Planning issues

This topic is to discuss any issues that arise during the planning phase

My first post here. Apologies in advance if what I am suggesting has already been discussed elsewhere and is well understood by folks, and apologies if I’m posting in the wrong place!

Twelve Hour Schedule Mirroring for Geo-Diversity

For much of computer science the geographical center of gravity is in the North Atlantic. On the other hand, the fastest growing regions are roughly 12 time zones away, in Asia. Achieving geo-diversity depends on us providing equality of access to participants regardless of their geographical location. I propose a very simple idea which I’ll call twelve-hour-mirroring (THM) which addresses the problem of geo-diversity in a way that is complementary to many of the other concerns associated with virtualizing conferences. The key insights are: a) that if we consider a ‘no-go zone’ within any 24 hour window during which it is unreasonable to expect anyone to attend conference events, e.g. 10pm-8am, then THM will ensure that there will be an ‘edition’ of each event that falls outside their no-go zone regardless of any attendee’s geography, and b) that there is an asymmetry of inconvenience between the presenter and the audience, both numerically (one presenter, large audience) and temporally (the presenter presents just once or twice, but the audience attends many events). My hope is that a mechanism such as THM can help us wind back some of the entrenched geo-biases in our community.

A few specifics…

Assumptions:

While I’m entirely open to entirely rethinking what a conference is, this simple proposal addresses a model of a conference that somewhat reflects our current conferences. I’m assuming that they are either fully virtual or hybrid (a physical conference that supports virtual attendance). I’m also assuming that non-interactive content is recorded, so can be duplicated at low cost. I’m further assuming that we have technology that supports interaction online, and that interaction happens formally through moderated Q&A, panel sessions, etc, and informally through ‘hallway’ conversations among attendees.

Examples:

A. A fully virtual conference constructs a 12-hour schedule where primary delivery (eg talks) are pre-recorded, and Q&A and panels are live. Speakers and attendees notify organizers of no-go zones ahead of time (noting that some papers will have no no-go zone due to geo-diverse authorship), and organizers will optimize the schedule accordingly. Authors might be required to be available for just one of the two Q&As for their paper. Keynote speakers would be carefully scheduled to maximize the chance that they can reasonably do two live Q&A’s. Likewise for panels (and/or the two editions of each panel may have different panelists). Sessions are chaired by people in time-appropriate geo-diverse locations.

B. A hybrid conference located in London schedules events from 0800 BST to 1800 BST (marked in green here). The conference has two general chairs, one based in London, the other San Francisco. The mirrored schedule runs twelve hours behind, from 2000-0600 BST, leaving 2 x 2 hour breaks. The keynotes are in-person, running from 0800-0930. The keynotes are recorded and replayed from 2000-2100, with the speakers doing a second live Q&A from 2100-2130. Paper presentations may be pre-recorded or in-person, with a recorded version shown 12 hours later. Paper Q&A’s are live and duplicated live whenever the authors are willing and able (which may not be difficult for geo-diverse papers). Activities in the second edition (2000-0600) are run from a geo-diverse location with oversight from the San Francisco-based co-chair (1200-2000 PST), for example.

There are many other ways it could pan out, of course.

A virtual attendee looks at the schedule and choses between first and second editions of talks according to what works best in their time zone. They engage on line and in hallway discussions with those who are online at the same time as themselves. Assuming eight hours of sleep, virtual attendees will find that 1/3 of events will be during their normal sleep time (so they’ll have no choice of edition) and for the remaining 2/3 events, they can choose which of the two editions suits their schedule best.

There may be events such as townhall meetings which are not straightforwardly mirrored live. In such cases, they should be scheduled for a time that reaches the biggest set of attendees and a recording replayed 12 hours later.

1 Like

Hi Steve. ICSE is doing 3 hour sessions in different time bands. We’ll be following the sun. Everyone will be able to participate in 2/3 of the conference without jetlag, and be synchronous with 2/3 of the participants.

The time bands are:

  • Pacific (midnight-03:00 UTC): East Asia, Oceania, Americas
  • Indian (07:00-10:00 UTC): Europe, Africa, Middle East, East Asia, Oceania
  • Atlantic (15:00-18:00 UTC): Americas, Europe, Africa, Middle East, West Asia

Hello,

Having had this discussion on Facebook with Jeremy Gibbons and being in the “odd one out” time zone, I strongly support the idea of having the speakers do “two talks” - it can be within 2 out of 3 ICSE 8-hourly segments or every 12-hours - but as long as there is an opportunity for people around the world to be able to do live Q&A/breakout discussions with authors or interested parties.

