Oliver Hennhöfer

Oliver Hennhöfer

Research staff at Hochschule Karlsruhe working on reliable machine learning systems.

Intelligent Systems Research Group/Karlsruhe, Germany

I work on statistical methods and software for machine learning systems that need to stay reliable under uncertainty. My current focus is conformal inference for anomaly detection, uncertainty quantification, and online false discovery rate control.

I am especially interested in settings where data are scarce, noisy, non-stationary, or arrive sequentially. Recent work studies low-data conformal anomaly detection, shift-aware weighted procedures, and practical Python interfaces that turn anomaly scores into calibrated p-values and FDR-controlled decisions.

In parallel to the theory, I build research software for reproducible experiments and applied workflows. The current toolchain includes nonconform for conformal anomaly detection and online-fdr for sequential multiple testing when hypotheses arrive over time.

Interests

Anomaly Detection/Conformal Inference/Uncertainty Quantification/False Discovery Rate Control/Sequential Testing/Research Software

Projects

  • nonconform

    Conformal anomaly detection and change-point detection for turning anomaly scores into calibrated p-values, with split, cross, and weighted conformal workflows plus FDR-controlled decisions.

    Python, conformal inference, anomaly detection, FDR control

    GitHubDocsPyPI

  • online-fdr

    Sequential multiple-testing procedures for online FDR and FWER control, including one-by-one APIs for LORD, SAFFRON, ADDIS, alpha-spending, and batch baselines.

    Python, online FDR, FWER, sequential testing

    GitHubDocs

  • oddball

    A lightweight collection of anomaly detection datasets for reproducible experiments.

    Python, datasets, benchmarks

    GitHub

  • aberrant

    Online anomaly detection algorithms for streaming settings.

    Python, streaming, anomaly detection

    GitHub

Selected Publications

  • Conformal Anomaly Detection in Python: Moving Beyond Heuristic Thresholds with 'nonconform'

    Oliver Hennhöfer, Maximilian Kirsch, Christine Preisach

    Preprint, arXiv:2605.13642

    2026

    arXiv

  • Between Resolution Collapse and Variance Inflation: Weighted Conformal Anomaly Detection in Low-Data Regimes

    Oliver Hennhöfer, Christine Preisach

    Preprint, arXiv:2603.23205

    2026

    arXiv

  • Leave-One-Out-, Bootstrap- and Cross-Conformal Anomaly Detectors

    Oliver Hennhöfer, Christine Preisach

    IEEE ICKG 2024, pp. 110-119

    2024

    DOIarXiv