Selected work

Representative Papers

Monocular Visual Simultaneous Localization and Mapping: (R)Evolution From Geometry to Deep Learning-Based Pipelines

May 01, 2024IEEE Transactions on Artificial Intelligence

This paper addresses the robustness bottlenecks of monocular visual SLAM in realistic scenarios—namely dynamic scenes, underwater imaging, and high-speed motion. To this end, it proposes the first comprehensive survey framework that jointly ensures classification consistency and quantifiable evaluation. The work systematically traces the evolution of geometric and end-to-end learning paradigms, unifies the modeling of these three canonical challenges, and establishes a reproducible, quantitative evaluation benchmark. It rigorously delineates the boundaries between the two paradigms, integrating multi-view geometry, nonlinear optimization, CNNs/RNNs, domain adaptation, and robust feature learning to enable cross-environment performance analysis. Empirical findings reveal distinct failure modes of existing methods under varying imaging conditions, thereby providing a systematic benchmark and an extensible research roadmap for designing and deploying robust monocular SLAM systems.

7 citationsRead paper

Benchmarking Time Series Foundation Models for Short-Term Household Electricity Load Forecasting

Oct 12, 2024arXiv.org

Short-term household electricity load forecasting demands practical zero-shot generalization without task-specific training. Method: This study presents the first systematic evaluation of time-series foundation models—Chronos, TimesFM, and LagLlama—in this setting, benchmarking their zero-shot performance against a from-scratch trained Transformer under identical experimental conditions and evaluating rigorously using multi-scale error metrics (MAE/MSE). Contribution/Results: All foundation models achieve performance on par with task-specific models, with TimesFM attaining the best zero-shot accuracy under large input windows—reducing MAE by 12.3% relative to the strongest customized model. Critically, none require gradient updates or domain-specific fine-tuning; all operate effectively with only minimal historical data. The results demonstrate that time-series foundation models are both effective and deployment-efficient for fine-grained, low-resource load forecasting scenarios.

6 citations2 influentialRead paper

Information Leakage Detection through Approximate Bayes-optimal Prediction

Jan 25, 2024arXiv.org

Information leakage (IL) detection faces challenges including difficulty in estimating high-dimensional mutual information (MI), poor convergence, and limitations of conventional methods to binary sensitive attributes. This paper proposes the first general-purpose IL detection framework for arbitrary sensitive information, deeply integrating statistical learning theory with information theory. Instead of explicit MI estimation—prone to bias and instability—we employ the log-loss and classification accuracy of a Bayes-optimal predictor as principled surrogates for MI. Our method synergistically combines AutoML, Bayesian modeling, log-loss optimization, and information-theoretic quantification to achieve automated, robust MI approximation. Evaluated on synthetic benchmarks and real-world OpenSSL TLS datasets, our approach reduces MI estimation error by 37% and achieves an IL detection AUC of 0.92—substantially outperforming state-of-the-art baselines.

1 citationsRead paper

H-VAEP and H-xT: Valuing Offensive On-the-Ball Actions in Handball by Estimating Probabilities

Aug 13, 2026

This study addresses the limitations of traditional handball player evaluation, which relies on simplistic statistical metrics and fails to quantify individual offensive contributions within multi-player coordination. For the first time, the paper systematically adapts the football-derived expected Threat (xT) and Valuing Actions by Estimating Probabilities (VAEP) frameworks to handball, leveraging five seasons of event-tracking data from the German Handball Bundesliga to build an offensive action valuation model. It introduces a handball-specific pitch zoning scheme, H-xT, to enhance robustness, and refines the feature space and context window of H-VAEP to mitigate team identity leakage and better capture playmaking value. The resulting player ratings exhibit high stability, discriminative power, and intuitive interpretability, accurately identifying key organizers. The full implementation is publicly released to facilitate adoption by professional clubs.

0 citationsRead paper

Shapes from Examples: Foundations of Shape Learning in Recursive SHACL

Jul 30, 2026

This study addresses the problem of automatically learning shape expressions that conform to recursive SHACL semantics from positive and negative example nodes, thereby enabling effective validation of knowledge graphs. Focusing on the ELI fragment of description logics, the work establishes—for the first time—the theoretical foundations of shape learning under multiple semantics of recursive SHACL, including well-founded, stable, and supported interpretations, and precisely delineates the boundaries of learnability. The core contributions include proving that both the existence of a fitting shape expression and the computation of a most specific fitting expression are solvable in exponential time, while also providing polynomial-time algorithms for several practically relevant special cases. Furthermore, the paper establishes tight upper bounds on the associated computational complexity.

