E-Locus - Institutional Repository of the University of Crete

Home Collections School/Department School of Sciences and Engineering Department of Computer Science Post-graduate theses

Post-graduate theses

Current Record: 1 of 824

[Add to Basket]

Identifier

000463775

Title

Automated causal discovery library

Alternative Title

Βιβλιοθήκη για την αυτόματη εύρεση αιτιακών σχέσεων

Author

Ντρουμπογιάννης, Αντώνιος Χ.

Thesis advisor

Τσαμαρδινός, Ιωάννης

Reviewer

Τριανταφύλλου, Σοφία
Τόλλης, Ιωάννης

Abstract

In the realm of data science and statistics, understanding the causal relation- ships between variables is paramount for making informed decisions and predictions. Traditional methods of causal discovery, relying heavily on human expertise and manual analysis, pose significant challenges in terms of scalability, objectivity, and efficiency. To address these challenges, we introduce the concept of Automated Causal Discovery (AutoCD), a revolutionary approach that seeks to fully automate the application of causal discovery and causal reasoning methods. AutoCD is a python library, designed not only to replicate the analytical capabilities of an ex- pert human analyst but also to surpass them by leveraging advanced algorithms and computational power. The goal of AutoCD is to comprehensively deliver all causal information, including complex causal relationships and interactions, and to efficiently answer a user’s causal queries with precision. AutoCD represents a significant leap forward in the field of causal analysis by automating the intricate process of identifying causal relationships from vast and complex datasets. We describe the architecture of such a platform, which is built upon a robust framework that integrates state-of-the-art algorithms for causal discovery and reasoning. This architecture is designed to be flexible, allowing for customization and scalability according to the specific needs of various applications. Furthermore, we illustrate the performance of AutoCD on synthetic data sets, showcasing its ability to accurately and efficiently uncover causal relationships. The system’s general applicability is one of its core strengths, enabling it to be deployed across a wide array of causal discovery problems. From healthcare and epidemiology to economics and social sciences, AutoCD has the potential to transform how causal analysis is conducted, making sophisticated causal discovery accessible to a broader audience without the prerequisite of deep statistical knowledge. By automating the process of causal discovery, AutoCD not only enhances the efficiency and accuracy of causal analysis but also democratizes access to advanced causal reasoning capabilities, opening up new possibilities for research and decision-making across multiple domains.

Language

English

Subject

Artificial intelligence

Causality

Feature selection

Graphs

Machine learning

Python

Αιτιότητα

Γράφοι

Μηχανική μάθηση