10.18453/rosdok_id00005643 Hellwig, Jan Jan Hellwig http://d-nb.info/gnd/1402678932 https://orcid.org/0009-0001-2455-8135 Universität Rostock 2025 510 Mathematics 2025 en https://purl.uni-rostock.de/rosdok/id00005643 urn:nbn:de:gbv:28-rosdok_id00005643-5 In this work, we first discuss the question of whether the problem of modeling NMR time series has ambiguous or unique solutions in the continuous and discrete case and using several common types of model functions. In the second part, we propose an efficient and automated solution, by interpolating the parameters using cubic spline functions. Additionally, we improve on the optimization approach by introducing convolutional neural networks, which better initialize the optimization routines. We demonstrate the functionality of our algorithms on constructed and experimental data sets.