
Recently, an incident suspected of "academic fabrication" has caused a huge stir in domestic academic circles.
The eye of the storm comes from the just-concluded top academic conference ACM MM 2026. A Best Paper at this international flagship conference in the multimedia field has been pointed out to have many suspicious aspects, casting doubt on its possible fabrication and on the credibility of the title of Best Paper at an international flagship conference.
AI Technology Review (Leiphone) has sorted out the ins and outs of the matter and exclusively contacted the corresponding author of this paper.

On November 14, ACM MM 2026 closed in Brazil. Thanks to the arrival of the AI era, this conference has attracted much academic attention.
The protagonist of this controversy is one of the 17 Best Papers, titled "Clifford-Algebra-Based Geometric Feature Fusion for Real-Time UHD Low-Light Image Enhancement."

This is a work on how to use AI to make images brighter. The direction is popular, and many scholars have worked in this field for decades. What is surprising is that the first author of the paper, Xiaohan Wang, is a second-year undergraduate at Xidian University.
A sophomore's first paper won Best Paper at a flagship conference.

Scrolling down the list page, in the corresponding author column, the name is not followed by a university or research institute, but by Independent, an independent researcher.

In the context of academic circles, this affiliation usually corresponds to two kinds of people. One is those who do not need an institution, who are accomplished and whose titles are already more prominent than their institutions. The other is those who have no institution.
What was supposed to be an inspirational narrative of an independent researcher leading a genius girl in devoted research, braving an international flagship conference, and winning Best Paper, turned into a public opinion storm over whether there was academic fabrication because of what happened next.

According to an investigation by AI Technology Review, the earliest doubt appeared on November 14, when someone asked on Zhihu how to evaluate this Best Paper.

The questioner claimed to have done research in the field of low level vision for many years and was the first to reply to their own question, listing 5 problems.
First, deliberately downgrading comparison methods. The paper compares old methods that were not trained on the dataset with its own trained method.
Second, presenting synthetic data as real shots. The paper claims that the test data are photos taken in the real world, but the authors of the original dataset wrote that the data are computer-synthesized.
Third, stronger methods did not enter the competition. Methods stronger than theirs are only mentioned by name and not included in the comparison table; those included are all unsupervised methods. The paper's own score is 20.56, which cannot even beat an old method from 2018 (21.67).
Fourth, suspicious user ratings. The paper includes a user study, and all 48 sub-items neatly end in .x0 or .x5, with none falling in any other position, and 5 sub-items are exactly identical. The probability of such a distribution occurring naturally is 1 in 3.55 × 10³³.
Fifth, the core selling point does not live up to its name. The paper's title appears to piggyback on the advanced mathematics beacon "Clifford algebra," but if you unpack and calculate the formulas in the paper, it is actually just the most ordinary operation of "multiplying two numbers and then summing them," with little relation to advanced mathematics.
This post quickly attracted the attention of researchers in the field, and the corresponding author also personally responded, calling on the questioners to "boldly reproduce."

The back-and-forth academic exchange drew more attention.
Soon, another respondent (a master's student in electronic information at Tsinghua University) tried to reproduce it. He acknowledged the paper's engineering route but raised more comprehensive doubts about the research motivation and baseline comparison data, directly pointing out that the first author's academic training was insufficient, the corresponding author's guidance was inadequate, and the conference review was not strict.

Excerpted from the conclusion section of the Zhihu respondent; for the full answer see: https://www.zhihu.com/question/2104933122630476610
Because the paper is not open source, the doubts surrounding the data can be neither confirmed nor disproven, and the two sides are shouting at each other through a pane of frosted glass.
On November 17, someone shifted attention from the experimental section to the references: the paper's 12th citation points to a paper that does not exist; the 47th and 51st entries have author information that does not match the real literature.
The three errors are distributed in different positions in the list and are of different types, so it is unlikely to be a "slip of the hand."
To explain for readers unfamiliar with academic ethics: citation hallucination is a serious act of academic misconduct. The 12th fabricated reference alone is enough to constitute a retraction-level consequence, because in nature it means the authors implanted untraceable false support into the paper's chain of argumentation.
Discussion of the paper's technical details requires a threshold, while the citation issue is much more accessible to the public. As this accusation was confirmed, the discussion escalated.

Doubts about the paper quickly shifted to the people.
First the first author, Xiaohan Wang, a second-year undergraduate at Xidian University. The only publicly verifiable paper is this one. In this field, a teacher who graduated with a PhD many years ago may not even touch a Best Paper over an entire career, yet this sophomore won it with a first paper.
More discussion centered on the corresponding author, Zheng Zhuoran. The author information does not list a university or research institute, but "Independent." Publishing a paper as a corresponding author in the capacity of an independent researcher is in itself beyond reproach. However, someone released a screenshot showing that in August this year, Zheng had posted on Xiaohongshu, offering paper guidance services with clearly marked prices.

