Causal Invariance

Causal invariance focuses on identifying and leveraging causal relationships that remain consistent across different datasets or environments, improving the robustness and generalizability of machine learning models. Current research emphasizes developing algorithms that learn representations invariant to spurious correlations, often employing adversarial training or representation learning techniques to disentangle causal from non-causal features. This approach addresses limitations of standard models that fail when faced with out-of-distribution data or domain shifts, leading to improved performance in real-world applications such as robotics and visual recognition where data variability is common. The ultimate goal is to build more reliable and adaptable AI systems that are less susceptible to biases and better generalize to unseen situations.

Papers

August 17, 2023

Regularizing Adversarial Imitation Learning Using Causal Invariance
Ivan Ovinnikov, Joachim M. Buhmann
Adversarial Training Imitation Learning Adversarial Imitation Power Regularization Causal Invariance

July 4, 2022

Invariant and Transportable Representations for Anti-Causal Domain Shifts
Yibo Jiang, Victor Veitch
Domain Shift Causal Structure Causal Direction Dynamical Invariant Causal Invariance

April 26, 2022

Causal Transportability for Visual Recognition
Chengzhi Mao, Kevin Xia, James Wang, Hao Wang, Junfeng Yang, Elias Bareinboim, Carl Vondrick
Image Classification Visual Representation Causal Graph Distribution Generalization Visual Recognition Causal Dynamic Learning Causal Invariance

March 22, 2022

Out-of-distribution Generalization with Causal Invariant Transformations
Ruoyu Wang, Mingyang Yi, Zhitang Chen, Shengyu Zhu
Distribution Generalization Causal Feature Invariant Causal Feature Invariant Causal Explicitly Mentioned Causal Fact Causal Invariance

Causal Invariance

Papers

Regularizing Adversarial Imitation Learning Using Causal Invariance

Invariant and Transportable Representations for Anti-Causal Domain Shifts

Causal Transportability for Visual Recognition

Out-of-distribution Generalization with Causal Invariant Transformations