0 citationsRead paper

Recent publications

Latest Papers

H-VAEP and H-xT: Valuing Offensive On-the-Ball Actions in Handball by Estimating Probabilities

Aug 13, 2026

This study addresses the limitations of traditional handball player evaluation, which relies on simplistic statistical metrics and fails to quantify individual offensive contributions within multi-player coordination. For the first time, the paper systematically adapts the football-derived expected Threat (xT) and Valuing Actions by Estimating Probabilities (VAEP) frameworks to handball, leveraging five seasons of event-tracking data from the German Handball Bundesliga to build an offensive action valuation model. It introduces a handball-specific pitch zoning scheme, H-xT, to enhance robustness, and refines the feature space and context window of H-VAEP to mitigate team identity leakage and better capture playmaking value. The resulting player ratings exhibit high stability, discriminative power, and intuitive interpretability, accurately identifying key organizers. The full implementation is publicly released to facilitate adoption by professional clubs.

0 citationsRead paper

Shapes from Examples: Foundations of Shape Learning in Recursive SHACL

Jul 30, 2026

This study addresses the problem of automatically learning shape expressions that conform to recursive SHACL semantics from positive and negative example nodes, thereby enabling effective validation of knowledge graphs. Focusing on the ELI fragment of description logics, the work establishes—for the first time—the theoretical foundations of shape learning under multiple semantics of recursive SHACL, including well-founded, stable, and supported interpretations, and precisely delineates the boundaries of learnability. The core contributions include proving that both the existence of a fitting shape expression and the computation of a most specific fitting expression are solvable in exponential time, while also providing polynomial-time algorithms for several practically relevant special cases. Furthermore, the paper establishes tight upper bounds on the associated computational complexity.

0 citationsRead paper

Nudging Sustainable Choices through LLM-Generated Recommendation Explanations

Jul 28, 2026

This study addresses the limited efficacy of conventional sustainability disclosures in driving pro-environmental behavior. Bridging behavioral nudging theory and large language models (LLMs), the authors propose an innovative approach that automatically generates framed explanatory messages to promote sustainable choices. Randomized controlled trials were conducted in two distinct contexts—instant coffee purchases (low-involvement) and hotel bookings (high-involvement). Findings reveal that merely providing sustainability information does not significantly influence user decisions. In contrast, LLM-generated explanations incorporating descriptive social norms or loss/gain framing significantly increase the adoption of sustainable options while reducing perceived decision burden. This work represents the first integration of LLMs with behavioral nudges, uncovering a critical mechanism for translating informational cues into actionable behavior.

0 citationsRead paper

LLM-SoccerArena: Benchmarking LLMs on Real-World Predictions in Sports

Jul 27, 2026

Current evaluations of large language models (LLMs) predominantly rely on static, retrospective benchmarks, which fail to capture their predictive capabilities under real-world uncertainty. To address this gap, this work introduces the first prospective, real-time evaluation protocol and open-source platform tailored for unresolved events. The framework employs a factorial experimental design to systematically assess LLM performance across varying information access modalities, prompting strategies, and prediction horizons, while integrating timestamp logging, pattern validation, tool-use tracing, and cost tracking. Experiments on 104 matches and 15 tournament-related questions from the 2026 FIFA World Cup reveal that models with web access only marginally outperform those without (a Brier score improvement of 0.023), highlighting the current limitations of LLMs in forecasting real-world outcomes.

0 citationsRead paper

Semantic-Aware Task Clustering for Constructive and Cooperative Multi-Tasking

Jul 23, 2026

In multi-task semantic communication, improper modeling of inter-task semantic relationships often leads to negative transfer and destructive collaboration, degrading overall performance. To address this issue, this work introduces semantic relationship modeling into multi-task clustering for the first time and proposes a semantic-aware two-stage optimization framework. The approach first groups tasks via hierarchical density-based clustering to achieve semantic alignment, followed by end-to-end joint training within each cluster. This strategy effectively fosters constructive collaboration among related tasks, significantly improving accuracy and substantially mitigating negative transfer compared to both unclustered multi-task learning and independently trained baselines.

0 citationsRead paper