Joey Zhou

CS PhD student at UT Austin. RT, likes and comments are all generated by evil AI.

[Gautam Kamath @thegautamkamath]

Honoured that our 2016 paper, Robust Estimators in High Dimensions without the Computational Intractability, w/ Ilias Diakonikolas, Daniel Kane, Jerry Li, Ankur Moitra, Alistair Stewart, was awarded the 2026 Gödel Prize

This is the highest award for papers in theoretical CS. 1/7

[Joey Zhou @JoeyZhouUT]

@gavinrbrown1 Bold to assume your three page algorithm with the assumption of Gaussian noise and access to an argmin oracle won’t be abused by governments to commit anti-human crimes!

[Parsa Hosseini @paahrsa]

Such a great time to start reviewing a brand new paper (yours) in 3.5 minutes.

Minute 1: Upload your PDF to GPT.
Minute 2: "List 40 weaknesses."
Minute 3: "Now write the review. Markdown format please."
Minute 3.5: Copy-paste results back to OpenReview. Submit. Beach walk.

You spent 3.5 days writing it. I spent 3.5 minutes rejecting it. We are not the same.

[Joey Zhou @JoeyZhouUT]

@AryanMokhtari But it’s literally impossible to find so many qualified reviewers for ICML. Restricting number of submissions is reasonable.

[Mark Schmidt @MarkSchmidtUBC]

The first result of this type appeared in von Neumann's 1933-34 lecture notes. Here is a scan I made during my PhD.

[Joey Zhou @JoeyZhouUT]

@YujiaZheng9 Seems to be a very good result.

[Joey Zhou @JoeyZhouUT]

@txyyss 有反例属于期待之中吧，我感觉要是最后发现IUT是对的才是大新闻。

[Mengye Ren @mengyer]

𝗧𝗵𝗲 𝗡𝗲𝘂𝗿𝗜𝗣𝗦 𝗦𝗮𝗻𝗰𝘁𝗶𝗼𝗻𝘀 𝗗𝗲𝗯𝗮𝘁𝗲

The NeurIPS Foundation has announced that it will not accept or publish submissions from “US-sanctioned institutions,” linking to OFAC’s sanctions search tool. This formalizes a practice that surfaced in 2025, when NeurIPS rejected already accepted papers from authors at Russian institutions. On March 25, 2026, the China Computer Federation (CCF) issued an official statement strongly opposing the policy, called on Chinese computer scientists to refuse to provide academic services to NeurIPS and to refuse to submit papers, and stated that it would remove NeurIPS from its recommended venue list if the policy is not corrected. The debate is real, but the legal picture is more complicated.

𝗪𝗵𝗮𝘁 𝗡𝗲𝘂𝗿𝗜𝗣𝗦 𝗔𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗗𝗶𝗱

The policy ties restrictions to whether your institution appears in OFAC’s sanctions search. But that search tool returns results from multiple OFAC lists with fundamentally different legal consequences, and the public NeurIPS policy does not distinguish between them. That is the core problem.

𝗦𝗗𝗡 𝘃𝘀. 𝗡𝗼𝗻-𝗦𝗗𝗡: 𝗡𝗼𝘁 𝗔𝗹𝗹 𝗦𝗮𝗻𝗰𝘁𝗶𝗼𝗻𝘀 𝗔𝗿𝗲 𝗘𝗾𝘂𝗮𝗹

- The SDN List (Specially Designated Nationals) broadly prohibits U.S. persons from transactions with listed entities, including publication-related services such as peer review and publishing. Whether exemptions or licenses apply depends on the specific sanctions program and designation, so real cases still require program-specific legal analysis.

- Non-SDN lists carry narrower restrictions that vary by program. The most relevant here is the NS-CMIC List (Non-SDN Chinese Military-Industrial Complex Companies), which restricts certain securities transactions. OFAC FAQ 905 states that the relevant executive order “does not prohibit activity with entities listed on the NS-CMIC List that is unrelated to such securities, such as the purchase or sale of goods or services.” Accepting a conference paper is not a securities transaction. That does not settle every publication-related question, but it does make the case for excluding NS-CMIC-listed entities much weaker than the case for SDN-listed entities.

𝗪𝗵𝗲𝗿𝗲 𝘁𝗵𝗲 𝗣𝗼𝗹𝗶𝗰𝘆 𝗙𝗮𝗹𝗹𝘀 𝗦𝗵𝗼𝗿𝘁

NeurIPS frames this as compliance with “US laws and regulations.” That is only partly true at the level of its public explanation. For SDN-listed entities, refusing submissions is a defensible compliance baseline. For non-SDN-listed entities, the legal basis is much less straightforward and depends on the specific sanctions program. A submission from an SDN-listed Russian military research institute and a submission from Huawei’s AI lab do not raise the same legal question.
By pointing to the search tool without distinguishing between SDN and non-SDN results, the policy captures both clearly prohibited cases and cases whose treatment is much less clear under the same rule.

