Can We Get Serious?

This post represents my personal opinion (and only my opinion) with respect to turning STOC 2017 into a “Theory Festival.”

FOCS and STOC is where the entire theory community is supposed to come together for the benefit of the fruitful exchange of ideas between sub-areas of theory (which as Boaz mentioned, has been one of the sources of the community’s success). But this beautiful vision is threatened by two growing trends to worry about:

  1. More and more researchers in specific sub-communities abandon FOCS and STOC for the benefit of specialized venues that provide more effective exposure to their work.
  2. Even the FOCS-STOC experience gets fragmented as people tend to attend talks in areas closer to their own research (it is hard to know what to attend out of so many talks outside of one’s area).

The second issue could be addressed with a strong plenary program that would focus the attention of the entire community. This does not seem to be controversial and so I believe that STOC 2017 will have such a program, which is great. The second issue is much more problematic. We can no longer expect all of the strongest work in theory to be submitted to FOCS/STOC and I would think that having strong specialized venues should be encouraged. It is therefore difficult to expect researchers from area X that was “lost to FOCS/STOC” to attend the event just for the FOCS/STOC papers. In my mind, the only way to get them back is to offer a strong program in area X in addition to the plenary program that would allow them to stay connected with other areas of theory. This will require some effort and does not come for free as it will reduce the centrality of FOCS/STOC talks within the more general event. So this is the place to ask: are we really serious about a pan-theory event or are we content with keeping FOCS/STOC a place where a fraction of the theory community meets? Are we serious enough to give up a little bit of our entitled attention?

Currently, a large group of good intentioned individuals, led by the SIGACT executive committee, is considering designs for STOC 2017. This group is trying to be as attentive as possible to the voices coming out of the entire community. The problem is that in such a committee work it is very tempting to go for a conservative small change that will not upset anyone (but may gain too little). As the set of reasonable constraints and goals is impossible to completely satisfy, it is tempting to concentrate on not losing anything STOC already is rather than on gaining something new for the sake of what STOC can become. Being a romantic, I wish there was a hope to be just a little bit bolder!

To be less vague, let me conclude with the design of a pan-theory event that I would have chosen. This would be an event that would be more attractive for me to attend and much more attractive for researcher from the areas of theory that are underserved by STOC:

A STOC PC will select its program as usual (but with full flexibility regarding the number of accepted papers). An organizing committee will design the program of the theory festival. This committee will have representatives from the STOC PC but also from many other subareas of theory. The program will have three parts:

  1. A plenary session designed by the organizing committee and aimed to present only what the entire community should hear about.
  2. An afternoon session in many parallel tracks containing a collection of 3-5 days workshops (featuring STOC papers as well as other papers). These workshops will be curated by the organizing committee. For natural areas (various Algorithms areas, Complexity, ML, Econ, Cryptography, Quantum Computing, etc) representatives of the organizing committee will be setting up a workshop program (as they would do for an Oberwolfach/Simons/Banf … workshop). The committee can also accept other workshop proposals.
  3. An evening session that will include poster sessions (having guaranteed spots for all STOC papers).

The main challenge I see is to allocate a venue with many rooms for the parallel sessions. STOC papers will get a little bit less “formal attention” but will probably get on average more “substantial attention.” In particular, papers that deserve the attention of the entire community will be presented in the plenary session, while other papers will get the attention of their respective subcommunity (as they currently do). In addition, the poster session will give a more relaxed and interactive forum for cross-area interactions.

19 thoughts on “Can We Get Serious?

  1. I actually like your proposal better but for one concern having to do with a separate organizing committee. I’m repeating what Eric Vigoda commented on in the original thread which I think addresses this issue really well (pasted from original): “The fact that there will be a separate committee to decide which papers get longer/shorter talks is disturbing to me. Are they going to read all of the accepted papers before deciding? Are they going to read the reviews as well and get a summary of the PC discussion? Are their decisions going to be based on their selective and subjective knowledge of who they think is a good speaker? This means that a timely talk by someone at say MIT can have a big impact on their talk length.”

    It is good that the organizing committee will have some representatives from the PC, but may be the entire PC should select the accepted papers (or perhaps even some rejected ones :)) which will get plenary slots (and hence additional quality-points which will undoubtedly get highlighted).