Cheers,
Alex.

Thanks Crista.

The ICSE approach avoids the geo-bias we see in the PLDI’20 program.

On the downside, it means everyone (equally!) will miss 1/3 of the program. Not great if that includes the sessions you desperately wanted to attend (eg I want my students to attend the PLDI’20 memory management session, which is scheduled for 2:00am-3:30am AEDT).

However, the ICSE schedule is entirely complementary with the 12-hr mirroring. :grinning:

The second edition mirror becomes:

  • Pacific (12:00-15:00 UTC)
  • Indian (19:00-22:00 UTC)
  • Atlantic (03:00-06:00 UTC)

Thus someone in Sydney would be able to catch the second edition of your Atlantic session (1pm - 4pm AEDT), having missed the first edition (1am - 4am AEDT). They’d be online and engaging with others who missed the first edition of that session.

This nicely illustrates one of my points. The THM proposal is complementary to many other ideas (such as what ICSE is doing) and it does ensure everyone has the opportunity to engage with the entire conference.

Thanks Alex.

The ‘magic’ of duplicating exactly 12 hours later (rather than 8 or some other number) is that because it is 180 degrees out of phase it ensures everyone gets an equality of opportunity to attend one of the two sessions regardless of geography (and it is also conceptually simple to think of 12 hour mirroring).

Thanks, Steve, for your careful argument. I appreciate the encouragement to think outside the box.

Some of the conferences happening now are following a “take the standard conference format, plop it into Zoom, and jettison the parts that don’t fit” approach. My sense is that they basically turn into jumbo-length webinars and no one is very happy.

Other conferences are getting a bit more creative: e.g., I’ve seen at least one recently where all talks were pre-recorded and made available (together with papers) a week ahead of the meeting; the actual talk slots during the meeting are used for a 4-minute summary of the talk followed by 20 minutes of live Q&A.

Getting back to your specific proposal, I’ve been thinking about it today and trying to generate reasons for not adopting something like it, and then seeing whether I really believed them :-). Here are some that I ended up not believing:

  • It imposes some overhead for speakers, who may have to get up one or more times in the middle of the night to give their talk. I don’t find this too persuasive, since it also saves a lot of jet lag and travel overhead, and besides it should be possible to schedule people’s talks so that only rarely will someone be forced to speak in what is actually the middle of the night for them.
  • It might dilute the energy of the conference (number of people online at any given moment interacting in either social or technical spaces) by spreading it over more hours. On the other hand, it will attract more people, so maybe this is a wash.
  • One might worry that people in each hemisphere will only interact with people in that hemisphere. But frankly, if the conference is in the middle of the night in one hemisphere, participation from that region is going to be small anyway. Might as well have the people from that region talking to each other (and maybe getting up early or staying up late to encounter others).

However, I do see one remaining downside that seems more serious. Running the whole conference twice, offset by 12 hours, entails a significant overhead for organizers, since there will need to be effectively a whole second organizing team that will be awake when the other is asleep, with all the attendant difficulties of scale and communication. This overhead may simply be too much for smaller meetings, and even for larger meetings it may not be worth the organizational effort if the community is very concentrated in one part of the world.

The exact numerical dividing line between “smaller” and “larger” and between “very concentrated” and “at least somewhat geo-diverse” is not clear to me!

Best,

  • Benjamin

P.S. Sorry for the formatting mistake toward the end of that — still learning how Discourse works!

Hi Benjamin,

Thanks for that thoughtful response. I apologize that the following is long.

Virtualization places us at a critical juncture with respect to geo-diversity. On one hand, virtualization could severely harm those in geo-poor regions. On the other, virtualization gives us the opportunity for geo-equality like never before.

I am optimistic that we’ll be able to take ourselves down the second of those routes, but I believe that it will not happen unless we make a very clear commitment to geo-diversity from the start. I agree that unfortunately this won’t come for free.

You might think I’m overstating things by suggesting that virtualization could bring real harm to those in geo-poor regions.

This week, I find myself faced with the not-at-all-hypothetical question of whether I should ask my students to stay awake from 10pm to 10am every day next week.

What do I do?

On the one hand, we have any semblance of care for my students’ wellbeing. On the other, the once-a-year opportunity to participate in PLDI, ISMM and the associated workshops, tutorials, keynotes, and amazing ‘ask me anything’ sessions, which run from 10pm to 10am next week.

