
With Daniel Manela and Linying Yang!

Oxford lecturer and statistician who just wants to tell the world about numbers. Especially when causality is involved.

With Daniel Manela and Linying Yang!

Please join us at our UAI virtual poster on: Testing generalizability in causal inference July 22nd, Poster session 1, 16:00-18:30 BRT, 19:00-21:30 UTZ

Using simulations and statistical testing, our framework is robust and avoids over-reliance on conventional metrics, providing statistical safeguards for decision making.

- A testing framework that statistically evaluates the transportability of both mean and distributional regression methods, and complements MSE.

- A statistical framework to evaluate the generalizability of high-dimensional causal inference algorithms by targeting low-dimensional causal margins.

In this paper, we propose a statistical method to test model generalizability. Our contributions include - A method to simulate (semi-)synthetic data across two domains, with explicitly conditional outcome density and the causal margin of target domain.

**Poster Alert** Generalizability is a common problem in causality, as we often want to generalize causal findings from one domain to another. However, current metrics like MSE can only do model comparison rather than answer directly whether a model is generalizable.

Validating causal models often relies on simplified assumptions. Frugal Flows address this by using normalising flows to generate datasets that exactly meet user-defined causal constraints, ensuring rigorous benchmarking.

Frugal Flows generate synthetic datasets with: •Realistic data patterns •Customisable ATEs and treatment heterogeneity •Unobserved confounding effects These features make them ideal for benchmarking causal models.

Frugal Flows allow for inference, and offer a new approach to causality by: •Leveraging normalising flows for flexible data generation •Exactly parameterising ATEs and confounding effects •Bridging machine learning and causal inference

Excited to share our latest research: Marginal Causal Flows for Validation and Inference. We introduce Frugal Flows—a normalising flows based generative model that directly parameterises causal effects, enabling realistic causal dataset simulation. arxiv.org/abs/2411.01295

What’s purple and commutes? An abelian grape.
What's your favorite math joke?

What's your favorite math joke?

Everyone knows it’s gold!
The question of identity that is tearing the Lib Dems apart: is the party yellow or orange? All Britons Yellow: 57% Orange: 24% Lib Dem voters Yellow: 55% Orange: 41% ygo-assets-websites-editoria…

Postdoc positions in Oxford to work with me, @cwcyau, Chris Holmes, George Nicholson and others. Very exciting project! (Advertised through psychiatry, but jobs not tied to that area!)
New Oxford-Aarhus-Copenhagen collaboration - not yet launched - 2 postdocs, 2 senior postdocs and postdoc coordinator in Oxford. Jobs 173031-3 at jobs.ox.ac.uk Methods-focussed in group Holmes, Visscher, Evans, Nicholson, @cwcyau, @ten_photo etc, Denmark ads soon

New Oxford-Aarhus-Copenhagen collaboration - not yet launched - 2 postdocs, 2 senior postdocs and postdoc coordinator in Oxford. Jobs 173031-3 at jobs.ox.ac.uk Methods-focussed in group Holmes, Visscher, Evans, Nicholson, @cwcyau, @ten_photo etc, Denmark ads soon