    How about modifying your proposal to say the entire PC selects a pre-assigned number of papers for plenary talks (just like special-issue selections) and then passes the information to the organizing committee. The organizing committee then takes care of scheduling these papers and additional new ones in the plenary talks and assigning the remaining non-plenary accepted papers to other (3 or 4 say) parallel track sessions (with 20 min slots) as they seem fit.

    PS: Eric’s comment –

    1. The procedure for selecting papers to each one of the tracks is not an inherent part of the proposal, and it can be played with. The main thing is that when the PC decided if to accept or reject a paper, I think that it is better that this decision is made separately of any of the scheduling questions (in the same way that we are separating the decision on best paper awards and special issue till other the PC finalizes its accept/reject decision).

      But I really don’t see any issue here. You gave the question of a special issue which is a great example. This is handled by a subset of the PC which solicits input from the entire PC (and doesn’t necessarily require a PC vote, though there is no harm in my mind in having such a vote). In the same way, the organizing committee can have a substantial representation of the PC (including the PC chair) and ask for as much input as needed from the entire PC. How much time is allocated to STOC papers could be pre-assigned or flexible.

      I understand that the idea that speaker’s reputation will have an impact on plenary talk assignments is somewhat scary. But I think we should get over it. It is absolutely fine that the talent and investment of speakers will be rewarded, but more importantly, I think it is important to pay a bit more attention to what the audience needs rather than obsessing about the author’s rights (which should be preserved in the prestige of the publication venue and in the workshops and in the poster sessions). Of course, it is very important to give junior researchers a chance to prove themselves as speakers, and this can be an explicit goal of the committee. By the way, it is not clear to me that the STOC papers need to always be presented in the plenary session as a collection of short talks. At times I can imagine a talk of the sort “trends in X as reflected by the STOC program,” and these talks need not be given by the authors. As I said, it is OK in the plenary session to be a bit more audience-centric rather than author-centric.

    2. “and assigning the remaining non-plenary accepted papers to other (3 or 4 say) parallel track sessions (with 20 min slots) as they seem fit.”

      This is actually very different than my proposal. The remaining STOC papers will be scheduled as part of the workshops. So Cryptography papers in the Cryptography workshop, Computational Geometry papers in the Computational Geometry workshop, ML papers in the ML workshop etc. Hopefully, there will be more specialized audience in these areas than in a usual STOC because there will be an entire afternoon program devoted to these areas and others (that usually get very little attention), as well as the opportunity to mix with other areas of theory.

      1. Actually scheduling the remaining papers as part of workshops was what I had in mind too. Would it still be possible to get 25min slots for all the accepted papers as part of these multiple workshops? Seems like it should be possible to me.

  2. I’d just like to add a few last notes to my earlier comments:

    1. As Omer points out we need to be a bit bolder. In fact the committee is being hampered by its limited mandate: “Fundamental problems observed with STOC/FOCS, committee of notables asked to select new font for proceedings”. Ok, pardon the hyperbole in there, but I think it illustrates the core of why the actions from this committee will have a rather limited effect.

    2. I agree with Boaz’s premise, but disagree with his conclusion. Indeed STOC/FOCS is in the hands of very capable and talented chairs, hence it is locally optimal and from year to year it only needs epsilon changes. This however does not preclude that there are vast improvements beyond the self imposed epsilon neighborhood for changes. A much better global optimal lies out there. This is not speculation: those of us old enough witnessed that better STOC/FOCS in the late 80s, one that was attended by all and captured the vast majority of excellent papers in TCS. All we need to do is take the one big sized step to take us to that place, and no epsilon steps won’t get us there. I think this is also what Omer is trying to get at.

    3. Along these lines certain solutions and proposals are declared impossible without any particular reasoning behind. In reality there is no fundamental reason (other than lack of will) why STOC/FOCS cannot contain (most) all good theoretical works or publish about twice the number of papers without any loss of quality. Who knows, maybe we don’t want to go in that direction but there is nothing impossible about them.

    4. I hope a new committee is assembled with a much broader mandate:

    Do whatever it takes to keep STOC/FOCS as flagship, must-attend, best-of-the-best venue in TCS.

    A new committee that contains both notable TCSers that publish and those who do not publish in STOC/FOCS. One that contains people who are mostly authors/presenters and those that would be, if certain areas were once again include. A committee that also included people who would be mostly audience trying to stay abreast with the latest and greatest in the field. A committee that includes both senior and junior researchers, and, thus fixing a glaring omission, a committee with people from outside the USA.

