Our research
At the University of Freiburg, we are building the tools to estimate the uncertainty of neural networks (NNs), including natural language models (LLMs), and to measure their resource usage especially their carbon footprint. We also test methods to improve the precision of the different models, using Retrieval Augmented Generation (RAG) models.
At the national laboratory: Deutsches Elektronen-Synchrotron (DESY) in Hamburg and Berlin-Zeuthen, we study more native data formats for machine learning (ML) applications, and investigate reduction methods of the size and complexity of ML applications, while keeping their precision the similar.