Curriculum Vitae

Doctoral researcher at the Pattern Recognition in Embedded Systems group, Department of Computer Science, TU Dortmund University. Seven peer-reviewed publications, four as first author, one first-author manuscript under review, and a best-paper award at the HIP workshop at ICDAR 2026.

tim.hallyburton@cs.tu-dortmund.de · github.com/TimHal · Google Scholar · ORCID

Research Profile

I build label-efficient recognition systems for low-resource and historical collections. My current method expands pseudo-labels over an unlabelled pool starting from one to five expert-selected exemplars per class, using reconstruction error ratios as the selection criterion. Five years of production software engineering before the PhD taught me how to design and scale systems.

I am looking for a 3–6 month research internship starting in January 2027, with flexibility for a May–July 2027 start. Expected PhD completion: December 2027.

Research Experience

Research Associate (Wissenschaftlicher Mitarbeiter), TU Dortmund University — since 06/2024
Pattern Recognition in Embedded Systems Group · Dortmund, Germany

• Primary researcher on the DFG–SNSF project “Combining Image and Graph-based Neural Networks for Handwriting Recognition”, in collaboration with the AIBEX group at the University of Fribourg.
• Developed PAUL (Prototype-Anchored Unsupervised Learning): per-class autoencoders anchored by one to five expert-selected exemplars iteratively expand pseudo-labels over an unlabelled pool, using reconstruction error ratios as the selection criterion. Evaluated on EMNIST, KMNIST and two newly introduced Omniglot-derived benchmarks against SCAN, DEC, IDEC, JULE and centroid/GMM baselines across DINO, CLIP and convolutional features, reaching 66.9 ± 1.6% accuracy on EMNIST Balanced against 53.3 ± 0.9% for the strongest baseline. Best Paper at the HIP workshop, ICDAR 2026; implementation released.
• Benchmarked nine commercial and open-weight LVLMs for segmentation-free handwriting recognition across four scripts and three difficulty levels. Page-level input beat line segmentation on every handwritten script tested and at lower inference cost, cutting Gemini 2.5 Pro’s CER on Cyrillic from 22.6% to 3.9% — except on the most degraded historical collection (Bentham).
• Co-authored the learned set-aggregation module (X-VLAD) of FormWR, a 387k-page writer-retrieval benchmark with sparse handwriting on printed forms. Transferred to the HisFragIR20 competition benchmark, the method reaches 78.8% mAP and 97.9% Top-1, improving on prior published work by +21.6% mAP and +12.7% Top-1 from a 256-dimensional embedding.
• Ongoing collaborations with the University of Fribourg (A. Fischer, A. Scius-Bertrand, A. F. S. Neto) and interdisciplinary work on the ethics of LLM-based companionship in elder care.

Industry Experience

Software Developer, MotionMiners GmbH — 01/2019 – 05/2024
Full-time and part-time · Dortmund, Germany

• Owned backend and database architecture for the customer-facing MotionMiners SOLUTIONS platform from prototype to public release; the platform now serves more than 100 industrial partners. Work covered graph databases and query languages, object storage, semantic search and, later, LLM integration.
• Built an internal fleet-management system tracking several thousand hardware beacons, with heuristic battery-lifetime monitoring.
• Designed an internal knowledge base integrating existing platforms; evaluated technologies and web standards and advised decision makers during early development.

Technical Student and User, European Organization for Nuclear Research (CERN) — 04/2017 – 10/2018
Full-time, ALICE experiment, grid computing · Geneva, Switzerland

• Developed and administered the ALICE Grid Environment Framework (AliEn) and its Java successor (JAliEn).
• Built simulation software for projecting storage and computing resource requirements of future LHC experiments (published, EPJ Web Conf. 2019).

Working Student, IMAGO IT GmbH — 10/2014 – 03/2017
Full-stack development · Kaiserslautern, Germany

• Primary developer of the original cross-platform iOS/Android Planspiel Börse app, a stock-market simulation used in schools; authored roughly 90% of the codebase, and the app reached around 100,000 users.
• Further full-stack work on e-learning platforms and interactive web applications.

Education

Ph.D. Candidate in Computer Science, TU Dortmund University — since 06/2024
Pattern Recognition Group, advisor: Prof. Dr.-Ing. Gernot A. Fink · Dortmund, Germany

• Dissertation: combining visual and graph-based representations for character and handwriting recognition in low-resource settings. Expected completion 12/2027.
• Funded by the DFG–SNSF project “Combining Image and Graph-based Neural Networks for Handwriting Recognition” (DFG 528122871 / SNSF 217594).
• Summer School on Document Analysis (SSDA) 2026, Vall de Núria.
• International Semantic Web Research Summer School (ISWS) 2025, Bertinoro: knowledge graphs for reliable AI.