    Once the goals are clearly stated as above, the solutions are fairly easy to see, and indeed they have all been posted already by many in the earlier thread, including members of the committee: collocate, have more invited presentations, increase the number of papers in ways that do not lower quality (e.g. by targeting neglected areas), add workshops, add tutorials, host the conference in cities that are within easy reach to all, both within the USA and from overseas, have poster sessions for students, etc.

    1. Just wanted to voice my support for Alex’s suggestions. He Seems to have the most sensible perspective on all this.

    2. Indeed Alex. This is a call to think big, to put the audience in the center and to be serious about diversity.

      I am all for an increase in number of acceptances, but I am not sure it is the answer in terms of diversity (it is more likely that we will just accept more of the same). But my suggestion allows for an almost arbitrary increase in the number of papers, and at the same time gives an important place to papers that appear in other venues (Econ Journals, Nature, Crypto or other venues that are dear to authors). By decoupling the roles of STOC as “journal of record” and “premier theory event,” as Sanjeev rephrased, the question of number of acceptances becomes a bit less important (and quite orthogonal).

  3. Let me try a brief rephrasing of what Omer is saying.

    Whenever any changes is suggested, people’s reactions tend to be colored by feelings like: “How does it affect me as a STOC author?” “How does it affect the tenure/promotion process in theoretical CS.”

    Omer is suggesting decoupling the roles of STOC as “journal of record” and “premier theory event.” So the PC could still accept 100 papers (or whatever number the community wants) to its proceedings. Then reconfigure the event format to be something different than just a collection of 100 20-min talks.

    FWIW currently our committee’s mandata/instructions from Sigact is to not tinker substantially with the figure of 95+ papers, with each of them getting 20-min talks. This somewhat limits what else is possible in those 5 days.

    So the core question for you from Omer is whether you agree/disagree with these ground rules that the core of the conference remains sessions of 20-min talks on all STOC papers.

    I’ll give my opinion in an upcoming post.

    Sanjeev Arora

      1. I also like Omer’s idea and, frankly, do not really find it too radical (which is good!). People interested in getting STOC/FOCS “stamp of approval” still get it, people wanting to learn “what cool have happened OUTSIDE of my area in the last (half a) year” have an opportunity to do that.

        Of course, the key point to discuss is the way the selection of plenary program is performed. After all, no matter how you perform such selection, it will make some accepted papers/authors seem more distinguished/recognized than others.

        This is obviously a very contentious issue. However, I don’t really think that there is a way to avoid having to do such selection, no matter how flawed and unfair it might seem.

        After all, at the end of the day, each one of us has a (very) limited attention span. It is impossible for us (or, at least, for me) to go to all the talks (even if there is no parallel sessions) and remain sane.

        So, one way or another, we *do* perform such selection individually and, e.g., the reputation of the speaker (as unfair as it might seem) *is* an important factor in this choice.

        In the light of this, I much prefer to have this selection be a result of a careful process that involves the whole (or a part) of the PC who have seen all the papers, than to do it on my own by browsing titles and abstracts of the papers.

  4. Raghu: The way I at least envision the plenary session is that apart from long survey talks from scientists both inside and outside TCS (including mathematicians, economists, systems researchers etc..) there would be about 25 short talks on recent works.

    Some of these would be works that appeared in the proceedings of the same STOC, but most would be works that appeared in other conferences such as FOCS, ICALP, SODA, CRYPTO, NIPS, PODS, etc. etc.. etc… including perhaps even works that were not yet formally published at all.

    These talks would be selected by a separate committee that might consult with the PC’s of these conferences but would not outsource the decision to these PC but rather try to construct a diverse program that would inform the audience of recent advances across all of theoryspace. It’s not about selecting the “top 25” papers but about assembling a holistic program. For example, if both the STOC, FOCS and SODA best papers are on the same topic, then this doesn’t mean that this committee needs to allocate three talks to them.

    Alex: While as a program chair I thought that my job was to not to radically change the conference, I agree with Omer’s sentiment that one can and should be bold with the event (as opposed to its role as a “journal of record” where it’s not prudent to make big changes quickly).

