Dr. Teodor Chiaburu

Verifying the Linear Representation Hypothesis: How Interpretable Are Vision SAEs?

Teodor Chiaburu, Franz Motzkus, Frank Haußer, Felix Bießmann

Under Review

We adapt the Autointerpretability Score (AIS) - previously shown to align with human judgments in NLP - to vision tasks and validate our approach in a dedicated user study. We evaluate SAE concept quality using both standard metrics and our adapted AIS. We find that established interpretability metrics for SAEs correlate neither with one another nor with AIS, indicating that no single reference-free metric, whether grounded in the LRH or not, is sufficient for verifying the interpretability of vision SAEs. We argue these findings support recent calls for more verifiable, ground-truth-anchored design and evaluation of explanation methods. Download preprint here.

Uncertainty in Explainable Artificial Intelligence (PhD Thesis)

Teodor Chiaburu

Defended with grade "very good" on the 10th of March at the Einstein Center for Digital Future, Berlin

Download thesis here.

No Single Metric Tells the Whole Story: A Multi-Dimensional Evaluation Framework for Uncertainty Attributions

Emily Schiller, Teodor Chiaburu, Marco Zullich, Luca Longo

Presented at WCXAI 2026, Fortaleza

We propose concrete implementations for the correctness, consistency, continuity, and compactness properties to evaluate uncertinty attributions. Additionally, we introduce conveyance, a property tailored to uncertainty attributions that evaluates whether controlled increases in epistemic uncertainty reliably propagate to feature-level attributions. We demonstrate our evaluation framework with eight metrics across combinations of uncertainty quantification and feature attribution methods on tabular and image data. Although most metrics rank the methods consistently across samples, inter-method agreement remains low. This suggests no single metric sufficiently evaluates uncertainty attribution quality. Download paper here.

Uncertainty-Guided Expert-AI Collaboration for Efficient Soil Horizon Annotation

Teodor Chiaburu, Vipin Singh, Felix Bießmann, Frank Haußer

Presented at ECAI 2025, CLEAR-AI Workshop, Bologna

In this work, we apply conformal prediction to SoilNet, a multimodal multitask model for describing soil profiles. We design a simulated human-in-the-loop (HIL) annotation pipeline, where a limited budget for obtaining ground truth annotations from domain experts is available when model uncertainty is high. Our experiments show that conformalizing SoilNet leads to more efficient annotation in regression tasks and comparable performance scores in classification tasks under the same annotation budget when tested against its non-conformal counterpart. Download paper here.

SoilNet: A Multimodal Multitask Model for Hierarchical Classification of Soil Horizons

Teodor Chiaburu, Vipin Singh, Felix Bießmann, Frank Haußer

Published in Geoderma, 2026

In this work, we propose SoilNet - a multimodal multitask model to tackle this problem through a structured modularized pipeline. Our approach integrates image data and geotemporal metadata to first predict depth markers, segmenting the soil profile into horizon candidates. Each segment is characterized by a set of horizon-specific morphological features. Finally, horizon labels are predicted based on the multimodal concatenated feature vector, leveraging a graph-based label representation to account for the complex hierarchical relationships among soil horizons. Our method is designed to address complex hierarchical classification, where the number of possible labels is very large, imbalanced and non-trivially structured. We demonstrate the effectiveness of our approach on a real-world soil profile dataset. All the code and experiments can be found in our repository. Download paper here.

Uncertainty Propagation in XAI: A Comparison of Analytical and Empirical Estimators

Teodor Chiaburu, Felix Bießmann, Frank Haußer

Presented at WCXAI 2025, Istanbul

This paper introduces a framework to measure uncertainty in XAI explanations, which stems from input data and model parameter variations. We use analytical and empirical methods to estimate how this uncertainty affects explanations and evaluate their robustness across different datasets. Our study identifies XAI methods that don't reliably handle uncertainty, emphasizing the need for uncertainty-aware explanations in critical applications and revealing limitations in current XAI techniques. All the code and experiments can be found in our repository. Download paper here.

