Bernd Freisleben

As a long-term scientific project on the history of the Italian language, the Lessico Etimologico Italiano (LEI) represents one of the most important and ambitious historical and etymological dictionary projects ever undertaken. The LEI, which started in 1968, documents and analyzes every single word of the Italian language and all Italian dialects from their beginnings to today. Until 2018, the editorial process used the traditional lexicographical method of creating annotated index cards to collect information. Each index card contains text areas, handwritten annotations, and/or textual stamps. Of particular interest are the etymon, context (the source text), and textual stamp, which contains the abbreviated title of the book or source from which it was copied or taken. In this paper, we present a novel approach for efficiently processing a large number of scanned index cards to accelerate the philological work required for producing the LEI. For this purpose, a deep learning workflow for automatic detection, alignment, and recognition of textual stamps on digitized index cards is proposed. To support large-scale indexing under an initially incomplete stamp inventory, we introduce an embedding-based retrieval workflow that enables the identification and integration of previously unseen stamps during operation. Our experimental evaluations show excellent results for stamp detection and stamp recognition, with a mean average precision of 98.80% and an accuracy of 97.02%, respectively. In addition, we compare our stamp recognition approach with two recent Large Language Model (LLM) approaches for OCR. Our best approach achieves 97.02% accuracy, which surpasses the OCR performance of the two LLMs (91.60% and 89.47%, respectively). Furthermore, we compare our approach to a recent multimodal large language model (MM-LLM) on a benchmark dataset and observe substantially lower accuracy for the LLM (up to 84.58%) compared to our approach (98.61%).

Background/Objectives: Accurate identification of dental malocclusions from routine clinical photographs can be time-consuming and subject to interobserver variability. A YOLOv11-based deep learning approach is presented and evaluated for automatic malocclusion detection on routine intraoral photographs, testing the hypothesis that training on a structured annotation protocol enables reliable detection of multiple clinically relevant malocclusions. Methods: An anonymized dataset of 5854 intraoral photographs (frontal occlusion; right/left buccal; maxillary/mandibular occlusal) was labeled according to standardized instructions derived from the Index of Orthodontic Treatment Need (IOTN) A total of 17 clinically relevant classes were annotated with bounding boxes. Due to an insufficient number of examples, two malocclusions (transposition and non-occlusion) were excluded from our quantitative analysis. A YOLOv11 model was trained with augmented data and evaluated on a held-out test set using mean average precision at IoU 0.5 (mAP50), macro precision (macro-P), and macro recall (macro-R). Results: Across 15 analyzed classes, the model achieved 87.8% mAP50, 76.9% macro-P, and 86.1% macro-R. The highest per-class AP₅₀ was observed for Deep bite (98.8%), Diastema (97.9%), Angle Class II canine (97.5%), Anterior open bite (92.8%), Midline shift (91.8%), Angle Class II molar (91.1%), Spacing (91%), and Crowding (90.1%). Moderate performance included Anterior crossbite (88.3%), Angle Class III molar (87.4%), Head bite (82.7%), and Posterior open bite (80.2%). Lower values were seen for Angle Class III canine (76%), Posterior crossbite (75.6%), and Big overjet (75.3%). Precision–recall trends indicate earlier precision drop-off for posterior/transverse classes and comparatively more missed detections in Posterior crossbite, whereas Big overjet exhibited more false positives at the chosen threshold. Conclusion: A YOLOv11-based deep learning system can accurately detect several clinically salient malocclusions on routine intraoral photographs, supporting efficient screening and standardized documentation. Performance gaps align with limited examples and visualization constraints in posterior regions. Larger, multi-center datasets, protocol standardization, quantitative metrics, and multimodal inputs may further improve robustness.

Physiological parameters have the potential to serve as valuable early warning indicators for the conservation of animal populations. However, measuring physiological adaptations in wildlife is often challenging, due to intrinsic differences causing natural variations of physiological measures between individuals across species. This study is aimed at addressing this by investigating the influence of intrinsic factors, including sex, age, body condition, and the incubation of eggs on the H/L ratio of a forest bird community. As physiological measure, we used the heterophil to lymphocyte (H/L) ratio of individuals belonging to different species in the forest bird community, which was assessed using a novel deep learning approach based on convolutional neural networks (CNNs) applied to whole blood smear scans. Using phylogenetically controlled Bayesian analyses across the bird species, we found higher H/L ratios in adult birds than in juveniles and observed slightly higher H/L ratios in females than in males. While body condition had no effect on the H/L ratio, incubating birds tended to have higher H/L ratios than non‐reproductive birds, regardless of their sex. Furthermore, we found a robust phylogenetic signal of the H/L ratio in the studied bird community. Our results reveal significant general patterns of the effect of intrinsic factors on the H/L ratio across a bird community.

This paper presents RL4CEP, a reinforcement learning (RL) approach to dynamically update complex event processing (CEP) rules. RL4CEP uses Double Deep Q-Networks to update the threshold values used by CEP rules. It is implemented using Apache Flink as a CEP engine and Apache Kafka for message distribution. RL4CEP is a generic approach for scenarios in which CEP rules need to be updated dynamically. In this paper, we use RL4CEP in a financial trading use case. Our experimental results based on three financial trading rules and eight financial datasets demonstrate the merits of RL4CEP in improving the overall profit, when compared to baseline and state-of-the-art approaches, with a reasonable consumption of resources, i.e., RAM and CPU. Finally, our experiments indicate that RL4CEP is executed quite fast compared to traditional CEP engines processing static rules.