  5. I really like Omer’s proposal. The idea of having the PC select the ~100 STOC accepts is good – it preserves the status quo in terms of prestige/publication count/etc. But having afternoon workshops in which both STOC accepts and other invited talks would be presented seems like a clear improvement.

    As a STOC attendee from the field of quantum computing, I can’t recall having ever gained anything from the quantum computing session at STOC: there are four papers being presented and I know them already. But I’m also the only one attending this session, that the organizers felt obliged to put together because, well, there is this quota of 5 quantum papers we just got used to accepting. Now, turning this into a small workshop, with perhaps twice as many talks selected by a knowledgeable figure in the field would allow to include recent and more diverse (or more subjectively chosen, which is fine) work to be presented, and that would definitely raise my motivation for attending.

    The primary reason I attend STOC is to learn about works, and meet researchers, from *outside* my area. But if on top of that I could get a half-day’s worth of good talks focused on my area, well — cherry on top!

  6. Hi Omer. Not sure why you think we are not serious.
    In any case, I am not sure which sub-communities you have in mind (aside from Formal Verification) when you talk of such abandoning FOCS/STOC. My view is that we have all abandoned it, at least to some extent, once we realized that the PC is *not* focused on providing a good program but rather on assigning due credits. I assume you know my opinion on this matters — see, e.g.,

    Moving from the analysis to the suggested correction, I fear that the separation between the PC and the OP somewhat reflects the problem and is likely to worsen it. You envision a two-stage selection process such that in the first stage “good/worthy” papers are selected (supposedly ignoring the consideration of their fitting a program) and in the second a program is selected from among these. I think the crucial issue is the making up of a program and one should focus on it, and have one PC devoted to this. The PC should not deal with assigning due credits and/or recognition, but rather with selecting a program.

    (I ignore the suggestion that the 2nd stage be done with the participation of representatives of sub-communities who don’t frequent FOCS/STOC, since I’m not clear who you have in mind, and in any case I think it is not a good idea…)

    1. P.s.: FV is actually right in splitting, since TCS is really justifiably divided to TOC and TOP. The term “theory” by itself is almost meaningless, or can mean anything. E.g., a relatively recent breach of sociology calls itself “theory and critique”….

  7. Indeed Thomas makes a great point. The biggest worry in increasing parallelism is that it might lead to sparsely attended sessions. But there are already sparsely attended sessions in the current format. Often in the crypto or quantum session you see an audience of 20 or so experts plus a few more poor souls who wondered in the room by mistake and are having a hard time following the presentation, since the speaker knows the room is 80% experts. (This is tongue in cheek of course – e.g., in FOCS 2010 I was one of those “poor souls” in the quantum session and ended up learning of a cool problem connected to my work from the paper of Harrow and Montanaro.)

    Even in a more parallel session these 20 experts will be there, and maybe even more of them if there is a workshop on this topic and/or the plenary program also has some non STOC papers from CRYPTO or QIP. Moreover, it’s quite likely that at least one of the 25 plenary papers will be on each of these topics, so *every* attendee will learn some presentation (which is intended for a general audience) on these areas, and will have at least some awareness of what’s going on there.

  8. I like Omer’s proposal a lot. And the workshops could have even 30-minute talks, which I think would appeal to experts more. This is my favorite proposal, but let me comment about other issues.

    I oppose mandating a large number of acceptances, like 130. I think we should instruct the program committee to keep the acceptance threshold at a similar level minus epsilon, and they can accept up to 130 or some upper bound, but they can accept considerably less if they deem that appropriate. Keeping the quality high is more important than trying to increase attendance through increased acceptances.

    I’m open to trying a poster session. However, I’m not keen on a lot of 5-minute talks which are just advertising the posters. This sounds just as exhausting as 20-minute talks, with little time to explain interesting ideas. I think 10-minute talks could work much better; I thought they worked well at Oberwolfach. I think the current 20 minutes is also reasonable, but I’d probably vote for 30-minute talks in 3 parallel sessions. This is most similar to Omer’s suggestion, and for talks closer to my area, I’d certainly prefer 30 minutes. Besides, two 30-minute talks is probably less tiring than three 20-minute talks. I think any format will be tiring, but people should also be skipping some talks to chat with colleagues. I’m also fine with the program committee choosing different talk lengths for different papers, as long as each paper gets at least 10 minutes.

    But again, I’d really vote for Omer’s proposal, in which case a lot of the last paragraph is moot.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s