Covid-19 has been spreading rapidly since it was detected in December in Wuhan, China. I didn’t anticipate in January, when I first heard the news, that schools and offices would be closed in the US, Italy would go on to put the country under quarantine, travel between the US and Europe will stop, and stock markets will suffer greatest dip since 1980s. This is one thing about exponential growth that human intuition is often wrong. Current estimates for the reproduction factor (average number of people infected by a person) of Covid-19 is 2.56 which is greater than 1. This means the disease can rapidly infect a large number of people at an exponential growth rate. Conclusion, the disease is quite contagious. Further, so far the mortality rate from Covid-19 seemed to be much higher than normal flu, hence we should not mistake it for just another flu. Therefore, it is important for people to take precautions. Even if you value working from office/school or enjoying outside, you run the risk of catching the disease and spreading it unknowingly, possibly to older people or sick people who might experience fatal symptoms. This is one reason why young and healthy people should not dismiss Covid-19 lightly.
Many universities have moved their classes fully online for the remaining Spring semester, companies including Microsoft are recommending people in the affected areas (e.g., Seattle, North-East US) to work from home. Conferences are moving online too. New York Academy of Sciences (NYAS) just announced that they will hold their conference virtually (probably the first in their history) and ICLR, an important machine learning conference, will be held as a webinar.
I feel this is a great idea. Virtual classes and conferences are something that shouldn’t have needed a pandemic to happen. However, some people have apprehensions about virtual meetings and conferences that I want to dispel in the remaining post.
Hosting (Machine Learning) Conferences Remotely
This section is specific to machine learning conferences but might apply more generally. A conference has two formal aims:
- In-person peer review, where people would ask authors question about their paper.
- Exchange of ideas and fostering of collaborations
Both of these can happen virtually. Machine learning conferences consist of four things: (i) an oral session where a selected group of papers are orally presented, (ii) poster sessions where paper are presented as posters, (iii) informal chit-chat and food, (iv) panel sessions.
Hosting Oral Talks
This is quite easy. Imagine a software like Microsoft Teams or Google Hangout. Everyone joins the presentation from wherever in the world they want. They can see the slides, go back to the previous slides if they want, sync again the current one, record and see it back. During the QA session, people can ping the organizer with questions and random selection will pick 2-3 questions for each talk. No big deal!
Hosting Poster Talks
Poster sessions require a bit more work since many poster sessions take at the same time. One solution is to create a login for people where they can see all the poster slides at once. They can then click on a poster and they will be taken to the video session where the author is discussing that poster.
Panel Discussion and Chit-Chat
Panel discussions take place as oral talks. A common webinar will start where a preselected group of people have audio access and they can discuss a topic. People can post questions and moderators can select which questions to ask. For chit-chat, people can create public (or private) groups where others can join (or be invited) in a fixed time slot, to discuss a topic or just for banter. Sessions should be recorded to ensure no harassment occurs although private groups should not be made public or investigated unless there is a charge. Private discussions should be deleted after a week’s time.
Why Virtual Conferences are actually better
My main emphasis of this post is that virtual conferences should not be viewed as a sad alternative to a physical conference in case of a pandemic, but in fact should be viewed as a viable alternative which has many advantages over a physical conference. I list the main advantages below.
1. People from countries with weak passports do not suffer
If you are a Swiss national from a Swiss university, planning to go to ICML, you sit in an airplane a day before the conference and you fly there. You probably have never heard of visa or you get one when you land. However, if you are from a country with weak passport such as China and India (both countries have heavy attendance at conferences), then you probably first fly to another city to apply for visa. You generally pay $100-$300 for the visa fee and only if you get it do you get to attend the conference. Further, many people have travel restrictions like they cannot travel outside without renewing their visa for their host country etc. So many people end up skipping conferences. Obviously the majority burden of this falls on people from non-western countries and maybe for this, it has never been discussed much before.
2. Conference Registration Fee will go Down
Conferences are a rich person’s game, if you are not reimbursed. Machine learning conferences charge between $500-$1000. They are generally held in touristy places, e.g., ICML 2016 was held in Times Square (why?). This means they have to charge more fee to recover the cost.
Many professors have a simple rule: you do not get reimbursed for a conference if you do not have an accepted paper. This means students will have to pay from their pockets. Now think what that means for a student from a poor country who wants to get into machine learning. This essentially stifles knowledge dissemination.
3. Many More People can Attend Conference
A virtual conference should cost $0. Companies will be more than happy to provide conference softwares. I believe a small fee of $20 will be charged just to discourage spammers from joining the call and bringing down the system. But this means essentially many more students can attend the conference and interact with people. Again this is different from talks being made offline. Virtual interactive sessions are not the same as offline non-interactive content.
4. Oral Talks and Poster Sessions are More Accessible
I do not see any advantage of an in-person oral talk over a virtual session. If anything, not having to stare at a tiny screen from the back of a room is actually a good thing. You can look at the slides clearly, you can go back to a slide, go forward, pause, resume etc. This is much better, right?
Poster sessions are also benefitted. I remember that when doing posters, I would dread a spot closer to the toilet or in the corner. People would, therefore, skip talks to rush to the poster area to put their posters at the preferred spot. With virtual session that is a thing of the past 🙂
5. More Convenient for People
You do not have to travel: people with travel anxiety, differently abled people, or people with kids or sick partner feel more convenience. I feel newcomers will also be more at-ease asking questions virtually than having to make their way through a huge crowd to reach the microphone.
Overall, I have always questioned the material benefit of spending $2000, a whole week of traveling, to give a 5min talk or a 1hr poster session that happens at the same time as dinner. Sure meeting friends and exchanging ideas in-person is great but this should not be compulsory and not a reason to inconvenience a good number of people and to stifle knowledge dissemination. Nevertheless, I feel there is some cost to removing in-person meeting. This can be more than remedied by either hosting a physical conference and virtual conference together with people deciding what they want. Or, one can have many workshops around the area where people can meet each other.