Conferences are the life-blood of our academic community. It is where my students meet potential employers and collaborators. It is where they are inducted into the academic community. This is where I’ve had countless spontaneous interactions that have impacted my career and that of my students profoundly. To be deprived of that (due to travel expenses or time zones) is to be deprived of a crucial part of our research experience.

If I happen to live in a geographically unlucky area (along with about 2B Chinese, Indonesians, Japanese, Koreans, and others who reside in or near this timezone), I must make the call about whether I’ll attend the 2am memory management session or not, whether I’ll say no to invitations to participate at the conference, whether I’ll say no to requests to chair sessions at 4am.

The point goes deeper. There’s a broader question of our community’s consciousness to geo-diversity in our day-to-day activities. I routinely meet people at 6am and late in the evening, but I have elected to withdraw from communities that refuse to make any allowances for those of us who are neither in Europe or North America.

In response to your point about the costs of mirroring, I’ll make a few points:

  • In terms of having more than one team, the same is true for any geo-diverse approach, including the one Crista describes above.
  • The need for more than one team opens up more opportunities for geo-diversity:
    • Geo-diverse conference general chairs
    • Geo-diverse student volunteers (a huge boon for students in Asia)
    • Geo-diverse chairing of sessions etc.
  • Most of the activities will be duplicated via recording, which given the right technology should not be expensive in time or $.

I know you, Crista, myself, and others here are passionate about reducing the carbon footprint of our activities. I am optimistic that done well, this has the opportunity to greatly improve geo-diversity which will be a huge win for our community. But I can’t help but worry that the move may have precisely the opposite effect. The release of the PLDI’20 schedule made those worries very real.

I don’t wish to be prescriptive about how we achieve it, but I do hope that we’re able to commit to geo-diversity as an underpinning principle behind virtual conferences.

There’s a lot more to say, but I’ve written too much already.

Thanks again for reading and engaging.

–Steve

Hi Steve,

Yeah, it’s a tricky trade-off, between being geographically diverse, and having a feeling of there being one conference that everyone’s attending.

As Benjamin suggested, we may want to rethink how technical material is presented at a conference, e.g. replace traditional talks by pre-recorded material with longer live Q&A sessions. This reminds me of the “flipped classroom” approach to teaching. These might be more amenable to geo-dirversity, for example the conference could be a large series of small-group sessions, where authors suggest times they’re available to lead discussions of their paper, then people sign up? Authors could be encouraged to spread their sessions around (and indeed it would be in their best interest to do so).

The thing we’d lose by doing something like that would be the shared experience of going to the same talk, perhaps keynotes, award presentations, etc. would play that role?

Alan.

Hi Alan,

I agree that we’ll be doing a lot of rethinking of how we present things. Still, I think the 12-hour mirroring proposal is largely orthogonal to that.

My experience presenting during COVID was that we prepared the talk ahead of time and then handled Q&A live. I think those Q&As become particularly important.

In the absence of other concrete information, my base assumption was that for fully virtualized conferences, something along those lines would be happening. In the case of a fully virtualized conference, I would expect that the two 12 hour mirrors would be fully (or near fully) symmetric in terms of their status and engagement level.

For hybrid conferences, the talks could still be pre-recorded, or alternatively, recorded during the ‘first edition’ and then replayed during the second. Thus for hybrid conferences, the editions would not be fully symmetric, with the in-situ edition being the primary one.

I think the details are interesting, but the more substantive point is that I see ourselves at a tipping point.

If this community won’t make a commitment to geo-diversity, then I believe that we will not only lose a great opportunity, but will actually harm the diversity of our community (e.g. almost no one from my region will participate in PLDI this year).

–Steve

An off-topic aside on geo-diversity of conferences: I think one way to reduce travel in my community (SIGPLAN) would be to break the nexus between publication venue (PLDI, POPL, etc) and conference location and time. I could imagine having three SIGPLAN conferences a year, roughly four months apart, each in one of three geo-distinct areas (like Crista’s regions). Each conference would run for about a week and would have all the major SIGPLAN venues represented (so there’d be PLDI, POPL, OOPSLA at each of them). The conferences would all be hybrid/virtualized, with the tacit assumption that people would normally only travel to their ‘local’ one, and only do so annually (at most). There are a lot of interesting details, including how we avoid creating geo-silos, but I’ll save that for another discussion.

To streamline the conversation, I’d suggest that we ignore hybrid conferences for now. Virtual conferences are already terra incognita, and hybrid seems an order of magnitude harder to get right. We will need to figure them out too, but not for the next year or so.

