September 13, 2019
tl;dr: I have been too busy this year to run SOCML 2019, but there will be a "SOCMLx" in Canada. Next year I intend for there to be a SOCML 2020 and a full SOCMLx program.
First, some background for anyone who doesn't know what the Self-Organizing Conference on Machine Learning is: at lunch at OpenAI in 2016, several people discussed their complaints about existing machine learning conference. OpenAI CTO Greg Brockman suggested that I use OpenAI funding and the OpenAI office building to run a conference that avoided all of these shortcomings, so I teamed up with OpenAI operations manager Kate Miltenberger and ran the first Self-Organizing Conference on Machine Learning in fall 2016 in San Francisco. SOCML has several goals. Compared to other machine learning conferences, SOCML has a more diverse group of participants, both because of extensive outreach efforts and because of travel grants for participants from underrepresented groups. (Some of the other machine learning conferences have improved considerably in this respect since 2016). SOCML also has a strong emphasis on inclusion of all participants. The conference has a strong code of conduct ensuring that everyone feels safe and respected. SOCML has group discussions and poster sessions but no talks, so there is no division of participants into speakers and audience. Elite established researchers and newcomers to the field interact freely in all discussion sessions. SOCML is always kept small (all SOCMLs have had fewer than 300 participants) so that it is feasible to meet everyone there at least briefly. There are no proceedings and thus no peer review. At conferences with proceedings, it is common to find that most papers being presented are already well-known to most participants who are interested in them, because the papers have been on arxiv for months already. At SOCML, participants come to discuss whatever is most interesting to them in the present moment. Altogether, this focuses SOCML on the key benefit of in-person conferences: social interaction with the other people at the conference. To some extent, SOCML aims to recapture the human-scale, interactive, realtime feel of the smaller machine learning conferences before the exponential increase in popularity of deep learning.
SOCML 2016 was an experiment. Following overwhelmingly positive feedback, I wanted to repeat it. After my move to Google in early 2017, the Google university relations team took on nearly all the hard work of running SOCML. With relatively minimal guidance from me, Negar Saei ran SOCML 2017 in Sunnyvale and Rainier Aliment and Jennifer Ye ran SOCML 2018 in Toronto.
This year, I have changed not just jobs but careers entirely, moving from basic research and academic publishing to business leadership of product-oriented research and development. These changes mean I need to find a new model for running SOCML but have also meant that I have much more on-the-job learning to do than usual. On top of this, my wife and I welcomed a new baby to our family this year. I never consciously decided to not run SOCML 2019, but I now find myself in September with no SOCML ready to go---it is too late for people to make travel plans and arrive at a SOCML even in December. I know that SOCML means a lot to many people, and I'm sorry there won't be one this year.
I do intend for there to be more SOCMLs in the future. If you'd like to be part of the SOCML foundation, please contact me.
Besides the main SOCML, I also intend to launch a SOCMLx program next year. The name is inspired by the TED conferences: there is a main TED conference, but also TEDx regional conferences with separate hosts. The SOCMLx program will be similar: designed to help hosts around the world run SOCMLx events smoothly, efficiently, and with all the core values of SOCML, for their local region or institution. This addresses one of the key shortcomings of SOCML: to stay human-scale, SOCML is small, but that means not many people can participate in any one SOCML event.
This year will not have a main SOCML, but this year will have the first SOCMLx experiment, in Canada. Based on feedback from this event, I plan to launch the full SOCMLx program next year.
Again, I am sorry there will not be a global SOCML this year, but rest assured, this is a temporary interruption, and SOCML will be back better than before.