AboutNews
Prof. Dr. Sören Auer und Prof. Dr. Ziawasch Abedjan zu Gast beim PhoenixD Kolloquium

Prof. Dr. Sören Auer and Prof. Dr. Ziawasch Abedjan speak at PhoenixD Colloquium

© Sonja Smalian/PhoenixD

The next PhoenixD Colloquium takes place Monday, November 1st, 2021 from 10 am till 12 am at Room M11 in Building 1104 on Welfengarten Campus.    

Prof. Dr. Sören Auer from Technische Informationsbibliothek (TIB) will talk about "Semantic Research Data Management in the National Research Data Initiative (NFDI)". In this talk he will give an overview on the concepts and implementation of semantic Research Data Management for the National Research Data Initiative (NFDI). Auer will introduce vocabularies and ontologies for establishing a common understanding of research data and showcase their use in the context of the NFDI initiatives NFDI4Ing, NFDI4Chem and NFDI4DataScience. He will give an overview on three open technology components, ready to be used in PhoenixD:

  • Open Research Knowledge Graph (ORKG) for organising scientific contributions in a knowledge graph: https://www.orkg.org

Prof. Dr. Ziawasch Abedjan from FG Datenbanken und Informationssysteme, LUH, will adress the topic of "Democratizing Data Science through Example-Driven Data Preparation".

Data scientists spend about 80 % of their time preparing data. Data preparation encompasses various tasks including discovery, extraction, transformation, and cleaning. 

Most of these tasks need heavy user supervision in the form of predefined configurations, such as rules, parameters, or patterns. Defining these type of configurations makes most data preparation tools hard to use and less accessible. Machine learning techniques provide the opportunity to learn the configurations and ultimately the preparation task itself. However to define data preparation as a machine learning task, a model is required that can generalize from a small training dataset to very large data. In this talk, I will surface the state-of-the-art in learning-based tools to support the data scientists with a special focus on data cleaning. I will discuss how user supervision can be reduced to a handful of example corrections using effective feature representation, label propagation, and transfer learning methods. Our cleaning systems internally leverage an automatically generatable set of base detectors and correctors and learn to combine them for highest utility. In practice, with a small number of 20 user-annotated tuples, our systems outperformed state-of-the-art techniques.