PROVIDER
The collaborative project PROVIDER develops an AI-supported early warning system for the early detection, analysis, and simulation of potential supply bottlenecks in Germany. The focus is on supply chains and supply systems that are relevant for everyday supply security, but do not necessarily belong to classical critical infrastructure. Many of these goods, logistics, and service chains are highly optimized, internationally interconnected, and particularly susceptible to disruptions from extreme weather events, geopolitical conflicts, production failures, transport issues, or societal crises.
PROVIDER aims to transition from reactive post-hoc analyses to a continuously operating, predictive system. For this purpose, heterogeneous data sources, current news flows, public data, economic information, and knowledge graphs are combined. On this basis, dynamic simulations are created to estimate potential bottlenecks and cascade effects over a period of several months. The project combines Knowledge Graphs, Large Language Models, agent-based simulation, Deep Reinforcement Learning, and Adversarial Resilience Learning.
The sub-project of Leuphana University Lüneburg focuses on neuro-symbolic AI for explainable information extraction and event analysis. The goal is to develop methods to detect, semantically classify, and make relevant events from structured and unstructured data sources (especially news flows) usable for the parameterization and evaluation of simulations. A special focus is on the use and adaptation of Large Language Models, the linkage with knowledge graphs, and the development of LLM-based agents that can generically access external interfaces.
Furthermore, Leuphana is developing an interactive evaluation tool with a natural language interface within the project. Users without computer science expertise should be able to query simulation results, visualize values and trends, and trace the origin and reasoning of the results. Thus, the sub-project contributes to making complex simulation results understandable, explainable, and usable for decision-making processes.
Objectives
- Develop an AI-supported early warning system for supply bottlenecks.
- Combine heterogeneous data sources including knowledge graphs and large language models.
- Create dynamic simulations to estimate cascade effects.
- Develop an interactive evaluation tool with a natural language interface.