Reflections on 2019 USENIX Conference on Operational Machine Learning (OpML ’19)

By Kurt Andersen, USENIX Liaison to OpML ’19.

On May 20th, USENIX was excited to hold the very first Operational Machine Learning conference: OpML ’19. As over 220 attendees gathered that morning, it became clear that this is a topic which affects a lot of different organizations, and that a lot of work in the domain is in progress in both industry and academia.

The chairs for the conference were Nisha Talagala and Bharath Ramsundar. You can read a bit more about the background and intent of the conference in the article which they wrote for ;login: magazine in the Spring 2019 issue.

As the hallway filled, the big topic of conversation was a general level of excitement to hear the morning’s opening keynote speaker, Michael Jordan from UC Berkeley. He was deemed the “most influential computer scientist” in 2016 by the Semantic Scholar project based on an analysis of the published literature and from the buzz amongst the conference attendees, people could hardly wait to hear what he would say.

Michael talked about the need to move beyond a task-based approach to ML to one built on a model of inter-cooperating agents based on a market (econometric) framework in order to unlock the next level of scale, and in order to respect the unique motivations of the people whom the machine learning processes are supposed to serve.

The morning continued with a mix of talks from academic researchers and industry practitioners alongside a select set of tutorials. After lunch, another highlight were two panels. The first brought together practitioners from LinkedIn, AirBNB, Facebook, Netflix, and Google to discuss scaling machine learning to handle real world problems. Joel Young from LinkedIn organized and moderated the panel. With the permission of the panelists, he recorded the session and has made the recording and a transcript of the discussion available at https://www.linkedin.com/pulse/how-experts-do-production-ml-scale-joel-young/. The second panel brought together experts from LinkedIn, Mosaic, and HealthIQ to discuss how regulation (GDPR & CCPA as examples) affects machine learning lifecycles as well as how they guard against bias in the application of ML

The afternoon was also packed with a long list of almost twenty additional talks. The combination of industry and academic content resulted in the sessions being full and the hallways empty until the evening poster session when attendees were able to discuss what they had heard with the presenters.

We are grateful for the hard work of the chairs and program committee, as well as the support of the six sponsor companies that made this inaugural conference possible with 61 proposals that resulted in a very content-rich program.

With USENIX’s Open Access policy, the academic papers can be found linked from the conference program page and the presentation slides from many of the sessions are also linked to the respective sessions. We are looking forward to future opportunities to bring together the community of interest in this rapidly evolving field. You can stay up to date by following @USENIX on Twitter, and connecting with USENIX on Facebook and LinkedIn.