Use What You Know: Causal Foundation Models with Partial Graphs

Arik Reuter, Anish Dhir, Cristiana Diaconu, Jake Robertson, Ole Ossen, Frank Hutter, Adrian Weller, Mark van der Wilk, Bernhard Schölkopf, [0.5em] University of Cambridge, Cambridge, United Kingdom · 2026

Read Use What You Know: Causal Foundation Models with Partial Graphs by Arik Reuter, Anish Dhir, Cristiana Diaconu, Jake Robertson, Ole Ossen, Frank Hutter, Adrian Weller, Mark van der Wilk, Bernhard Schölkopf, [0.5em] University of Cambridge, Cambridge, United Kingdom (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-14972.txt.

Description

Reuter and colleagues ask how causal foundation models can use what experts already know. Standard causal estimates are built for one set of assumptions at a time. Newer causal foundation models promise one unified step for discovery and inference, but they accept no domain knowledge, which can weaken their answers. The team shows how to condition such models on causal facts, from a full graph down to partial ancestral facts.

Their tests compare ways of feeding that knowledge in. Learnable biases placed in the attention step, joined to a graph-convolutional encoder, work best for both full and partial facts. With this conditioning, a general model matches specialists trained on single causal structures. The paper frames the result as a step toward all-in-one causal models that answer queries from data while still using expert knowledge.

Catalog record

CREATORArik Reuter, Anish Dhir, Cristiana Diaconu, Jake Robertson, Ole Ossen, Frank Hutter, Adrian Weller, Mark van der Wilk, Bernhard Schölkopf, [0.5em] University of Cambridge, Cambridge, United Kingdom
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE80 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-14972

Browse the full library · About the Tower of Babel