Rastislav Hronský
mail@rhronsky.com · hrasto.github.io · GitHub · Tilburg, Netherlands
Research Interests
language representation · tokenization · language model generalisation and domain drift · input representation and language codecs · distributional semantics · psycholinguistics and human language processing · evaluation methodology for NLP systems
Open to pivoting into: algorithmic monitoring, speech, accessible language technologies, robotics, energy grid management and load prediction, niche R&D.
Education
Thesis: Between Digital Text and Language Model: Role of Context in Language Sampling, Segmentation, and Representation. Thesis summary. Supervised by Emmanuel Keuleers. Topics: tokenization, evaluation, context-dependence, segmentation, unsupervised sentence representations, psycholinguistic modelling. Collaboration: systems for generating Chinese pseudocharacters.
Experience
Taught Deep Learning (MSc Data Science & Society) and Data Structures & Algorithms (BSc CSAI). Responsibilities: teaching, practicals, assignment design, examination design and grading. Supervised 7 MSc theses (DSS); second-reader for additional theses.
Research at the intersection of NLP and psycholinguistics. Trained and evaluated neural LMs (Transformers, LSTMs, CNNs) and classical NLP models (LDA, word2vec, LSA, n-grams) for text generation, word & sentence representation. PyTorch, HuggingFace. Designed and deployed text classification models for KPN (customer contact analytics). 5 peer-reviewed publications.
Courses: Data Mining for Business & Governance; Machine Learning.
Implemented t-SNE and DTW distance measures in the Visplore visual analytics system. Developed components for a recurrent pattern (motif) analysis dashboard for time-series data. C++, OpenGL, GTK.
Back-end development with Symfony (PHP, MySQL); Ionic mobile app development.
Built a server-side data visualisation dashboard for the Energy Management department. LAMP stack, ChartJS.
Front-end implementations (HTML, CSS, JS, Google Maps API), PHP programming.
Publications
See the full publications page for links and details.
Hronský, R. & Keuleers, E. A. (under review). Addressing Sample Bias of Language Data in Vocabulary Creation and Evaluation.
Hronský, R. & Keuleers, E. A. (2024). Tokenization via Language Modeling: The Role of Preceding Text. CAWL Workshop at LREC-COLING.
Hronský, R. & Keuleers, E. A. (2023). Role of Context in Unsupervised Sentence Representation Learning: The Case of Dialog Act Modeling. Findings of EMNLP.
Hronský, R. & Keuleers, E. A. (2022). Does the Choice of a Segmentation Algorithm Affect the Performance of Text Classifiers? BNAIC/BeNeLearn.
Hronský, R. & Keuleers, E. A. (2021). Word Probability Re-Estimation Using Topic Modeling and Lexical Decision Data. Proceedings of the Annual Meeting of the Cognitive Science Society.
Skills (fresh)
Skills (refreshable)
Languages
English (fluent) · German (fluent) · Slovak (native) · Dutch (elementary) · Chinese (elementary)
Personal
Running (marathon PB 3:05) · Road cycling & mountain biking (raced as junior; currently spinning instructor) · Bike touring (Tilburg → Bratislava) · Weight lifting (5×5 @ 130 kg squat) · Interest in nutrition