Openly providing paid services for a specific research direction and promising "quality and quantity guaranteed" has already gone beyond the scope of normal academic exchange and has a clear commercial transaction nature. "5000 before acceptance, 5000 after acceptance" implies that the service price is directly tied to whether the paper is accepted. This is very much like a contingency fee model, that is, paying the balance after the paper is successfully published.
The highlighted "full-process guidance" is also in a gray area. If the substance is completing core work for the student (such as proposing the idea, writing the code, and revising the paper to a publishable state), then this crosses the line into ghostwriting. Even if it is only providing ideas and revisions, combined with the precondition of payment, its nature can easily slide toward academic misconduct.
A single stone stirs up a thousand waves, and online discussion about Zheng Zhuoran's identity has intensified. Some suspect that he takes private jobs at postgraduate recommendation agencies, helping students with recommended admission by guiding papers; others who claim to be close to him say that he once taught at Qilu University of Technology, was dismissed because of resume fraud, and now falsely tells students that he is from National University of Defense Technology.
Public information shows that Zheng Zhuoran received his PhD from Nanjing University of Science and Technology, did a postdoc at Sun Yat-sen University, and then joined Qilu University of Technology.
Whether his current teaching position still exists, there is currently no publicly verifiable information. On his academic homepages (Open Review and Research Gate), the affiliated institution is shown as Nanjing University of Science and Technology; his Zhihu bio reads "PhD in computer science, focusing on low level vision tasks and representation learning," and his Xiaohongshu bio reads "computer science postdoc (Sun Yat-sen University), faculty member in the computer science division, computer vision," none of which explicitly mentions a teaching position.

AI Technology Review (Leiphone (official account: Leiphone)) noted that Zheng Zhuoran's latest work (submitted to arXiv on November 3) is eight pages long and has three references. There are two people in the author column, one is him, and the other is signed "DouBao," exactly the same pinyin as ByteDance's AI assistant Doubao.

On the morning of the 18th, Zheng Zhuoran publicly stated: "Because my supervision was inadequate, we have decided to retract the paper." Subsequently, he deleted multiple replies on platforms such as Xiaohongshu and Zhihu one after another, and cleared the content of his Xiaohongshu homepage. Currently, regarding this controversy, the replies he has left on public platforms are basically only the announcement of the retraction.


Someone noticed that Zheng Zhuoran's postdoctoral advisor, Ren Wenqi, suddenly removed him from his research group roster.


On the evening of the 18th, Zheng Zhuoran accepted an interview with AI Technology Review, his voice sounding somewhat weary.
Zheng Zhuoran clarified the content circulating online one by one. First, the reason for marking Independent (independent researcher) is that during the period when he was supervising this work, Zheng had voluntarily resigned from Qilu University of Technology and had not found a job, so there was no institution to list.
Regarding the online rumor that he was "expelled for resume fraud," he firmly denied it: the information can all be found online, including which papers he has published—how could it be fraud? Zheng Zhuoran said the real situation is that after leaving Qilu University of Technology, he took a gap period, and then entered National University of Defense Technology.
Regarding online questions about interest exchanges and improper procedures involved in his supervision of the paper, his answer was that there were no interest exchanges; out of friendship, he supervised this first author. As for the so-called joining a paper mill, Zheng's exact words were: "I can swear a poisonous oath that I have not joined any paper mill or organization, even though they have approached me many times."
At the same time, Zheng mentioned that his original intention for doing research has always been very pure, and that he and his students are very tight on hardware—the GPUs were rented on Xianyu, and when the server expired, the raw data would be lost, so he has always been relatively pushy with students, with group meetings held every few days.
Zheng mentioned that some people attacked his undergraduate degree as merely Zaozhuang University. He does not avoid his undergraduate background at all, and believes it does not affect his pursuit of research.
A paper from submission to Best Paper must pass through four gates: reviewers' scoring, the AC (Area Chair) making the acceptance decision, the program committee's final review, and finally the awards committee selecting the Best Paper. Some netizens said that the paper had already been explicitly rejected at the AC review stage, but could not withstand three reviewers giving full marks of 5, implying that the paper team had connections with the conference side.
Zheng's version is a completely different story: the paper's initial score was in the middle-to-lower range, and it was the revised manuscript that turned things around; in the middle of this year, when the acceptance news was received, the AC's comment was to recommend it as a poster; one or two months later, an email was received saying the paper had been rated oral; only the day before the award ceremony did he receive a text message learning that the paper had been selected as Best Paper.

