eprintid: 10142668 rev_number: 6 eprint_status: archive userid: 699 dir: disk0/10/14/26/68 datestamp: 2022-02-01 14:11:22 lastmod: 2022-02-01 14:11:22 status_changed: 2022-02-01 14:11:22 type: proceedings_section metadata_visibility: show sword_depositor: 699 creators_name: Yang, Mengyue creators_name: Liu, Furui creators_name: Chen, Zhitang creators_name: Shen, Xinwei creators_name: Hao, Jianye creators_name: Wang, Jun title: CausalVAE: Disentangled Representation Learning via Neural Structural Causal Models ispublished: pub divisions: C05 divisions: F48 divisions: B04 divisions: UCL keywords: Science & Technology, Technology, Computer Science, Artificial Intelligence, Imaging Science & Photographic Technology, Computer Science note: This version is the author accepted manuscript. For information on re-use, please refer to the publisher’s terms and conditions. abstract: Learning disentanglement aims at finding a low dimensional representation which consists of multiple explanatory and generative factors of the observational data. The framework of variational autoencoder (VAE) is commonly used to disentangle independent factors from observations. However, in real scenarios, factors with semantics are not necessarily independent. Instead, there might be an underlying causal structure which renders these factors dependent. We thus propose a new VAE based framework named CausalVAE, which includes a Causal Layer to transform independent exogenous factors into causal endogenous ones that correspond to causally related concepts in data. We further analyze the model identifiabitily, showing that the proposed model learned from observations recovers the true one up to a certain degree. Experiments are conducted on various datasets, including synthetic and real word benchmark CelebA. Results show that the causal representations learned by CausalVAE are semantically interpretable, and their causal relationship as a Directed Acyclic Graph (DAG) is identified with good accuracy. Furthermore, we demonstrate that the proposed CausalVAE model is able to generate counterfactual data through “do-operation” to the causal factors. date: 2021-11-13 date_type: published publisher: IEEE official_url: https://doi.org/110.1109/CVPR46437.2021.00947 oa_status: green full_text_type: other language: eng primo: open primo_central: open_green verified: verified_manual elements_id: 1936690 doi: 10.1109/CVPR46437.2021.00947 isbn_13: 9781665445092 lyricists_name: Wang, Jun lyricists_name: Yang, Mengyue lyricists_id: JWANG00 lyricists_id: MYANB08 actors_name: Flynn, Bernadette actors_id: BFFLY94 actors_role: owner full_text_status: public pres_type: paper series: IEEE Conference on Computer Vision and Pattern Recognition publication: 2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021 place_of_pub: Nashville, TN, USA pagerange: 9588-9597 pages: 10 event_title: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) event_location: ELECTR NETWORK event_dates: 19 Jun 2021 - 25 Jun 2021 issn: 1063-6919 book_title: 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) citation: Yang, Mengyue; Liu, Furui; Chen, Zhitang; Shen, Xinwei; Hao, Jianye; Wang, Jun; (2021) CausalVAE: Disentangled Representation Learning via Neural Structural Causal Models. In: 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). (pp. pp. 9588-9597). IEEE: Nashville, TN, USA. Green open access document_url: https://discovery.ucl.ac.uk/id/eprint/10142668/1/2004.08697.pdf