A0213
Title: Harnessing geometric signatures in causal representation learning
Authors: Yixin Wang - University of Michigan (United States) [presenting]
Abstract: Causal representation learning aims to extract high-level latent causal factors from low-level sensory data. Many existing methods often identify these factors by assuming they are statistically independent. In practice, however, the factors are often correlated, causally connected, or arbitrarily dependent. The purpose is to explore how one might identify such dependent causal latent factors from data, whether passive observations, interventional experiments, or multi-domain datasets. The key observation is that, despite correlations, the causal connections (or the lack of) among factors leave geometric signatures in the latent factors' support -the ranges of values each can take. Leveraging these signatures, it is shown that observational data alone can identify the latent factors up to coordinate transformations if they bear no causal links. When causal connections do exist, interventional data can provide geometric clues sufficient for identification. In the most general case of arbitrary dependencies, multi-domain data can separate stable factors from unstable ones. Taken together, these results showcase the unique power of geometric signatures in causal representation learning.
