Retrospectives @NeurIPS 2019
Schedule
NeurIPS 2019 Retrospectives Workshop Schedule
Friday, December 13 Vancouver Convention Centre West 114 + 115
| Opening Remarks | 9:00-9:10 Video (7:25) | |
| Invited Talk - Leon Bottou | 9:10-9:30 Video (Talk 8:25, Q&A 25:45) | |
| Invited Talk - Melanie Mitchell | 9:30-9:50 Video (Talk 29:10, Q&A 45:05) | |
| Invited Talk - Zach Lipton | 9:50-10:10 Video (Talk 48:45, no Q&A) | |
| Coffee-break | 10:10-10:25 | |
| Meta-analyses | 10:25-10:55 | Video (1:00) |
| Invited Talk - Veronika Cheplygina | 10:55-11:15 Video (Talk 37:53, Q&A 51:46) | |
| Panel | 11:15-12:15 Video (55:45) | |
| Lunch | 12:15-13:45 | |
| Invited Talk - Emily Denton | 13:45-14:05 Video (Talk 00:00, Q&A 20:10) | |
| Invited Talk - Percy Liang | 14:05-14:25 Video (Talk 23:08, Q&A 44:30) | |
| Retrospective Lightning Talks | 14:25-15:00 | |
| Coffee+posters | 15:00-16:00 | |
| Invited Talk - David Duvenaud | 16:00-16:20 Video (Talk 00:00, Q&A 18:05) | |
| Invited Talk - Michael Littman | 16:20-16:40 Video (Talk 22:07, Q&A 40:15) | |
| Meetup-style brainstorming session | 16:40-17:40 Video (43:25) | |
| Closing remarks | 17:40-17:45 |
Panel
A discussion with Yoshua Bengio, Joelle Pineau, Melanie Mitchell, Gael Varoquaux, and Jonathan Frankle.
Meta-analyses
All Meta-analyses are for 10 minutes.
- Smarter prototyping for neural learning (Prabhu Pradhan) Video 14:25
- Advances in deep learning applied to skin cancer detection (Andre Pacheco) Video 25:45
- Unsupervised minimax (Juergen Schmidhuber) Video 1:30
Retrospective Lightning Talks
All Lightning Talks are for 5 minutes.
- An Intriguing Failing of Convolutional Neural Networks and the CoordConv Solution (Rosanne Liu) Video (47:00)
- Learning the structure of deep sparse graphical models (Zoubin Ghahramani) Video (53:00)
- Lessons Learned from The Lottery Ticket Hypothesis (Jonathan Frankle) Video (1:00:33)
- FiLM: Visual Reasoning with a General Conditioning Layer (Ethan Perez) Video (1:06:58)
- Deep Ptych: Subsampled Fourier Ptychography via Generative Priors (Farhad Shamstad)
- Deep Reinforcement Learning That Matters (Riashat Islam)
- DLPaper2Code: Auto-Generation of Code from Deep Learning Research Papers (Anush Sankaran) Video (1:13:30)
- Conditional computation in neural networks for faster models (Emmanuel Bengio) Video (1:19:19)
Description of the “Meetup-style” session at the workshop:
A large part of the advantage from hosting a workshop is to be able to leverage a diversity of expertise from the extended ML community and we believe that there is a way to do so that is very interactive and engaging via the proposed “meetup-style” session in the schedule. We believe there is a lot to be gained from getting feedback from a diversity of ML researchers and practitioners. With that in mind, this session is meant to be a facilitated discussion group that will bring together different stakeholders that can benefit from ML retrospectives. The aim of the session is to organically through discussion, around a prior set of questions, surface potential best practices and ways to improve the effectiveness and usefulness of ML retrospectives. We believe that this will be the genesis for a great community driven standard for improving the quality and dissemination of the research work that we all do. This session will be co-led by Abhishek Gupta, who has organized over 40 similar sessions through the Montreal AI Ethics Institute.
Proposed schedule for the one hour meetup-style session:
- 0:00-0:05 Introduction, format and how the session will work
- 0:05-0:10 Break out into small groups
- 0:10-0:40 Facilitated group discussion around a set of motivating questions
- 0:40-0:55 Short summaries from some volunteer groups
- 0:55-1:00 Wrap-up
Here is a link for the inspiration for the session along with thoughts on the ethos behind it here.