NeurIPS 2021 Tutorials

Meire Fortunato and Marc Deisenroth

An important part of the Neural Information Processing Systems (NeurIPS) conference is its established tutorial program. We are pleased to announce this year’s line-up of outstanding tutorials. This guest blog post is written by the Tutorial Chairs for NeurIPS 2021, Meire Fortunato and Marc Deisenroth, to shed some light on the tutorials’ objectives, format, topics, and decision making process.  

List of  2021 Tutorials

In 2021, we will reduce the number of parallel tutorial tracks and avoid overlap of tutorials and the main conference. Therefore, we decided to go for invitation-only tutorials. When selecting the tutorial speakers, we paid attention to diversity in topics and speakers. 

  1. Cesar Mattos (Federal University of Ceará), Felipe Tobar (University of Chile): The Art of Gaussian Processes: Classical and Contemporary
  2. Irina Higgins (DeepMind), Toni Creswell (DeepMind) and Seb Racaniere (DeepMind): Do structural priors still matter in the age of transformers?
  3. Karen McKinnon (University of California, Los Angeles), Andrew Poppick (Carleton College): Machine Learning and Statistics for Climate Science
  4. Maria Schuld (Xanadu & University of KwaZulu Natal): Machine learning with quantum computers
  5. Lilian Weng (Open AI), Jong Wook Kim (Open AI): Self-supervised learning: self-prediction and contrastive learning
  6. Shirley Ho (Flatiron Institute & New York University & Princeton University): AI for Science
  7. Timnit Gebru (Black in AI), Emily Denton (Google): Beyond Fair ML: Towards Just, Equitable, and Accountable Machine Learning Systems
  8. Toshiyuki Ohtsuka (Kyoto University): Real-Time Optimization for Fast and Complex Control Systems
  9. Vukosi Marivate (University of Pretoria), David Adelani (Saarland University): A journey through the opportunity of Lower Resourced Natural Language Processing — An African Lens
  10. Wee Sun Lee (National University of Singapore): Message Passing in Machine Learning

Changes in 2021

Tutorials on a single day

One of our leading objectives was to minimize scheduling conflicts. Therefore, we decided to 

  • reduce the number of parallel tutorial tracks from three to two 
  • avoid overlap with the main conference. 

The tutorials will therefore run on a single day, concurrent with the EXPO, but without overlapping with the main conference. With a 4-hours slot for each tutorial (tutorial, Q&A, buffer), this gives us room for 10 tutorials this year.

Early publication of tutorials

Our aim is to make the recordings of the tutorials available one week prior to the conference,   so that  people have the opportunity to watch tutorials in advance, at their own pace.

Tutorials by invitation only

In the last few years, there has been a mix between invited tutorials (last year: 8) and contributed tutorials (last year: 9). We definitely wanted to keep 6-8 invited tutorials, which guarantees a spread of topics and allows us to invite people who would (most likely) not apply themselves, for example, because their main research topic is not in core machine learning or because they consider themselves ineligible. Given that we only have space for approximately 10 tutorials this year, we opted for invitation-only tutorials. 

Goals for this Year’s Tutorial Program


The central objective of the tutorial program is to feature topics that are of interest to a considerable portion of the NeurIPS community. At the same time, we aim to have minimal topic overlap with the tutorials at the NeurIPS, ICML, and ICLR conferences in the last 5 years. To compose a broad list of interesting and emerging topics, we looked at recent workshops at different machine learning conferences. Finally, we seek to have a program that includes fresh perspectives from speakers outside of our core community. This is important, because the NeurIPS community is expanding, and we are seeing interdisciplinary research, in which machine learning methods are being used in other fields, as well as methods and ideas from other fields being applied to machine learning. We wanted the tutorials to be an opportunity to promote this flow of ideas. 


Equally important as the topics, the tutorial speakers should be experts in their respective areas. They should be experts who know the topic inside and out, and have demonstrated the ability to share that knowledge in understandable language and clear presentation. They should present information deep enough that even those in the audience, who are active researchers in related fields, would still learn a lot. In addition, the tutorial speakers should provide a broad overview of a topic, which goes beyond their own research, because the tutorials are not focused research talks, but rather comprehensive lectures that encompass multiple perspectives. That’s a lot to ask! 

Finally, when considering speakers, we were looking for researchers whose careers can benefit from a NeurIPS tutorial (e.g., early/mid-career faculty, pre-tenure) or whose voices are unfamiliar to the broader ML community.


Prioritizing diversity and inclusion in ML is an important commitment. Diversity goes beyond familiar dimensions, such as age, seniority, gender, race, nationality, and current geographical location. Diversity also means enriching the ML community by offering a broad range of perspectives on how to tackle challenges within ML and think about the implications for society. This includes prioritizing speakers and topics that have not received as much attention in recent years, but also embracing perspectives from outside the mainstream ML community, which brings viewpoints that are of interest and benefit to the ML community. Diversity was therefore a key factor in the selection of both tutorial speakers and topics. 

Selection Process of Invited Tutorials

Both tutorial chairs independently drew up a list of potential speakers and shared the list with each other. In coming up with our lists, we looked at the last years’ publications, workshops, and tutorials presented at NeurIPS and related venues. We also asked for recommendations from colleagues. We then discussed each candidate from both lists. In analyzing the potential speakers, we reviewed their papers to understand their expertise and watched their talks to appreciate their presentation skills. We also took time to (relatively) get to know candidates, who are not so much in the focus of the ML community. In making our shortlist, we ensured that the candidates spanned a wide range of topics and did not retread recently covered ground. 

We emailed the General Chair, Diversity & Inclusion Chairs, and the rest of the Organizing Committee for their comments on this shortlist. Following a few adjustments based on their feedback, we emailed the potential speakers and asked whether they would be willing to give a NeurIPS tutorial. All of them accepted the invitation.

The tutorials are an integral part of NeurIPS and play a unique role in the ML community. We are very grateful to everyone who helped us shape this year’s tutorials. We are extremely honored to serve in this important role, and we hope that you are as excited as we are about this year’s tutorial program. 

Meire Fortunato (DeepMind) and Marc Deisenroth (University College London)

NeurIPS 2021 Tutorial Chairs