The system interface for the accepted paper, provided by Zheng Zhuoran
The key to the revision was highlighting the application value: this method can be deployed on Android phones and has commercial prospects, and the multimedia application track values practicality. After the revision, all three reviewers gave 5 points.
That the algorithm was not fraudulent is the point he most wants to clarify. Zheng Zhuoran cited the reproduction results of a Tsinghua master's student on Zhihu, saying that the paper is basically fine in terms of performance and parameter count across all aspects. The award was selected by a review panel led by foreign parties, and Zheng Zhuoran's side had no way to manipulate it: "If there really were manipulation, it wouldn't have come to finding an acquaintance in Brazil to accept the award on our behalf."
When asked about hallucinated citations, Zheng Zhuoran said he has many students and cannot check every reference in every manuscript one by one; he reminded students countless times, but problems still occurred. The first author was submitting for the first time and lacked experience; he believes the responsibility lies mainly with himself, and at the same time plans to suspend research work and start anew a year later.
In addition, Zheng hopes to convey three messages to the public through us. First, he is considering entering industry and hopes to produce some new work. At the same time, "I hope the academic community gives the first author another chance." Finally, he emphasized that "Xidian University is a good school."
Regarding this controversial paper, all authors of the team unanimously agreed to retract it, and a joint email has been sent to the conference chair's mailbox requesting retraction, noting that "if no reply is received within two days, we will urge again for expedited handling."

In the view of AI Technology Review, the reason this paper was able to pass through four review gates and reach the main conference's highest podium is that the award structure and review ecosystem of ACM MM 2026 itself provide some more structural explanations.
In 2024 and 2025, the ACM MM main conference awarded only one Best Paper each year, with a separate Best Student Paper. In December 2025, the ACM MM Steering Committee passed new rules: considering the increase in submissions, in 2026 it will select about 5% Outstanding Papers from accepted papers by theme, and then select from them Best Papers amounting to about 1% of total accepted papers.

Excerpted from the "Report of the December 2025 vote of the ACM Multimedia Steering Committee," hereinafter referred to as the "Awards Guidelines"
This year the ACM MM main conference accepted a total of 1703 papers and awarded 17 Best Papers, consistent with the officially announced selection ratio.
But that a single academic conference had as many as 17 "best" papers still exceeds people's imagination of top papers.
In the past two years, submissions to top AI conferences have climbed all the way. ICLR's annual submissions rose from 1013 in 2018 to nearly 20,000 this year, an increase of nearly 18 times in eight years. Submissions have surged, but reviewers cannot keep up. A researcher who has served as an area chair at a top conference said that in the past a reviewer's load was two or three papers, but now it is often ten.
Review quality is visibly declining. This year ICML conducted its first full public reproducibility audit: of 168 oral papers, only 8 could reproduce more than 80% of their conclusions, less than 5%. Hallucinated citations are not a problem unique to this controversial paper either; just this year, ICLR also saw AI-generated papers with fabricated references.
Citing a nonexistent reference requires no professional ability to verify, yet it passed through four gates—reviewers, AC, program committee, and awards committee—all the way smoothly, and finally appeared on the public version of a Best Paper.
Rather than blaming the dereliction of a particular reviewer, it is better to say that this is a systemic failure of the review system under the flood of submissions.
AI Technology Review noted that the "Awards Guidelines" reformed last year impose stricter recusal on the general chairs and PC to ensure the independence of award selection.
The AwC's responsibility is to "discuss and decide the Best Paper," and the document contains no provision for "verifying the authenticity of citations in candidate papers." This means that if citation problems are not stopped at the review stage, theoretically there is also no mechanism to intercept them at the award stage.
Beyond the procedural "non-verification," what is also worth noting is the actual carrying capacity of the award process. 17 Best Papers may be another side of the failure. From a single best to one per track, it appears to recognize more good work, but the actual situation may be: when "best" can be issued in batches, its value is diluted, and some checks that should be stricter ultimately become superficial.
Currently, regarding the controversial paper, ACM MM 2026 Program Co-Chair Yan Junchi confirmed to AI Technology Review that the chair group has received the retraction request email and will decide in the near future whether to make an official public response.
Will this storm, triggered by a "sophomore student + independent researcher" winning a top conference's best paper, ultimately end with the retraction of one paper, or will it become an opportunity to force top conferences to reform their review and award mechanisms? AI Technology Review will continue to follow.
NeurIPS 2026 is about to be held. Researchers, students, and readers interested in the frontiers of AI are welcome to join our exchange group.
Here, you can meet peers, exchange research, share admissions information, research collaboration, meal invitations, submissions, team formation for attending the conference, and more, and find potential partners.
During the conference, we will also continue to bring on-site reporting updates, interpretations of guest speeches, summaries of frontier directions, and forum highlights.
? Group entry portal: scan the QR code to join the group or add WeChat szhsn610, with the note: NeurIPS + institution/school + name + direction.


Sophomore "Genius Girl" Just Won ACM Best Paper, Then Found Herself in Fabrication Scandal; Corresponding Author Responds Exclusively

This is an original Leiphone article; reproduction without authorization is prohibited. See the reprint notice for details.