call “ The most shameless ” Of , But also belong to Text single line detection This piece of , The informant said ： They just changed the color .

generally speaking , Each table in the official data is multi row , It's not good for the future training .

therefore , The research team is right 3000 Images have been redrawn , Split the connected rows into a single row .

and IBM Of PDF Cells It is to take them directly in GitHub On the trained model .

besides , According to the informant's description ,IBM They will also 3 This rule is based on open source code , Forcibly disassemble into 9 Bar rule .

at present ,IBM Fang hasn't responded yet . Netizens also hold various opinions .

Some people think that , It looks like a personal opinion , The author is a little emotional .

Other netizens read the whole evidence , It's hard to believe IBM It's an original article , It is recommended to report directly to CVPR Project committee complaints .

One More Thing

Last , Let's briefly review the magic of this year CVPR.

CVPR 2022 The number of submissions is as high as 8161 piece , Compared with last year 7093 Sub submissions have increased 15%, among 44.59% The author is from China .

among , share 2064 Papers were received , The receiving rate is 25.28%. In the received papers , Yes 342 Were selected as Oral.

The best paper went to ETH Zurich、 University of Washington 、 Georgia institute of technology 、 Researchers from Czech University of technology and other institutions .

And the best student thesis Award , Is awarded to Interns in Dharma hall .

Professor lifeifei won the award of Thomas S. Huang A Memorial Award .

Besides, I have to admit , This year, CVPR It's really lively , After all, it is near 6000 People come to the offline conference .

So , Another magical thing happened —— Many people get new crowns because of this .

Netizens also ask questions ： What is the CVPR High acceptance rate or high infection rate of new crown ？

reference ：
[1]https://www.reddit.com/r/MachineLearning/comments/vlpnuw/d_ibm_zurich_research_plagiarised_our_paper_and/
[2]https://twitter.com/e2v_sde_parody/status/1540087877308239874
[3]https://arxiv.org/pdf/2105.01848.pdf
[4]https://arxiv.org/pdf/2203.01017.pdf
[5]https://www.youtube.com/watch?v=UCmkpLduptU&t=95s
[6]https://www.fmkorea.com/4760102853
[7]https://www.zhihu.com/question/539432448/answer/2543861341
[8]https://www.hankyung.com/society/article/2022062674031
[9]https://www.reddit.com/r/MachineLearning/comments/vjkssf/d_how_to_copy_text_from_more_than_10_previously/