Key Insights and Subtle Shifts Emerge from Massive NeurIPS 2018 Conference
The recently renamed NeurIPS conference draws nearly 8,000 attendees and highlights subtle architectural advancements like ODENets alongside a dominant theme of sanity checks for existing AI approaches.
The newly renamed NeurIPS conference brings together nearly 8,000 registrants in Montreal to explore over 1,000 accepted papers on artificial intelligence. Despite its massive scale and strong tech industry presence, the 32-year-old event remains fundamentally an academic science conference. The sheer volume of topics makes it impossible for any single attendee to absorb everything, leading to highly varied personal takeaways.
Unlike previous years that feature major architectural breakthroughs, this year sees more subtle but significant advancements. The standout innovation comes from Neural Ordinary Differential Equations, or ODENets, which earn top paper honors for their impressive performance in image recognition, continuous time models, and density estimation tasks. These continuous time models show particular promise for deep learning applications in video analysis and high-sampling-rate connected devices.
A dominant theme throughout the conference is the push for sanity checks to improve the fundamental understanding of existing AI approaches. However, the author expresses disappointment that the broader ethical issues facing the field, such as systemic bias, privacy concerns, and diversity, do not receive more prominent attention during the proceedings.