Going back to fully virtual conferences, two big questions in my mind are:

  1. Is organizing a “looking glass conference” with a 12 hour offset is (as one might hope) "almost free” in the sense that all you need is to choose a looking glass GC that can work well with the main GC and then they can organize their own team of student volunteers, etc., or (as one might fear) does it become expensive because things don’t scale linearly (e.g. due to additional communication overhead)?
  2. What is the impact of the mirror event on the social aspects of the main event? Does it draw people away from the main event and thus dilute the pool of people that are online interacting at any given time, for example, or does making the whole thing twice as big make it harder to find people you want to talk to? Does it make the social spaces more exciting or more confusing? etc.
  • Benjamin
  1. I think it will take some work to figure out how to make things scale and how to run things smoothly. It will also depend a great deal on who is selected. Alex may have something to say about this. He (Wellington NZ) is working with Jan Vitek (Boston) on OOPSLA’21. I’m optimistic that if we put the right people in place we’ll get processes worked out that streamline things and make them robust to such issues. However it’s terribly hard to guess without first hand experience. This is the unexpected upside to COVID — we’re being forced to experiment where once we might have shrunk back for fear of failure.

  2. I think that it is inescapable that if we aim for geo-diversity we need to duplicate events (otherwise there will always be a non-trivial fraction of the community for whom the event is completely unreasonable). If we duplicate events, we dilute their focus. I don’t think we can escape that. The other side of the coin is that we’re making the event more accessible, so the total number of attendees should increase and be more diverse. However, I see no reason why the dilution factor need be worse than 50%.

PS. I agree with your call to streamline the conversation — it’s complex enough just discussing virtual! However, the question of hybrid conferences is actually an immediate and pressing concern. Just this week I’ve been in separate conversations with organizers of PLDI’21 and OOPSLA’21, both of whom need to nail down their plans. In the case of PLDI’21 it is very pressing. They’re making hard decisions this week forced by contractual matters involving large deposits etc, and this all rests on whether they’ll have traditional, virtual or hybrid conference next year.

PS. I agree with your call to streamline the conversation — it’s complex enough just discussing virtual! However, the question of hybrid conferences is actually an immediate and pressing concern. Just this week I’ve been in separate conversations with organizers of PLDI’21 and OOPSLA’21, both of whom need to nail down their plans. In the case of PLDI’21 it is very pressing. They’re making hard decisions this week forced by contractual matters involving large deposits etc, and this all rests on whether they’ll have traditional, virtual or hybrid conference next year.

I’ve also been involved in conversations with both groups. My strong feeling is that at least PLDI 21 should be virtual, not hybrid (physical-only is out of the question, IMO).

Why is hybrid hard? Think back to the last Zoom call you were in where a substantial number of people were in the same room but a larger number were remote. Was it any fun for the remote folks? Didn’t think so.

I think that it is inescapable that if we aim for geo-diversity we need to duplicate events (otherwise there will always be a non-trivial fraction of the community for whom the event is completely unreasonable). If we duplicate events, we dilute their focus. I don’t think we can escape that. The other side of the coin is that we’re making the event more accessible, so the total number of attendees should increase and be more diverse. However, I see no reason why the dilution factor need be worse than 50%.

I’m surprised you conceded 50%. :slight_smile:

It’s not actually clear to me that the dilution factor needs to be anywhere near that large. From the point of view of the main conference, one might imagine that everything could run pretty much as it would if there were no mirror conference except that when they arrive in the morning the discussion forums have gotten filled up with interesting exchanges between a bunch of people that those in the main conference generally don’t know.

Of course, if the mirror conference is that separate from the main conference, then one could argue there’s not much point in having them both at the same time or calling them the same thing: Only the presenters would actually have the chance to interact with people on both sides of the mirror. So I guess one would want to try to create more mixing, which I guess means more synchronous interaction. But doesn’t that defeat the point of letting people sleep at their normal times?

I find I am becoming confused about how a mirrored event would actually work, when it gets down to the details!

Oh, I agree. The social asymmetry of the hybrid model makes it harder than fully virtual. No question. It seems clear that those who are physically present will always have a very different (better) experience than those remote.

Oh, there’s a disconnect here.

Yes, the schedule is mirrored. But it is not the case that there are two separate conferences.

Each individual will participate in sessions according to their needs, picking and choosing among first and second edition of any given event.

Those closer to the primary time zone will presumably engage in ‘first edition’ more, and those at the antipode will likely do the opposite. However, most will pick and chose according to what works best for their timezone (and their own constraints and preferences).