Towards a Global Spatial Machine Learning Model for Seasonal Groundwater Level Predictions in Germany

Stefan Kunz, Alexander Schulz, Maria Wetzel, Maximilian Nölscher, Teodor Chiaburu, Felix Biessmann, Stefan Broda

Published in HESS - Hydrology and Earth System Sciences, 2025

Accurate groundwater level predictions are essential for a sustainable groundwater management. This study applies two machine learning (ML) models—N-HiTS and TFT—to seasonally predict groundwater levels for 5,288 monitoring wells across Germany. Both approaches provided good predictions across diverse hydrogeological conditions, whereby N-HiTS outperformed the TFT. Both models showed better perforance in areas with high data density, in lowlands, and when distinct seasonal dynamics occurred. Download paper here.

Evaluating Time Series Models for Urban Wastewater Management: Predictive Performance, Model Complexity and Resilience

Vipin Singh, Tianheng Ling, Teodor Chiaburu, Felix Bießmann

Presented at SpliTech 2025, Split

We propose a protocol for evaluating neural network architectures for combined sewer systems time series forecasting with respect to predictive performance, model complexity and robustness to perturbations. In addition, we assess model performance on peak events and critical fluctuations, as these are the key regimes for urban wastewater management. To investigate the feasibility of lightweight models suitable for IoT deployment, we compare global models, which have access to all information, to local models, which rely solely on nearby sensor readings. Download paper here.

Data-driven Modeling of Combined Sewer Systems for Urban Sustainability: An Empirical Evaluation

Vipin Singh, Tianheng Ling, Teodor Chiaburu, Felix Bießmann

Presented at the German Conference of Artificial Intelligence 2024, Würzburg

We present a comprehensive empirical evaluation of several state-of-the-art time series models for predicting sewer system dynamics in a large urban infrastructure, utilizing three years of measurement data. We especially investigate the potential of Deep Learning models to maintain predictive precision during network outages by comparing global models, which have access to all variables within the sewer system, and local models, which are limited to data from a restricted set of local sensors. Download paper here.

Confident Teacher, Confident Student? A Novel User Study Design for Investigating the Didactic Potential of Explanations and their Impact on Uncertainty

Teodor Chiaburu, Frank Haußer, Felix Bießmann

Presented at ECML 2024, AIMLAI Workshop, Vilnius

We propose an experimental design to evaluate the potential of XAI in human-AI collaborative settings as well as the potential of XAI for didactics. In a user study with 1200 participants we investigate the impact of explanations on human performance on a challenging visual task - annotation of biological species in complex taxonomies. Download paper here.

Uncertainty in XAI: Human Perception and Modeling Approaches

Teodor Chiaburu, Frank Haußer, Felix Bießmann

Published in Machine Learning and Knowledge Extraction (MAKE), 2024

We review established and recent methods to account for uncertainty in ML models and XAI approaches and we discuss empirical evidence on how model uncertainty is perceived by human users of XAI systems. We summarize the methodological advancements and limitations of methods and human perception. Finally, we discuss the implications of the current state of the art in model development and research on human perception. We believe highlighting the role of uncertainty in XAI will be helpful to both practitioners and researchers and could ultimately support more responsible use of AI in practical applications. Download paper here.

CoProNN: Concept-based Prototypical Nearest Neighbors for Explaining Vision Models

Teodor Chiaburu, Frank Haußer, Felix Bießmann

Presented at WCXAI 2024, Valletta

We present a novel approach that enables domain experts to quickly create concept-based explanations for computer vision tasks intuitively via natural language. Leveraging recent progress in deep generative methods we propose to generate visual concept-based prototypes via text-to-image methods. These prototypes are then used to explain predictions of computer vision models via a simple k-Nearest-Neighbors routine. The approach can be evaluated offline against the ground-truth of predefined prototypes that can be easily communicated also to domain experts as they are based on visual concepts. All the code and experiments can be found in our repository. Download paper here.