M.Sc. Computer Science, TU Dortmund University — 03/2024
Final grade 1.3; completed alongside professional employment · Dortmund, Germany

• Thesis: “IMU Data Anonymisation with Generative Adversarial Networks” — removing identifying information from wearable time-series data while preserving downstream human activity recognition utility. Thesis grade 1.1; results published at ICPR 2024.

B.Sc. Applied Computer Science, University of Applied Sciences Kaiserslautern — 12/2018
Final grade 1.1 (valedictorian); completed alongside dual and full-time employment · Kaiserslautern, Germany

• Thesis: “Annotation and Reflection Based Metaprogramming in Java 8”. Thesis grade 1.0.

German grading scale: 1.0 (excellent) to 4.0 (sufficient); 5.0 is a fail.

Awards and Honours

• Best Paper Award, International Workshop on Historical Document Imaging and Processing (HIP) at ICDAR 2026, Vienna — 2026
• Doctoral Consortium Award, International Conference on Document Analysis and Recognition (ICDAR) 2026, Vienna — 2026
• Best Presentation Award, International Semantic Web Research Summer School (ISWS), Bertinoro, Italy — 2025
• Runner-up, Best Student Paper, ISWS 2025, for “OnToDate: Using Context-Aware PKGs and LLMs to Ensure Trust, Accountability, and Autonomy in Digital Spaces” — 2025
• Scholar of the German Academic Scholarship Foundation (Studienstiftung des deutschen Volkes), Germany’s national academic merit scholarship — since 2016
• Award for the Best Graduate, issued by Novatec Consulting GmbH, University of Applied Sciences Kaiserslautern — 2018

Publications

• Reconstruction Error Ratios for Prototype-Anchored Unsupervised Learning in Optical Character Recognition Best Paper Award International Workshop on Historical Document Imaging and Processing (HIP) at ICDAR, Vienna, Austria, 2026
• Writer Retrieval at Scale International Conference on Document Analysis and Recognition (ICDAR), Vienna, Austria, 2026
• Recent Advances in Information Extraction from Historical Archival Records International Conference on Document Analysis and Recognition (ICDAR), Vienna, Austria, 2026
• Talking as a Drug — Administering LLM Conversations for Medical Treatment in Elder Care CEUR Workshop Proceedings — presented at ETHICAIA, IJCAI 2026, Bremen, Germany, 2026
• Handwriting Recognition Using Large Vision-Language Models: Impact of Segmentation Under review Informatica — submitted June 2026, under review, 2026
• Segmentation-Free Handwriting Recognition in Historical Handwritten Documents Using Large Vision-Language Models 14th International Symposium on Information and Communication Technology (SOICT) — Springer CCIS, 2025
• Anonymisation for Time-Series Human Activity Data Pattern Recognition (ICPR), Lecture Notes in Computer Science, vol. 15315, Springer, pp. 17–32, 2025
• A Simulation Tool for ALICE Storage and Computing Resource Usage EPJ Web of Conferences, vol. 214, p. 03017, 2019

Teaching and Supervision

• Teaching assistant, Pattern Recognition tutorials, TU Dortmund — 2025
• Teaching assistant, Computer Vision tutorials, TU Dortmund — 2024
• Co-supervision of seven B.Sc. and M.Sc. theses in document analysis and representation learning — since 2024

Technical Skills

• Languages: Python, JavaScript/TypeScript, Java, Haskell, SQL
• Deep learning: PyTorch, PyTorch Lightning, PyTorch Geometric, Hugging Face, timm, Torchvision, scikit-learn, NumPy, Pandas
• Models: Vision Transformers, masked autoencoders, DINOv2, CLIP, CNNs, GNNs, VAEs, large vision-language models
• Experimentation and infrastructure: Slurm, multi-GPU training (8×GPU nodes), CUDA, Podman/Docker, MLflow, Hydra, Optuna, vLLM, FAISS
• Web and data: FastAPI, Flask, Angular, Vue, PayloadCMS, relational, NoSQL and graph databases
• Other: Git, Linux (Debian-based), LaTeX, Zotero/JabRef
• Spoken languages: German (native), English (fluent)