NeurIPS 2020 Sharing
We can all agree that 2020 has been a wild year. Even though being under such a tough condition, there’s still some silver lining and NeurIPS hosted from Dec 6th ~to 12th being one of them. This the first time for me attending NeurIPS conference. Usually a ticket to NeurIPS is quite difficult to get. NeurIPS 2018 tickets were sold in 12 minutes and 2019 it converted to perform a lottery ticketing system. This year, it is hosted entirely online while simulating the poster sections, roundtables and live Q&A with Gather.town virtual avatars and Rocketchat where you can actually “meet” people.
Because this being the first year, I wan’t fully prepared for what to expect during the conference. This year, there were 7 invited keynotes, 7 virtual poster sessions, 16 tutorials, 60 workshops, 20 demonstrations, 16 competitions as well as multiple tracks of contributed oral and spotlight presentations.
With so much content, I found it impossible to keep track of what’s happening without constantly checking the main schedule. The good thing though, is that most sessions are both live and recorded. Even if you missed part of the talk, you can still catch up by dragging the video back. By doing this, you can attend both sessions even if they are hosted at the same time. It’s almost like entering parallel worlds while you can listen to several different sessions at the same time.
There’s also multiple social talks and events via Zoom and Gathertown where there aren’t recorded. I am happy that I didn’t miss those sessions. Especially the roundtable session from Women in Machine Learning Workshop taken place on Wednesday (12/9) 3:30pm EST. There were 50 roundtables to choose from covering varies topics in research or career life and advice. I attended Navigating the job search, Putting machine learning research into practice, and Non-traditional paths to machine learning. Each topic took 30 minutes and the event was from 3:30pm to 5pm so I can at most attend 3 topics. It was a really nice way to get interaction with mentors and I hoped there were more time since there’s still a lot more sessions I wish I could go to.
The entire conference was more research focused (at least compare to KDD, the conference I went least year), so the participants are mostly PhDs or working in research institutes/departments. I often found myself sitting with a roomful of PhDs asking to apply for internship at research labs so I do feel out of place sometimes. I feel like most people has years of experience in research and lots of publications while I just mostly read papers and filter out the closest ones that fit our needs and bring them into our ecosystem. The New in ML workshop is meant to help with people just beginning so that draws my attention. I found the Q&A session Prof. Anima Anandkumar (from California Institute of Technology and NVIDIA) quite inspiring.
The conference covers a wide range of research topics, but can be roughly categorized into four main trends:
- Self-supervised representation learning (Workshop: Self-supervised Learning — theory and practice, Orals & Spotlights Track 27: Unsupervised/Probabilistic, Orals & Spotlights Track 12: Vision Applications)
- Knowledge representation for abstraction and reasoning (Workshop: Object Representations for Learning and Reasoning, KR2ML — Knowledge Representation and Reasoning Meets Machine Learning, Graph mining at scale; Orals & Spotlights: Representation/Relational; Tutorial Track1: Abstraction & Reasoning in AI systems: Modern Perspectives)
- Neuroscience inspired AI (Tutorial Track1: Where Neuroscience meets AI; Workshop: BabyMind: How Babies Learn and How Machines Can Imitate)
- Fairness AI (Workshop: Explaining Machine Learning Predictions: State-of-the-art, Challenges, and Opportunities)
Below is some highlights that I will go over more in depth on separate posts:
- Workshop: Graph mining at scale by Google
- Workshop: Self-supervised Learning — theory and practice
- Oral: Visual Applications
- Tutorial Track1: Abstraction & Reasoning in AI systems
- Workshop: KR2ML — Knowledge Representation and Reasoning Meets Machine Learning
There’s also some cools sessions that might not be as main streams such as the Machine Learning for Creativity and Design 4.0 Workshop. Seeing people using machine learning to produce music, analyze dance movements and create artwork is really fascinating. That’s also something I’m personally very interested in.
