@inbook{7227, author = {Weisser, Christoph}, booktitle = {Künstliche Intelligenz und Data Science in Theorie und Praxis: Von klassischen Algorithmen zu Generativer KI und Agentic AI Konzepte und Realisierung in Unternehmen}, editor = {Gillhuber, Andreas and Kauermann, Göran and Hauner, Wolfgang}, publisher = {Springer}, title = {{Von der KI-Plattform zur dialogfähigen Intelligenz: Die Zukunft industrieller Prognosen}}, doi = {https://link.springer.com/book/10.1007/978-3-662-73474-2}, year = {2026}, } @techreport{7217, abstract = {Typed decision models (TDMs) return a probability for each allowed answer instead of generating text, which makes them candidates for cheaply replacing LLM judges that compare two answers. Whether they share the biases of LLM judges is untested. We run the CALM benchmark's bias tests on eight TDMs and seven LLM judges of 2024 and 2026. Every TDM is biased, in three to eight of twelve tests. Emotional tone affects them most, and rewriting the better answer in such a tone lowers the accuracy of a typical TDM from 99% to 42%. TDMs are more biased than 2026 LLM judges. All eight change verdicts when the answers swap places, which the 2026 LLM judges hardly do, and several side with an answer that most people allegedly prefer or that a famous model allegedly wrote. Yet both kinds of judge are tripped up by the same prompts. The TDMs' probabilities show that most tests mainly flip verdicts the TDM was unsure of, and that good calibration does not protect against bias. As TDMs are more consistent and 37 times cheaper than 2026 LLM judges, our results suggest they can be attractive judges if their biases are tested for the task.}, author = {Lukassen, Fabian and Weisser, Christoph and Silbersdorff, Alexander}, title = {{Cheap, Consistent, Calibrated, but Biased? Typed Decision Models as Judges}}, year = {2026}, } @techreport{7152, abstract = {Property prediction for molecules and materials is bottlenecked by label scarcity, and in this regime the operative failure mode is not a low mean but an unpredictable tail: individual training runs collapse. The three dominant molecular representations fail in complementary ways?physicochemical descriptors are exact but fixed, graph neural networks learn topology but degenerate under sparse supervision, and language models supply semantic context but cannot compute exact quantities? yet combining them is itself the learning problem, because modality-level scalar gating commits the whole model to one trust weight per source. We introduce TAME, a tri-modal encoder whose element-wise Mixture-of-Experts router assigns an independent modality mixture to every hidden coordinate, stabilized by a balance?entropy regularizer, a closed-form gate initialization (? = log τ ) that is intended to remove router burn-in, and two-stage self-supervised graph pretraining. On scaffold-split BACE (≈ 1.5k molecules, 100 seeds) we evaluate two fusion topologies, each against its own graph-only control. In both, adding the text and descriptor experts contracts the seed-to-seed distribution of the threshold-free metrics?ROC-AUC σ falls five-fold in the flat topology and 1.6-fold in the hierarchical one, whose control was already the tighter of the two?and the collapse tail reaching ROC-AUC ≈ 0.45 that the weaker single-modality configurations carry is absent from the fused models while accuracy is retained. Each comparison is matched on everything but the fusion stage, isolating its effect. The router allocates a distinct mixture to each coordinate, which no scalar gate can represent, and the entropy weight moves that allocation continuously between graded mixing and a near-binary regime in which each coordinate is claimed outright by one modality. Fused-representation alignment tracks encoder quality without supervision, shifting from text to topology once pretraining lifts the graph embedding out of rank-1 collapse. The same three experts map onto crystal graphs, composition?structure descriptors and literature-mined text. These three modalities recur unchanged across molecular and inorganic chemistry, so TAME is a single interpretable recipe for turning volatile low-data predictors into reliable, self-explaining ones, wherever data are scarce and every failed prediction costs a real experiment.