What to do when you spot a paper on arXiv with the same essential material as yours?

Question

What do you do when you've sent a paper on your fancy new algorithm to a conference, and before the conference has replied to you, you spot a newly submitted paper on arXiv on the same algorithm?

Possible reactions I can imagine:

1. You immediately submit your work to arXiv and/or open-source your code to "prove" you were working on it too (or at least as much as that might be worth at this point)

2. You just wait and see if the conference accepts it (but then what?)

3. You withdraw your paper entirely -- you "lost"

4. You totally ignore it -- it's not "official" until it's peer-reviewed, so you might still be "first"

Furthermore, who typically gets credit if:

1. Your paper is accepted, and is first to be published outside of arXiv

2. Your paper is declined, and is not first to be published outside of arXiv

---

Update

I'm reading the other group's paper more carefully (I'd only had a chance to glance at it yesterday, and was alarmed because several of the key words and concepts were exactly the same as ours), and it seems like they might not have discovered the same algorithm after all -- it's difficult for me to tell because their notation and terminology varies considerably from ours, but there's a chance that we've found different algorithms, even though several key concepts are the same. I'll continue looking into it, but just thought I'd mention this to add more context. At least now I'm a little bit more hopeful.

Answer 1

It seems to me that the best answer is some combination of 1. and 2. Because you submitted the paper for review before the other work -- call it paper X -- appeared on the arxiv, the community will readily believe that your work does not rely on paper X. (At first I wrote "completely clear that your work does not rely", but that's too strong: it's possible that you had some prior contact with the authors of paper X and learned about their work before it was published. But from your description that didn't actually happen, so no problem there.)

So you are in a fortunate situation: because you submitted the work to the conference before the arxiv posting, you have established your independent priority. The fact that the report hasn't come back yet has nothing to do with that. With respect to the submission, it would be reasonable to just wait for the report -- I am assuming that since it is a conference, it will come back within a month or so? If your paper is accepted, then you should include in the published version and also in your conference talk the information that similar (or the same...) work was independently done in paper X.

However it would be a good idea to write immediately to the authors of paper X and let them know about your work. If you are in a field where the conference paper will be supplemented by a later journal paper, then depending upon the degree of similarity you may want to consider a joint publication. If not, then your journal papers should cocite each other: this establishes that "you both have priority", which is certainly possible, and then both works should be publishable. (But in my opinion a joint paper is the better option if the work is very similar: does the community need two versions of the same work? Can everyone be counted on to know about and value the two works equally? Better to join forces: that seals it.) Depending upon the response you receive and the timing it might be a good idea to post your submission to the arxiv as well, with a note explaining the chronology.

I disagree with both 3. and 4. First, it does not matter who did the work chronologically first but rather that each work was done independently and before the other was published. There does not need to be a "winner" and a "loser" here: you can both "win". It is good that research communities operate in this way, much better than your option 4.: no one has control over which referee report comes back first or which paper goes to press first or anything like that, so if this were the standard it would be at the very least quite unfair and in fact open to all kinds of ethical issues and abuses.

Note: One of the comments asks whether the work was stolen. It seems that the only plausible way for this to happen is for there to be some collusion between the authors of paper X and either the conference organizers or the chosen referees of your paper. This type of behavior is in my experience extremely rare, so I don't want to address it in my answer.

Answer 2

This actually happened with a paper I worked on. We handled it by:

1. Immediately submitting our own version to arXiv, including a short mention of the other paper.
2. Informing the other authors of our result, and offering to write a joint journal paper.

Submitting your own version as soon as possible strongly suggests that it was an independent discovery, especially if the presentation is completely original. It also sends a signal that you're not trying to hide anything.

By writing a joint 'final' version, both parties can share the credit. In our case, the papers had been submitted to different conferences, so we thought a joint journal version would be the most appropriate. In the end, both papers were rejected from these conferences, but the other authors were able to strengthen the original result, while we generalized it. This meant that we were able to write a very strong merged paper, which was accepted to the most important conference in the area.

Answer 3

If your paper gets accepted in this conference, you win. The submission date is before the arxiv uploading and no one can claim you plagiarized the arxiv preprint.

If your paper gets rejected, you probably lost. In subsequent submissions you have to cite the original arxiv preprint, make extra effort and experiments to differentiate your work from theirs (by augmenting your original work) and claim that both works have reached independently to those parallel findings. Still, this lowers your work's novelty and might lead to another rejection. In that case, the other side might lost too, because your original rejection might also signify that the algorithm is not that seminal or important.

So, you should consider in what ways you can expand your work to actually provide novel content in comparison to the arxiv preprint, in case of rejection. In case of acceptance, you have nothing to worry about.

UPDATE: I really liked the other answers. Submitting to arxiv the OP's paper as soon as possible is probably the best thing to do. Also sharing co-authorship (in case of rejection) is of course the ethical / right thing to do and that is what the OP should do. But:

• Case 1. There is some foul play on the other side. In that case, they do not want to share co-authorship but patent / steal the idea. In that case, co-operation is not likely to happen

• Case 2. No foul play involved and the OP's paper gets rejected. The other side has already patented those results by their arxiv preprint. They may even already submitted the paper to another conference (many times that is when you upload preprints). Why would they share co-authorship? Would the OP share co-authorship if his paper got accepted? Will he include the other paper in the related work section (of course he should) in his camera ready version (in case of acceptance), when most of the results are identical? According to his comments he is not going to do that (when he has nothing to lose by that if his paper got accepted). Why does everyone assume that the other side will cooperate? These are serious questions that are easily answered on an ethical basis but the practical side is always more complicated. And what if the other side is more famous / established than the OP? Sometimes in that case they may even refuse co-authorship on that fact alone. Co-operation and co-authorship usually happens between similar / equal parties but they are harder to achieve when the other side has more leverage.

I really hope things work for the OP. But if his paper gets accepted he should definitely cite the other work and explain the situation in his camera ready version.

Answer 4

Many theorems, algorithms, fundamental scientific ideas, etc bear the name of (or are attributed to) more than one person. This does not always happen because these persons worked together. Sometimes it happens because it is established that they worked on the same issue approximately during the same period and/or published approximately during the same period. An example that I can immediately give from Economics/Econometrics is in the sub-field of Stochastic Frontier Analysis: in 1977 two papers were published independently, laying the fundamentals of the field. Almost 40 years later, they are still mentioned together, when the author wants to refer to those that initiated the whole thing. These papers are

Aigner, D., Lovell, C. A. A., & Schmidt, P. (1977). Formulation and estimation of stochastic frontier production function models. journal of Econometrics, 6(1), 21-37.

and

Meeusen, W., & Van den Broeck, J. (1977). Efficiency estimation from Cobb-Douglas production functions with composed error. International economic review, 435-444.

Your algorithm and the other algorithm may be "cousins", and the existence of both may have positive externalities on the research and professional paths of all involved, since it makes for a more vigorous "look here!" shout to the scientific world. I would even consider promoting the other paper alongside yours.