Quantifying Consistency and Information Loss for Causal Abstraction Learning

Fabio Massimo Zennaro, Paolo Turrini, Theodoros Damoulas
5/7/2023
cs.AI

No official code found. Best implementation: FMZennaro/CausalAbstraction (4 stars, GPL-3.0, last commit 2024-08-16).

Abstract

Structural causal models provide a formalism to express causal relations between variables of interest. Models and variables can represent a system at different levels of abstraction, whereby relations may be coarsened and refined according to the need of a modeller. However, switching between different levels of abstraction requires evaluating a trade-off between the consistency and the information loss among different models. In this paper we introduce a family of interventional measures that an agent may use to evaluate such a trade-off. We consider four measures suited for different tasks, analyze their properties, and propose algorithms to evaluate and learn causal abstractions. Finally, we illustrate the flexibility of our setup by empirically showing how different measures and algorithmic choices may lead to different abstractions.

View on arXivView PDF

Code Implementations(1)

FMZennaro/CausalAbstractionCommunity100%
GPL-3.0

Abstraction of Causal Models

40Python, Jupyter NotebookJan 14, 20232 years agoGPL-3.0

Similar Papers

Abstraction between Structural Causal Models: A Review of Definitions and Properties

89% match2022

Jointly Learning Consistent Causal Abstractions Over Multiple Interventional Distributions

87% match2023

Causal Optimal Transport of Abstractions

87% match2023

Causal Abstraction Inference under Lossy Representations

86% match1 repo2025

Causal Abstraction in Model Interpretability: A Compact Survey

86% match2024

Learning Causal Abstractions of Linear Structural Causal Models

85% match2024

Cite this paper

@article{zennaro2023quantifying,
  title  = {Quantifying Consistency and Information Loss for Causal Abstraction Learning},
  author = {Fabio Massimo Zennaro and Paolo Turrini and Theodoros Damoulas},
  year   = {2023},
  eprint = {2305.04357},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2305.04357v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2305.04357.svg)](https://codeofpaper.com/paper/2305.04357)

Discussion