Heisenberg reflecting on the famous 1927 Solvay conference remembered that “the most intense discussions took place not during the conference sessions but over meals…” . Why is that most conferences suck so badly and can one do better?
The search for an antidote to conference ennui has been a decade-long journey for me. We have recently finished the 2nd Machine Learning JEDI workshop during which time I learned a great deal – like PCA for tensors and EMD analysis, which I did not even know existed beforehand. I also enjoyed myself very much, as always, but like most of the participants I suspect, am now exhausted — but in the good, “TEDache“, kind of way. This is the 15th JEDI-style workshop, school or conference we have run since early 2006. For years I have been meaning to write about it, why I think it works and how to run your own version, but have never got round to it because I was still experimenting, or perhaps I was just protective of something that has been an important part of my life. I think now is the right time to change that…
I will perhaps describe the history of how the JEDI concept was born elsewhere but here I will focus on the main idea. Boiled down to its essence the JEDI is an attempt to cure the usual conference woes: the one-way, top-down communication where sometimes 80% of the audience have their heads buried in their laptops because they are tired of poorly delivered talks and their bodies are aching after being in the same seat for hours on end with no movement except for the 30 minutes where they crowd around the coffee and biscuit table in a desperate search for relief and connection.
Typically conference organisers try to get round these horrible features by choosing a beautiful venue for the conference, only to force the participants to stay indoors for the majority of the day! In a way it is worse – you know there is beauty just outside but instead you are forced into dull, serial monogamy with a string of lacklustre speakers.
On the rare occasion that I attend “normal conferences” these days I am struck by how dysfunctional and painful they are for the most part, yet they continue as standard because of social norms. But can one do better? On our path to try to do better we have evolved, through trial and error, the following key elements to the JEDI recipe:
- Very few formal talks (< 15% of the total available time), but lots and lots of 2-way, 3-way and n-way informal discussions between participants.
- Participants work on a selection of open research problems. This is the core of the JEDI concept and ensures that people are really engaged in something that interests them. The fact that the problem is unsolved means the participants are doing real research, which is very exciting. Hopped-up on the excitement of the chase, participants learn using “just-in-time” methods and then immediately apply the learning to solve the problem in front of them. As Aristotle said “What we have to learn to do we learn by doing”.
- Research projects are done by teams that form spontaneously based on the interests of participants (everyone interested in project 1 works together, etc…). This mixes students, post-docs and faculty who then have to decide how to tackle the project together. Seeing how senior researchers tackle the problem provides a wonderful learning experience for younger participants. If enough progress is made during the JEDI, the paper is completed by the participants working long-distance, over the following few months or longer. Using this model eight papers have been published and numerous new research links made.
- Team building and bonding is actively encouraged as a direct outcome: long-term friendships and collaborations are great for humans and key for careers, especially for students who lack contact with top researchers. Typically all participants live and eat together in self-catered apartments and usually the cooking is done communally. In the recent Machine Learning JEDI 2 the dinners were made by the participants – we had a braai (barbecue), Indian curry and fresh pasta – as well as delicious breakfast burritos on the final morning. There is nothing like making a meal with someone to build rapport and break down formal barriers: while an MSc student may feel intimidated by a professor when it comes to science, they can probably match them in chopping an onion!
- Hold your workshop in a creative environment that nurtures creativity, hard work and social bonding (like walking on the beach) while getting away from distractions and “work associations”. Google’s policy of hiring good people and then treating them really well (not something typical of universities in my experience) inspired me to get out of the typical University department and go somewhere nice. Just as important, progress on JEDI research projects requires that the participants are fully engaged and dedicated to the project during the workshop. Trying this inside a department, where people can find you with trivial admin requests, kills the group dynamic and any real progress.
- The spirit of the workshop is based on the belief of the power of sharing and communal good. This is of course not new: Linux and the open source community is a wonderful example. However, in the realm of conferences it is still somewhat foreign. At their worst, conferences can be akin to a used-car auction in which car dealers try to sell their work to other dealers trying to do the exactly the same.
Of course many people have recognised the issues with conferences and this has spawned the “unconference” model. How is the JEDI model similar and different? Unconferences and JEDI share their community spirit and favour lots of participant interaction. Unconferences sometimes have a “hack” day where teams work to build something together. From this point of view JEDI workshops effectively are 80% hack while at many unconferences the hack component is less than 20%.
I was lucky to attend the famous Google Sci Foo Camp unconference this year which, like most unconferences, was almost all built around parallel discussion sessions with amazing participants from all spheres of science. My problem with unconferences in general is that they can feel a little superficial and random at times because although the agenda is created by the attendees, sessions are often hijacked by one or two people, and usually nothing deep is produced in the end. At JEDIs the focus is usually on trying to solve some hard but concrete problem and on really learning research skills and the difficulties this entails leads to connections that last long after the end of the workshop.
In the decade that we have been using the JEDI model participants have published the following papers:
- BBN constraints on scaling dark energy
- Age-dating of LRGs
- Photometric SNIa challenge and a followup paper
- Star-formation history of VIDEO galaxies
- Generalising the Fisher matrix
- Redshift remapping
- Bayesian inference for radio astronomy
The secret to getting publications out is to choose projects that are sufficiently well-defined, contained and small that the participants can see the end goal in sight, which motivates them to finish.
The JEDI model is very flexible: we have run them for undergraduates and inexperienced graduate students using simple problems in Mauritius, Namibia and Kenya, as half of a more standard cosmology school and for faculty at the SuperJEDI in Mauritius. I hasten to add that the JEDI model can definitely still be improved. I often leave them feeling a vague sense of dissatisfaction and a feeling that one could still do better. It is partly the hope that we will crack the code on the next one, along with the positive reactions of participants, as well as our own growth and enjoyment, that drives us to organise new events.
It seems appropriate to close with some quotations from a couple of the past participants:
“Thank you so much for organizing such a beautiful and original conference…I am also looking forward to test some of your ideas (like the speed meetings) as team building for the next edition of a school I am organizing. The whole conference has been a total boost in motivation. This is just so much more than I could expect from any conference” – participant at SuperJEDI 2013.
“The JEDI V conference was the most enriching experience I ever had so far.” – student at JEDI V
At the end of the SuperJEDI – the JEDI format conference targeted at established researchers – we asked participants to fill out a survey to rate various aspects of the conference on a scale from 1 to 5. Overall the SuperJEDI was rated 4.6 out of 5. How good is that? Well the SuperJEDI was held at a top Mauritian resort where the food, water sports and drinks were all included for free and which respondents rated as being 4.2, 3.5 and 3.9 respectively. Anytime your conference rating is beating the free food, drinks and water-sports on an island paradise, then you’re probably doing OK!
It is a pleasure to thank all the people who have helped organise and fund the JEDI workshops over the years and to thank the participants themselves, without which it would be a just an intellectual construct. There are far too many wonderful people to name but I would particularly like to thank Nadeem Oozeer who in the last few years has put in a great deal of effort to expand the JEDI into Africa: Mauritius, Kenya and Namibia, while bringing with him the true spirit of the workshops.
 The Strangest Man: the hidden life of Paul Dirac, Mystic of the Atom, Graham Farmelo.