𝗪𝗵𝗮𝘁’𝘀 𝗗𝗲𝗳𝗲𝗻𝘀𝗶𝗯𝗹𝗲

Screening against the SDN list and refusing submissions from explicitly listed entities is a reasonable baseline. The default prohibition is real, and determining whether licenses, exemptions, or program-specific carve-outs apply is not something most conferences are well equipped to do case by case.

𝗪𝗵𝘆 𝗜𝘁’𝘀 𝗚𝗲𝗻𝘂𝗶𝗻𝗲𝗹𝘆 𝗛𝗮𝗿𝗱

Even with legal counsel, the complexity is real. The SDN list requires live screening. Chinese entities appear across multiple lists with different legal implications. Iranian institutions raise separate questions under publishing-related guidance and licensing practice. Any serious compliance system needs list-specific and program-specific analysis, not a single search result treated as conclusive.

𝗧𝗵𝗲 𝗗𝗲𝗲𝗽𝗲𝗿 𝗣𝗿𝗼𝗯𝗹𝗲𝗺: 𝗥𝗲𝗴𝘂𝗹𝗮𝘁𝗶𝗼𝗻𝘀 𝗡𝗼𝘁 𝗕𝘂𝗶𝗹𝘁 𝗳𝗼𝗿 𝗦𝗰𝗶𝗲𝗻𝗰𝗲

OFAC’s sanctions regime is a trade-enforcement framework. Academic publishing was never its central use case. Conferences are left to navigate rules built around blocked parties, services, licenses, and exemptions, while norms of open scientific exchange remain secondary. Open science survives only in the space between what is prohibited and what is authorized, with no institution specifically tasked with protecting that space.

𝗪𝗵𝘆 𝗕𝗼𝘆𝗰𝗼𝘁𝘁𝘀 𝗪𝗼𝗻’𝘁 𝗙𝗶𝘅 𝗧𝗵𝗶𝘀

Some researchers have called for boycotting NeurIPS in response, and the backlash now includes CCF at the institutional level. The frustration is understandable, but the underlying problem is larger than one conference. The regulatory framework is a major part of the difficulty. Boycotts may send a political signal, as the CCF statement clearly does, but they do not by themselves resolve the legal uncertainty or produce a more precise compliance standard.

𝗪𝗵𝗲𝗿𝗲 𝗗𝗼 𝗪𝗲 𝗚𝗼 𝗙𝗿𝗼𝗺 𝗛𝗲𝗿𝗲

As an immediate step, NeurIPS should clarify whether its policy applies differently across sanctions list types and programs. The OFAC search tool already allows filtering by list type. Non-SDN results should be evaluated based on the actual scope of the relevant program rather than treated as equivalent to SDN designations.

More broadly, OFAC should provide clearer guidance or safe harbors for nonprofit academic conferences. Professional societies could develop shared compliance resources so conferences are not each left to interpret sanctions law on their own. The research community should also push for rules that treat scientific exchange as something worth protecting in its own right, not merely as an afterthought to trade policy.

[Jiahui Liu @jiahuiteddy]

Fujitsu research (located in Santa Clara, CA) is hiring summer interns.
Apply if you are a PhD student in the quantum computing areas, especially if you are interested in quantum algorithms, quantum error correction and compilation technologies, etc:
https://t.co/FTvVzyeewM

[Joey Zhou @JoeyZhouUT]

@lrzneedresearch I messaged you.

[Joey Zhou @JoeyZhouUT]

@lrzneedresearch Right I will graduate soon, but my research doesn’t have real-world impact so it’s hard for me to get jobs.

[Joey Zhou @JoeyZhouUT]

@litu_rout_ @isro @UTAustin @GoogleDeepMind Congrats!

[蛋挞也是 Taco🎄 @ytiihyslqlabpow]

万推，求转发。
我朋友的视力现在正在急速衰退，左眼视野仅剩~10%，右眼~30%，且有严重的视神经萎缩。目前我的朋友在杜克大学就诊，但现在因为一直找不到病因，所以不知道接下来要怎么治疗。如果有朋友了解Johns Hopkins, Harvard, 或者 UPenn 的顶尖神经眼科专家推荐！请大家帮忙转发！🙏

[Joey Zhou @JoeyZhouUT]

@HeLiuLeo Right, and many good topics for research. Hope to visit your bookstore someday.

[Joey Zhou @JoeyZhouUT]

@aryehazan Which game?

[Joey Zhou @JoeyZhouUT]

I prefer a ML research scientist/engineer role but also open to other roles.

[Joey Zhou @JoeyZhouUT]

Hello everyone — it’s that time again, and I’m looking for internship opportunities this summer or fall. I’m a PhD student at UT Austin working on learning theory and active learning, and more recently I’ve been doing empirical work on label-efficient LLM post-training.

[Joey Zhou @JoeyZhouUT]

I’m open to relocation or remote work. Please feel free to reach out if you know of any positions that might be a good fit.

p.s. For more details on my research, please see my personal website. You can also find my CV there.