Bellairs Invitational Workshop on Causal Inference and Representation Learning

March 11-18, 2022


Causal inference provides a language and a set of tools and principles that allow one to combine data and structural invariances about the environment to reason about a collection of unobserved mechanisms underlying the data. Representation learning and deep learning are highly popular paradigms in machine learning that excel at finding representations of high-dimensional data that support statistical prediction tasks as long as the data are i.i.d., and in particular as long as the test distribution is identical to the training distribution. These models do not naturally support reasoning about the effect of interventions, extrapolations, transportability, and counterfactuals.

In this workshop, we hope to study the foundations of causal inference and its relationship with deep learning, representation learning, and how recent advances in these fields can help each other.

Confirmed participants
* Organizers
Background Reading
Participants are encouraged to consult the following references in advance of the workshop:
  • Bareinboim, Elias, et al. "On Pearl’s Hierarchy and the Foundations of Causal Inference". In “Probabilistic and Causal Inference: The Works of Judea Pearl” (ACM Special Turing Series), in press. [link]
  • Pearl, Judea. "The seven tools of causal inference, with reflections on machine learning." Communications of the ACM 62.3 (2019): 54-60. [link]
  • Pearl, Judea, and Dana Mackenzie. The book of why: the new science of cause and effect. Basic books, 2018. [link]
  • Schölkopf, Bernhard, et al. "Toward causal representation learning." Proceedings of the IEEE 109.5 (2021): 612-634. [link]
Scientific Program

The scientific program will take place from Sunday, March 13 to Thursday, March 17 inclusively. Each day of the workshop will consist of a morning lecture (9:30 am-noon) and an evening lecture (7:30 - 9:00 pm). The rest of the day will be left open for discussions and collaborations.

Day 1 (morning)

Introduction to causal inference

Elias Bareinboim [Slides]

Day 1 (evening)

Introduction to Structural Causal Models

Elias Bareinboim [Slides]

Day 2 (morning)

Causal Representation Learning

Bernhard Schölkopf [Slides]

Day 2 (evening)

Do-Calculus and Moving Across Layers of the PCH

Elias Bareinboim [Slides]

Day 3 (morning)

Causal Fairness

Elias Bareinboim [Slides]

Day 3 (evening)

Causal modelling with kernels: treatment effects, counterfactuals, mediation, and proxies - Arthur Gretton [Slides]

On calibration and out-of-domain generalization - Uri Shalit [Slides]

Day 4 (morning)

Transportability and Data Fusion

Elias Bareinboim [Slides]

Day 4 (evening)

Open Problem Session

Dhanya Sridhar, Chandler Squires, Victor Veitch

Day 5 (morning)

Causal Reinforcement Learning

Elias Bareinboim [Slides]

Day 5 (evening)

Open Problem Session + Discussion

Eszter Vértes, Konrad Kording


Travel Protocols (COVID-19) Please see below for the various rules that are currently in place on the island. At the moment, you need to show a negative PCR or antigen test on arrival. If you need a test to go back home, there are a number of options very close to the workshop location. The easiest is to book an appointment online at the Lime Grove Lifestyle Centre in Holetown, which is a short (~10 min) walk away (book here). Other options are available (see p.24).

Venue The workshop will be held at Bellairs Research Institute of McGill University, Holetown, St. James, Barbados. See here for important information about the facilities.

For questions please contact