Learning Causally Disentangled Representations via the Principle of Independent Causal Mechanisms

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Komanduri, Aneesh, Wu, Yongkai, Chen, Feng, Wu, Xintao
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912000249954304
author Komanduri, Aneesh
Wu, Yongkai
Chen, Feng
Wu, Xintao
author_facet Komanduri, Aneesh
Wu, Yongkai
Chen, Feng
Wu, Xintao
contents Learning disentangled causal representations is a challenging problem that has gained significant attention recently due to its implications for extracting meaningful information for downstream tasks. In this work, we define a new notion of causal disentanglement from the perspective of independent causal mechanisms. We propose ICM-VAE, a framework for learning causally disentangled representations supervised by causally related observed labels. We model causal mechanisms using nonlinear learnable flow-based diffeomorphic functions to map noise variables to latent causal variables. Further, to promote the disentanglement of causal factors, we propose a causal disentanglement prior learned from auxiliary labels and the latent causal structure. We theoretically show the identifiability of causal factors and mechanisms up to permutation and elementwise reparameterization. We empirically demonstrate that our framework induces highly disentangled causal factors, improves interventional robustness, and is compatible with counterfactual generation.
format Preprint
id arxiv_https___arxiv_org_abs_2306_01213
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Learning Causally Disentangled Representations via the Principle of Independent Causal Mechanisms
Komanduri, Aneesh
Wu, Yongkai
Chen, Feng
Wu, Xintao
Machine Learning
Learning disentangled causal representations is a challenging problem that has gained significant attention recently due to its implications for extracting meaningful information for downstream tasks. In this work, we define a new notion of causal disentanglement from the perspective of independent causal mechanisms. We propose ICM-VAE, a framework for learning causally disentangled representations supervised by causally related observed labels. We model causal mechanisms using nonlinear learnable flow-based diffeomorphic functions to map noise variables to latent causal variables. Further, to promote the disentanglement of causal factors, we propose a causal disentanglement prior learned from auxiliary labels and the latent causal structure. We theoretically show the identifiability of causal factors and mechanisms up to permutation and elementwise reparameterization. We empirically demonstrate that our framework induces highly disentangled causal factors, improves interventional robustness, and is compatible with counterfactual generation.
title Learning Causally Disentangled Representations via the Principle of Independent Causal Mechanisms
topic Machine Learning
url https://arxiv.org/abs/2306.01213