How do you anonymize code for a conference submission? [D]

The question arrived with the specific dread of someone who has done everything right and then noticed the one thing they hadn't. Months of work, a polished manuscript, tight figures — and then the GitHub link sitting in the related work section like a name badge at a blind date. Double-blind review at the major venues demands identity concealment, and the research community has developed reasonable habits for the paper itself: strip the acknowledgements, remove institutional affiliations from the draft, replace self-citations with neutral language like "in prior work." But code is a different problem. A public repository carries your commit history, your username, your collaborators' names, sometimes your institution's organization account. Anonymizing a paper without anonymizing the code is, at best, a technicality and, at worst, a desk rejection. The thread on MachineLearning surfaced what practitioners actually do: spin up a throwaway anonymous GitHub account, strip identifying information from the README, scrub the commit history before the upload, and use services like Anonymous GitHub — a tool built specifically for this purpose that proxies repository content through a neutral URL. The consensus, such as it is in a single Reddit thread, runs toward Anonymous GitHub as the path of least resistance, with the caveat that you should audit your code for strings, comments, and file paths that might contain your name or your lab's name embedded in the logic itself.

If confirmed as common practice across the major ML venues, here is what this means: the anonymization burden on researchers is substantially higher than conference submission guidelines typically spell out, and the gap between what the rules say and what reviewers actually expect has widened quietly over years of open-source norms colliding with double-blind tradition. Junior researchers — the ones least likely to know about Anonymous GitHub, least likely to have a senior colleague mention it in passing over coffee — carry the highest risk of an inadvertent identity leak. That asymmetry matters because double-blind review exists precisely to level the playing field, and an undocumented procedural knowledge gap carves a channel of advantage straight back to well-connected labs. Venues that care about the integrity of blind review should probably say, explicitly and in the submission checklist, what they expect researchers to do with their code.

Watch for whether major venues — NeurIPS, ICML, ICLR — update their author guidelines to include explicit code anonymization instructions ahead of their next submission cycles, and whether Anonymous GitHub's maintainers report any significant uptick in usage that would confirm this is a quietly widespread concern rather than one researcher's late-night panic.

NewsHive monitors these sources continuously. All signal titles above link to the original reporting.

Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →