Causal Structure and Representation Learning with Biomedical Applications
2511.04790v1
cs.LG, cs.AI, stat.ML
2025-11-11
Авторы:
Caroline Uhler, Jiaqi Zhang
Abstract
Massive data collection holds the promise of a better understanding of
complex phenomena and, ultimately, better decisions. Representation learning
has become a key driver of deep learning applications, as it allows learning
latent spaces that capture important properties of the data without requiring
any supervised annotations. Although representation learning has been hugely
successful in predictive tasks, it can fail miserably in causal tasks including
predicting the effect of a perturbation/intervention. This calls for a marriage
between representation learning and causal inference. An exciting opportunity
in this regard stems from the growing availability of multi-modal data
(observational and perturbational, imaging-based and sequencing-based, at the
single-cell level, tissue-level, and organism-level). We outline a statistical
and computational framework for causal structure and representation learning
motivated by fundamental biomedical questions: how to effectively use
observational and perturbational data to perform causal discovery on observed
causal variables; how to use multi-modal views of the system to learn causal
variables; and how to design optimal perturbations.
Ссылки и действия
Дополнительные ресурсы: