Hi, my name is Nicholas Popovič.
I am a fourth-year PhD student in Michael Färber's group at TU Dresden & ScaDs.AI. Prior to that I was a member of the Web Science group at Karlsruhe Institute of Technology.
My research focuses on natural language processing and machine learning, particularly on representation learning for information extraction.
You can contact me via email, LinkedIn, Twitter, Bluesky, Google Scholar, or GitHub.
News
| 01/2026 | Kaleidoscope accepted to ICLR'26! Thanks and congrats to everyone involved! |
|---|---|
| 09/2025 | One paper accepted to EMNLP'25! |
| 07/2025 | Presented a paper at the XLLM workshop at ACL 2025. |
| 06/2025 | Held a tutorial session at the 11th International Summer School on AI and Big Data about Language Models. |
| 12/2024 | Gave a talk on 'Encoding or Decoding? NLU in the Age of Generative Language Models.' |
| 09/2024 | EMBER got accepted to EMNLP'24 as a main conference paper! |
spotlight
-
Extractive Fact Decomposition for Interpretable Natural Language Inference in One Forward Pass Nicholas Popovič, Michael Färber EMNLP, 2025, [PDF], [CODE], [WEBSITE] -
Embedded Named Entity Recognition using Probing Classifiers Nicholas Popovič, Michael Färber EMNLP, 2024, [PDF], [CODE], [WEBSITE] -
Few-Shot Document-Level Relation Extraction Nicholas Popovič, Michael Färber NAACL, 2022, [PDF], [CODE], [WEBSITE]
publications
The short summaries you see below were generated using Claude Opus 4.6. They might not be 100% accurate. Please take with a huge grain of salt :)
-
Benchmarking Uncertainty Calibration in Large Language Model Long-Form Question Answering Philip Müller, Nicholas Popovič, Michael Färber, Peter Steinbach preprint, 2026, [PDF], [CODE]AI-generated summary❓ Do AI models know when they're unsure?
💡 Barely—self-reported confidence is unreliable, but checking consistency across multiple answers helps. -
Kaleidoscope: In-language Exams for Massively Multilingual Vision Evaluation Salazar et al. ICLR, 2026, [PDF], [CODE], [WEBSITE]AI-generated summary❓ Can AI understand images in non-English languages?
💡 Poorly—especially for low-resource languages and image-heavy questions. -
⭐️ Extractive Fact Decomposition for Interpretable Natural Language Inference in One Forward Pass Nicholas Popovič, Michael Färber EMNLP, 2025, [PDF], [CODE], [WEBSITE]AI-generated summary❓ Can a small model fact-check by breaking claims into pieces?
💡 Yes—no expensive large model needed. -
DocIE@XLLM25: In-Context Learning for Information Extraction using Fully Synthetic Demonstrations Nicholas Popovič, Ashish Kangen, Tim Schopf, Michael Färber XLLM @ ACL (workshop), 2025, [PDF], [CODE], [WEBSITE]AI-generated summary❓ Can AI extract info from documents without human-labeled examples?
💡 Yes—by generating its own synthetic examples to learn from. -
The Effects of Hallucinations in Synthetic Training Data for Relation Extraction Steven Rogulsky, Nicholas Popovič, Michael Färber KBC-LM @ ISWC (workshop), 2024, [PDF]AI-generated summary❓ Does AI-generated training data with hallucinations cause problems?
💡 Yes—relevant fabrications hurt recall by up to 39%, but filtering helps. -
⭐️ Embedded Named Entity Recognition using Probing Classifiers Nicholas Popovič, Michael Färber EMNLP, 2024, [PDF], [CODE], [WEBSITE]AI-generated summary❓ Can we spot names and entities in AI text as it streams out?
💡 Yes—with only ~1% slowdown. -
VOCAB-EXPANDER: A System for Creating Domain-Specific Vocabularies Based on Word Embeddings Michael Färber, Nicholas Popovič RANLP (demo), 2023, [PDF], [CODE], [DEMO]AI-generated summary🛠️ Auto-builds specialized word lists for any field using word embeddings.
-
Towards solving fuzzy tasks with human feedback: A retrospective of the minerl basalt 2022 competition Milani et al. Competitions Track @ NeurIPS (workshop), 2022, [PDF], [CODE]AI-generated summary❓ Can AI learn vague tasks from human feedback?
💡 Somewhat—but vague goals remain very hard. -
⭐️ Few-Shot Document-Level Relation Extraction Nicholas Popovič, Michael Färber NAACL, 2022, [PDF], [CODE], [WEBSITE]AI-generated summary❓ Can AI find relationships in documents from just a few examples?
💡 It struggles—much harder than in single sentences. -
AIFB-WebScience at SemEval-2022 Task 12: Relation Extraction First - Using Relation Extraction to Identify Entities Nicholas Popovič, Walter Laurito, Michael Färber SemEval @ NAACL (workshop), 2022, [PDF], [CODE], [WEBSITE]AI-generated summary❓ Can we decode math symbols in papers by finding relationships first?
💡 Yes—and it works even with incomplete training labels. -
Which Publications' Metadata Are in Which Bibliographic Databases? A System for Exploration Michael Färber, Christoph Braun, Nicholas Popovič, Tarek Saier, Kristian Noullet BIR @ ECIR (demo), 2022, [PDF], [CODE]AI-generated summary🛠️ Explore which academic search engines actually index which papers—the gaps are surprising.
-
Ampacity forecasting from numerical weather predictions: a fusion of the traditional and machine learning methods Gabriela Molinar, Johannes Bassler, Nicholas Popovič, Wilhelm Stork IEEE ISGT-Europe, 2020, [PDF]AI-generated summary❓ Can we better predict a power line's safe capacity?
💡 Yes—combining physics with ML outperforms either alone. -
From data points to ampacity forecasting: Gated recurrent unit networks Nicholas Popovič, Gabriela Molinar, Wilhelm Stork IEEE BigDataService, 2018, [PDF]AI-generated summary❓ Can a neural network predict safe power line capacity?
💡 Yes—it learns time-based patterns from historical data.