Work

My academic interest in particular rests in machine intelligence, computational data analysis, and information technology. In the future, I may also be interested in addressing commercial and financial problems in industries by drawing inspiration from computer science and applied mathematics.

Remainder: Part of work is also tied to my papers referring to the “Academic Papers” on this page.

Undergraduate Research Projects Report (2024)

Intro: A Report for Hybrid-Based Causal Discovery with Machine Learning

Different to my previous research papers, I adopt relatively informal language and flexible structures during writing. This report is overall meant to be written for (technical) readers who wish to make assessment on my undergraduate studies in the specialization of machine learning and causality.

Personally, I know my knowledge in causation will be inevitably dampen by time. Thus, at least I also want to preserve them via text. The lesson I learned from the writing is that I know little about causation, but I feel a modicum of comfort that in this report I still kept making progress compared to the previous writing I did.

Popularization of Causal Science (2023-2024)

Intro: A Primer on Causal Diagram Learning: Interpreting Causation from the Causal Discovery Perspective

Since I was motivated by fundamental ideas of causality from a popular science book named ‘The Book of WHY’ (by Turing Award winner Judea Pearl), I attempt to further create connections in the book to other celebrated books by giant leaders in causation, centering around subjects of “causal diagram learning”.

I want to show that ideas of Causal Discovery are deeply rooted in the past by standing on the shoulders of these giants. (e.g. Causation, Prediction, and Search [2001] (Peter Spirtes, Clark Glymour); Causality [2008] (Judea Pearl); Elements of Causal Inference [2017] (Jonas Peters, Bernhard Schölkopf, etc))

Causality Algorithms Programming (2022-2023)

Intro: Cadimulc: Light Python Package for Hybrid-Based Causal Discovery

CADIMULC is a open-source github repo standing for CAusal DIscovery with Multiple Latent Confounders, which provides easy-to-use APIs to learn an empirical causal graph from raw data with relatively efficiency.

For example, It integrates the implementations of hybrid-based causality approaches such as the MLC-LiNGAM algorithm, along with the “micro” workflow of causal discovery (Causal Inference), such as data generation, learning results evaluation, and graphs visualization.

Research in Causal Science (2) (2023)

Intro: A Survey on Causal Discovery with Incomplete Time-Series Data

With the rapid growth of massive time-series data, inferring temporal-based causal relationships from data — Temporal Causal Discovery (TCD) — has become an important and challenging task in recent years, holding both scientific significance and commercial value for uncovering data generative mechanism.

Distinguishing from existing reviews, we focus on the latest research progress in the task of TCD with incomplete data, summarizing the philosophies and paradigms reflected in current research methods.

Research in Causal Science (1) (2022-2023)

Intro: Non-linear Causal Discovery for Additive Noise Model with Multiple Latent Confounders

How should we appreciate “causal structures” underneath the data from a complicate learning environment? An environment in which “generic data relations” are prone to be non-linear, and even impacts from the multiple unknown factors are persisting

Existing solutions towards generic causal discovery might be either theoretically elusive in formal representation or notoriously difficult in algorithmic computation. Such motivations have driven us to design a theory-guided and practically effective causal discovery algorithm.

Academic Papers

Publishments

Preprints

  • Chen, X., Chen, W., Cai, R., 2023. A Survey on Causal Discovery with Incomplete Time-Series Data. In Xuanzhi’s Personal Website. [paper] [slides]

  • Chen, X.*, Chen, W.*, Cai, R., 2023. Non-linear Causal Discovery for Additive Noise Model with Multiple Latent Confounders. In Xuanzhi’s Personal Website. [paper] [slides] [talk] [code]

  • Chen, X., 2023. Supplementary Material to: “Non-linear Causal Discovery for Additive Noise Model with Multiple Latent Confounders”. In Xuanzhi’s Personal Website. [paper]

Personal Essaies

  • Chen, X., 2024. A Report for Hybrid-Based Causal Discovery with Machine Learning: Inferring Causation Under Generic Conditions of Nonlinearity and Confounders. In Xuanzhi’s Personal Website. [paper]

  • Chen, X., 2024. A Primer on Causal Diagram Learning: Interpreting Causation from the Causal Discovery Perspective. In Xuanzhi’s Personal Website. [paper] [slides] [talk]