SIGMOD 18

Eric Brewer, VP of Infrastructure at Google and faculty member at UC Berkeley, talking about the effect of container technology on cloud computing, and Pedro Domingos, Professor at University of Washington, talking about machine learning-what works, what doesn',t, and where the field is headed. Like last year, the keynotes are followed by a plenary session of teaser talks, where each presenter gives a one-minute summary of their paper, to give attendees a high-level view of the conference and help them decide which sessions to attend. There are two changes in the session organization from recent years, whose goal is to make the program more compact and interesting for attendees. First, tutorials are presented during the main conference on Tuesday through Thursday, rather than on Friday after the main conference is over. Second, to ensure there are at most four parallel sessions in each time slot, each research paper presentation is allocated either 20 minutes or 10 minutes. The decision of long vs. short presentations had several phases. During the reviewing process, PC members were asked to recommend whether each paper, if accepted, should be a long or short presentation. Then research PC group leaders made a recommendation for each of the accepted papers they supervised -- definitely 20 minutes, 20 minutes if there',s time available, borderline, or definitely 10 minutes -- based on reviews, reviewer discussions, and their own judgment, without knowing the identity of authors. Their recommendation is not necessarily a quality metric. They recommended ',definitely 10', for some papers highly-rated by reviewers, because the topic was narrow, could be explained in 10 minutes, or couldn',t be explained in 20 minutes so extra time wouldn',t help. For borderline papers, the final decision was based on many factors, such as topic diversity, institutional diversity, and the time available in the relevant session.