As I type, a colleague (I’m on a Zoom faculty coffee chat) points out that they’re attending CVPR next week and that it is following exactly the 12-hour mirror model… I’ll be very interested to hear how it goes. Another colleague points out that rather than ‘geo-diversity’ we should perhaps use ‘chrono-diversity’.

Yes, the schedule is mirrored. But it is not the case that there are two separate conferences.

Each individual will participate in sessions according to their needs, picking and choosing among first and second edition of any given event.

Those closer to the primary time zone will presumably engage in ‘first edition’ more, and those at the antipode will likely do the opposite. However, most will pick and chose according to what works best for their timezone (and their own constraints and preferences).

Are you assuming here that each mirror runs for 12 hours each day? If this is the plan, then clearly we can expect a constantly moving window of participants dropping off to eat and sleep while others are waking up and joining.

But 12 hours is a long time — I would guess that an average conference-goer is going to be able to make themselves sit in front of a screen for maybe six solid hours at a stretch for five days in a row. So making each conference day last 12 hours means that people will be participating for around half of the events. Is this what you meant by 50% dilution factor?

(ICSE chose 12-hour conference days not because they wanted to force people to choose between missing events and burning out their eyeballs and sitting bones, but precisely to mitigate timezone issues. But dealing with timezones is what the mirror is for, so…?)

And if the main conference has six- or eight-hour days, then there’s a large gap before the mirror starts; it seems unlikely that very many people will want to straddle / pick and choose from both main and mirror. But maybe one needs to lay out the schedule in more detail to see if this intuition is well founded or not…

  • Benjamin

I fully agree that the “impossibility” of travel makes it much more important to accommodate different timezones. A 12h mirror of the live part of a conference seems like a very good idea, along with the idea of spreading the schedule over the whole 24h as I here ICSE plans to do.
This discussion did remind me of something else I sometimes miss: access to comments/discussions about articles. I hope in the future the ACM DL can associate with every article not only the PDF and surrounding artifacts but also the videos of the corresponding presentations, and an archive of the subsequent discussions.
The questions/comments/answers given live at a conference are currently only accessible to those few lucky enough to be present at the talk, but could be useful for the many people reading the paper years later.

1 Like

No, that’s not the assumption at all.

The model is simply that every live event is mirrored 12 hrs later, independent of the length of the schedule:

  • PLDI’20 has a 12 hour schedule, in which case the two mirrors create a full 24 hr schedule, with 12 hrs primary followed by 12 hrs secondary:
    • 12:00-00:00 UTC primary
    • 00:00-12:00 UTC secondary
  • ICSE (Crista’s post above) has 3 x 3 hr blocks spaced over a 24 hr window, which means primary and secondary are nicely interleaved:
    • 00:00-03:00 UTC ‘Pacific’
    • 03:00-06:00 UTC ‘Atlantic mirror’
    • 07:00-10:00 UTC ‘Indian’
    • 12:00-15:00 UTC ‘Pacific mirror’
    • 15:00-18:00 UTC ‘Atlantic’
    • 19:00-22:00 UTC ‘Indian mirror’
  • Fairly traditional 9 hour schedule with lunch break
    • 0900 - 1230 UTC primary
    • 1400 - 1800 UTC primary
    • 2100 - 0030 UTC secondary
    • 0200 - 0600 UTC secondary

The mirroring proposal can be applied to a schedule of any length.

Crista’s proposal has 9 hours of content per day. It works really well with mirroring. As you can see from the above, the participants would have quite a few ways to pick and choose among all three sessions, including dense (by taking three consecutive sessions), or sparse.

The mirroring approach addresses the timezone issue and is complementary to a schedule like the ICSE one. The ICSE approach has the nice property that all timezones are essentially equal (which is not the case with the PLDI schedule).

Maybe. However, most conferences run a longer schedule than that (typically 8-9 hrs with a lunch break etc in between).

I think most attendees would look at it the other way: with the mirroring they now get to chose which parts of the day they watch the conference. There are plenty of late risers who would be delighted to select a schedule that starts at noon. Earlier risers might get stuck in at 6:00am with a coffee. Some would prefer a really dense schedule, some would prefer 3 hour gaps. That’s the nice side-effect of the mirroring—not only do you fully accommodate people from all regions, you also give everyone twice the scheduling choice for 2/3 of the program, regardless of where they live.

As we see with ICSE, PLDI’20 and CVPR, in practice I expect conferences will run under different schedules if they are virtualized to what they used to do. The ICSE pattern is a nice one because it offers a lot of choice and does not (implicitly) treat some timezones as second-class.