RoCA: A Robust Method to Discover Causal or Anticausal Relation by Noise Injection
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Causal or Anticausal Relation Discovery, Semi-Supervised Learning
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Understanding whether the data generative process is causal or anticausal is important for algorithm design. It helps machine learning practitioners understand whether semi-supervised learning should be employed for real-world learning tasks. In many cases, existing causal discovery methods cannot be adaptable to this task, as they struggle with scalability and are ill-suited for high-dimensional perceptual data such as images. In this paper, we propose a method that detects whether the data generative process is causal or anticausal. Our method is robust to label errors and is designed to handle both large-scale and high-dimensional datasets effectively. Both theoretical analyses and empirical results on a variety of datasets demonstrate the effectiveness of our proposed method in determining the causal or anticausal direction of the data generative process.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8394
Loading