Harnessing Geometric Signatures in Causal Representation Learning

Explore the cutting-edge field of causal representation learning in this hour-long talk by Yixin Wang. Delve into the concept of extracting high-level latent causal factors from low-level sensory data, challenging the assumption of statistical independence among these factors. Discover how geometric signatures of latent causal factors can be leveraged to facilitate causal representation learning using interventional data, without making assumptions about distributions or dependency structures. Learn about the identification of latent causal factors up to permutation and scaling with perfect do interventions, and block affine identification with imperfect interventions. Gain insights into the unique power of geometric signatures in causal representation learning, covering topics such as motivation, identification of latent causal factors, correlated latent causal factors, learning with IOSS, and interventional causal representation learning. Conclude with key takeaways and participate in a Q&A session to deepen your understanding of this complex subject.

- Discussant Slide
- Introduction
- Motivation
- Identification of Latent Causal Factors
- Correlated Latent Causal Factors
- Learning Latent Causal Factors with IOSS
- Interventional Causal Representation Learning
- Takeaways
- Q&A