}, author = {Schiller, Robert P. and Weisser, Christoph and Müller, Klaus-Robert and Ochsenfeld, Christian and Semnani, Parastoo}, title = {{TAME: Element-wise Mixture-of-Experts Fusion for Reliable and Interpretable Molecular Property Prediction}}, doi = {10.26434/chemrxiv.15008270/v2}, year = {2026}, } @article{6958, author = {Kumar, Manish and Thielmann, Anton Frederik and Weisser, Christoph and Säfken, Benjamin}, journal = {Preprint}, publisher = {Arxiv}, title = {{From Uniform to Learned Knots: A Study of Spline-Based Numerical Encodings for Tabular Deep Learning}}, year = {2026}, } @article{7112, author = {Kumar, Manish and Thielmann, Anton Frederik and Weisser, Christoph and Säfken, Benjamin}, journal = {Transactions on Machine Learning Research}, publisher = {arxiv}, title = {{From Uniform to Learned Knots: A Study of Spline-Based Numerical Encodings for Tabular Deep Learning}}, doi = {10.48550/arXiv.2604.05635}, year = {2026}, } @article{7110, author = {Schlee, Michael and Lukassen, Fabian and Weisser, Christoph}, journal = {arXiv:2608.11753}, title = {{LabelFusion-TS: Fusing Large Language Models, Transformer Encoders, and Financial Time Series for Monetary-Policy Stance Classification}}, doi = {10.48550/arXiv.2608.11753 }, year = {2026}, } @article{7066, author = {Lukassen, Fabian and Weisser, Christoph and Kneib, Thomas and Silbersdorff, Alexander}, journal = {Preprint}, publisher = {arxiv}, title = {{CAFE: A Compound-AI Factorial Evaluation Framework}}, doi = {10.48550/arXiv.2607.10380}, year = {2026}, } @article{6924, abstract = {Das Positionspapier formuliert neun Thesen zur Weiterentwicklung der Statistikausbildung im Kontext von Data Science und Künstlicher Intelligenz. Im Zentrum stehen Data & Statistical Literacy, Datenqualität und -ethik, die Verbindung von Statistik und Machine Learning sowie die Stärkung einer eigenständigen Statistikdidaktik. Die Diskussionsbeiträge erweitern diese Perspektiven: Christina Elmer betont Wissenschaftskommunikation und AI Literacy, Helmut Küchenhoff projektbasiertes Lernen und Datenschutz, Christoph Weisser die industrielle Praxis und organisationsweite Datenkompetenz, Göran Kauermann die Verzahnung von Statistik und Data Science, und Rolf Biehler sowie Karin Binder die didaktische und institutionelle Weiterentwicklung. Die Replik greift diese Impulse auf und unterstreicht die interdisziplinäre Verantwortung für eine zukunftsfähige Statistikausbildung.}, author = {Berger, Ursula and Biehler, Rolf and Binder, Karin and Elmer, Christina and Ertz, Florian and Hotz, Thomas and Huber, Sarah and Ickstadt, Katja and Kauermann, Göran and Küchenhoff, Helmut and Lübke, Karsten and Münnich, Ralf and Schüller, Katharina and Skill, Thomas and Weihs, Claus and Weinert, Henrike and Weisser, Christoph}, issn = {1863-8163}, journal = {AStA Wirtschafts- und Sozialstatistisches Archiv}, publisher = {Springer}, title = {{Daten, Künstliche Intelligenz und Evidenz - neue Anforderungen an die Statistikausbildung an Hochschulen: Diskussion und Erwiderung}}, doi = {10.1007/s11943-026-00372-0}, year = {2026}, } @article{6957, author = {Amorin, Alvaro Paredes and Python, Andre and Weisser, Christoph}, journal = {LREC 2026 - The Fifteenth biennial Language Resources and Evaluation Conference - Financial Narrative Processing (FNP) Workshop}, title = {{Not All News Is Equal: Topic- and Event-Conditional Sentiment From Finetuned LLMs for Aluminum Price Forecasting}}, doi = {10.48550/arXiv.2603.09085}, year = {2026}, } @article{6959, author = {Schlee, Michael and Kivimaki, Timo and Mashiku, Melchizedek and Weisser, Christoph and Säfken, Benjamin}, journal = {LREC 2026 - The Fifteenth biennial Language Resources and Evaluation Conference - Financial Narrative Processing (FNP) Workshop}, title = {{LabelFusion: Fusing Large Language Models With Transformer Encoders for Robust Financial News Classification}}, doi = {10.48550/arXiv.2512.10793}, year = {2026}, } @article{6961, author = {Lukassen, Fabian and Herrmann, Jan and Weisser, Christoph and Säfken, Benjamin and Kneib, Thomas}, journal = {Preprint}, publisher = {Arxiv}, title = {{From XAI to Stories: A Factorial Study of LLM-Generated Explanation Quality}}, doi = {10.48550/arXiv.2601.02224}, year = {2026}, } @article{6960, author = {Lukassen, Fabian and Weisser, Christoph and Schlee, Michael and Kumar, Manish and Thielmann, Anton and Säfken, Benjamin and Kneib, Thomas and Silbersdorff, Alexander}, journal = {Preprint}, publisher = {Arxiv}, title = {{LLM-Augmented Change Point Detection: A Methodological Framework for Ensemble Detection and Automated Explanation}}, doi = {10.48550/arXiv.2601.02957}, year = {2026}, } @article{6962, author = {Amorin, Alvaro Paredes and Python, Andre and Weisser, Christoph}, journal = {Preprint}, publisher = {Arxiv}, title = {{Fine-Tuning of Lightweight Large Language Models for Sentiment Classification on Heterogeneous Financial Textual Data}}, doi = {10.48550/arXiv.2512.00946}, year = {2025}, } @inproceedings{6966, author = {Semnani, Parastoo and Bogojeski, Mihail and Bley, Florian and Zhang, Zizheng and Wu, Qiong and Kneib, Thomas and Herrmann, Jan and Weisser, Christoph and Müller, Klaus Robert}, location = {Copenhagen, Denmark}, publisher = {Openreview}, title = {{From Prediction to Proposal in Catalysis: Robust Evaluation, LRP Explanations, and Relevance-Guided Candidate Generation}}, year = {2025}, } @article{6920, author = {Reuter, Arik and Thielmann, Anton and Weisser, Christoph and Säfken, Benjamin and Kneib, Thomas}, issn = {2162-2388 }, journal = {IEEE Transactions on Neural Networks and Learning Systems}, number = {8}, pages = {14551 -- 14565}, publisher = {IEEE}, title = {{Probabilistic Topic Modelling With Transformer Representations}}, doi = {10.1109/TNNLS.2025.3538262}, volume = {36}, year = {2025}, } @article{6919, author = {Kant, Gillian and Alkaher, Talip and Kivimaki, Timo and Weisser, Christoph}, issn = {1556-1836}, journal = {Terrorism and Political Violence}, publisher = {Taylor & Francis}, title = {{The Word and the Bullet: Out-Grouping and Threat Framing as Predictors of Islamic State Targeting, 2015-2020}}, doi = {10.1080/09546553.2025.2561194}, year = {2025}, } @inproceedings{6942, author = {Thielmann, Anton and Weisser, Christoph and Säfken, Benjamin}, booktitle = {Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING)}, editor = {Calzolari, Nicoletta and Kan, Min-Yen and Hoste, Veronique and Lenci, Alessandro and Sakti, Sakriani and Xue , Nianwen }, location = {Torino, Italia}, pages = {8395--8405}, publisher = {Elra And Iccl}, title = {{Human in the Loop: How to Effectively Create Coherent Topics by Manually Labelling Only a Few Documents per Class}}, year = {2024}, } @article{6963, author = {Thielmann, Anton Frederik and Kumar, Manish and Weisser, Christoph and Reuter, Arik and Säfken, Benjamin and Samiee, Soheila}, journal = {Preprint}, publisher = {Arxiv}, title = {{Mambular: A Sequential Model for Tabular Deep Learning}}, doi = {10.48550/arXiv.2408.06291}, year = {2024}, } @article{6964, author = {Reuter, Arik and Thielmann, Anton and Weisser, Christoph and Fischer, Sebastian and Säfken, Benjamin and Khadka, Bishnu}, journal = {Preprint}, publisher = {Arxiv}, title = {{GPTopic: Dynamic and Interactive Topic Representations}}, doi = {10.48550/arXiv.2403.03628}, year = {2024}, } @article{6965, author = {Stemmler, Henry and Lenel, Friederike and Kis-Katos, Krisztina and Weisser, Christoph}, journal = {Preprint}, publisher = {Ssrn}, title = {{Solar Technology as a Shock-Coping Device: Evidence From Rural Tanzania}}, doi = {10.2139/ssrn.5046518}, year = {2024}, }