Interpretable Time Series Models for Wastewater Modeling in Combined Sewer Overflows

Teodor Chiaburu, Felix Bießmann

Presented at ISCSI 2023, Lisbon

In this work we specifically address the problem of sewage water polluting surface water bodies after spilling over from rain tanks as a consequence of heavy rain events. We investigate to what extent state-of-the-art interpretable time series models can help predict such critical water level points, so that the excess can promptly be redistributed across the sewage network. Our results indicate that modern time series models can contribute to better waste water management and prevention of environmental pollution from sewer systems. All the code and experiments can be found in our repository. Download paper here.

Low Cost Machine Vision for Insect Classification

Ingeborg Beckers, Danja Brandt, Teodor Chiaburu, Frank Haußer, Henning Schmidt, Ilona Schrimpf, Alexandra Stadel, Martin Tschaikner

Presented at IntelliSys 2023, Amsterdam

In this work, we present an imaging method as part of a multisensor system developed as a low-cost, scalable, open-source system that is adaptable to classical trap types. The image quality meets the requirements needed for classification in the taxonomic tree. Illumination and resolution have been optimized and motion artefacts suppressed. The system is evaluated exemplarily on a dataset consisting of 16 insect species of the same as well as different genus, family and order. We demonstrate that standard CNN-architectures like ResNet50 (pretrained on iNaturalist data) or MobileNet perform very well for the prediction task after re-training. Smaller custom-made CNNs also lead to promising results. Download paper here.

Multisensor Data Fusion for Automatized Insect Monitoring

Martin Tschaikner, Danja Brandt, Henning Schmidt, Felix Bießmann, Teodor Chiaburu, Ilona Schrimpf, Thomas Schrimpf, Alexandra Stadel, Frank Haußer, Ingeborg Beckers

Presented at SPIE Remote Sensing 2023, Amsterdam

This paper presents a multisensor approach that uses AI-based data fusion for insect classification. The system is designed as low-cost setup and consists of a camera module and an optical wingbeat sensor as well as environmental sensors to measure temperature, irradiance or daytime as prior information. Download paper here.

Towards ML Methods for Biodiversity: A Novel Wild Bee Dataset and Evaluations of XAI Methods for ML-Assisted Rare Species Annotations

Teodor Chiaburu, Felix Bießmann, Frank Haußer

Presented at CVPR 2022, XAI Workshop, New Orleans

Classification and XAI Experiments on a dataset of wildbee photos. Visit repository (link to paper, poster and dataset there).

VisPer - Visual Forensic Person Verification

Visit project website.

BGR - Soil Horizon Prediction

Localizing and classifying soil horizons in multimodal input (visual and tabular). Visit BGR website.

KInsecta and beexplainable

Investigating interpretability in ML neural networks trained to recognize wildbee species on images and wingbeat signals. Visit KInsecta website.

RIWWER

Reduction of the Impact of untreated Waste Water on the Environment in case of torrential Rain. Visit project website.

KI Festival

Heilbronn, 2026

Hat die KI Ihr Vertrauen verdient?

BHT Magazin, 2026

Read article here (in German).

Uncertainty in XAI and ML for Environmental Sciences

Forschungs- und Transfertag 2025, BHT

Towards Reliable Soil Horizon Classification: A Multimodal Approach Leveraging Hierarchical Taxonomy and Graph-Based Representations

KIDA-KON 2025, Braunschweig

Uncertainty in XAI

Summer School on Signal Processing 2024, Capri

Die KI bewirbt sich als Lehrkraft

Lange Nacht der Wissenschaften 2024, BHT

Explanations from Humans for Humans: The Potential of Transparent AI for Trust Calibration and Didactics

Einstein Center Digital Future 2023, Berlin

Das Summen der Pixel: Versteht die KI, was eine Biene ist?

re:publica 2022, Berlin

Vertrau mir, wenn du kannst: Ein Blick in die Black Box der KI

Lange Nacht der Wissenschaften 2022, BHT