Name of the participant: Zhen Han
Description of the IT research project: In the era of digitalization, corporates are looking to improve their production process using big data and artificial intelligence (AI). They aim to model data as effectively as possible for discovering useful information, making predictions, and supporting decision-making. A natural question is how to manage the enterprise knowledge and the data collected from the production process. Traditional artificial intelligence algorithms consider the input data as identically independent distributed. However, large quantities of data are in the form of graphs, resulting in it consisting of not only features vectors but also relations among entities. To this end, knowledge graph models provide a compact and comprehensive solution. Specifically, knowledge graphs represent related entities in the form of triples (subject, relation, object). Thus, domain knowledge can be illustrated as a network of real-world entities, i.e., objects or concepts. Moreover, the relations between entities might change over time. Thus, we need to not only model the structural dependencies between entities but also the temporal dynamics.
Existing approaches always deal with temporal information and stationary structural dependencies separately. However, many industrial scenarios require utilizing both static and dynamic multi-relational data to support decision-making and improve the industrial chain. For example, predictive maintenance is a key technology to minimize downtime. To achieve this, we need to model both the static machinery structural dependencies, e.g., hyponymy relations between different parts, the temporal information from the sensors on the parts, and the historical maintenance records of different parts. In the communication industry, the features of base stations, such as their maximum workloads and geographical positions, are stationary, but the user volume around the stations is time-dependent. Thus, to optimize the resource allocation and the network traffic, we need to analyze both static and temporal information.
Motivated by the above industrial challenges, the goal of this project is to use knowledge graphs to represent both static and temporal industrial data, which can power lots of reasoning tasks and improve a plethora of downstream applications such as predictive machinery maintenance, recommender system, demand forecasting in logistics, etc. Knowledge graph construction and reasoning are becoming more and more important and are penetrating lots of industries. In addition, this research project fits into the Federal Ministry’s Hightech Strategy 2025, whose goal is to put Germany at the top of the next technological revolution.
Software Campus partners: Ludwig-Maximilians-Universität München, HUAWEI
Implementation period: 01.04.2021 – 31.03.2023
