Publications | OdenseNLP

2026 TMLR

Transactions on Machine Learning Research

Synthetic Data Privacy

Disjoint Generation of Synthetic Data

Lautrup, Anton Danholt; Rajabinasab, Muhammad; Hyrup, Tobias; Zimek, Arthur; Schneider-Kamp, Peter

Read paper

2026 arXiv

arXiv preprint

Model Editing Research Tools

BrainSurgery: Reproducible and Reliable Declarative Weight Manipulations for Model Editing and Upcycling

Barmina, Gianluca; Pirchert, Annemette Broch; Blasi Núñez, Andrea; Poech, Lukas Galke; Schneider-Kamp, Peter

Read paper

2026 arXiv

arXiv preprint

AI Safety Post-training

PsychoSafe: Eliciting Psychologically-Informed Refusals in Large Language Models

Barmina, Gianluca; Torrielli, Federico; Harms, Sven; Nielsen, Jacob; Mächtle, Felix; Beltoft, Stine Lyngsø; Schneider-Kamp, Peter; Eisenbarth, Thomas; Poech, Lukas Galke; Lauscher, Anne

Read paper

2026 arXiv

arXiv preprint

Danish NLP Evaluation

SDUs DAISY: A Benchmark for Danish Culture

Nielsen, Jacob; Beltoft, Stine L.; Schneider-Kamp, Peter; Poech, Lukas Galke

Read paper

2026 ACL Anthology

LREC 2026

Danish NLP Data

Dynaword: From One-shot to Continuously Developed Datasets

Enevoldsen, Kenneth; Jensen, Kristian Nørgaard; Kostkan, Jan; Szabó, Balázs; Kardos, Márton; Vad, Kirsten; Heinsen, Johan; Blasi Núñez, Andrea; Barmina, Gianluca; Nielsen, Jacob; Larsen, Rasmus; van der Goot, Rob; Vahlstrup, Peter; Dalum, Per Møldrup; Elliott, Desmond; Galke Poech, Lukas; Schneider-Kamp, Peter; Nielbo, Kristoffer

Read paper

2026 arXiv

Technical report

Danish NLP Efficient NLP Reasoning

DFM Mimir v1: An Open HRM Delivering Frontier Performance at 1B Parameters Using Only Permissible Post-Training Data

Schneider-Kamp, Peter; Nielsen, Jacob; Barmina, Gianluca; Enevoldsen, Kenneth; Galke Poech, Lukas

Read paper

2026 ACL Anthology

LREC 2026

Evaluation Reasoning

SommBench: Assessing Sommelier Expertise of Language Models

Brach, William; Bedej, Tomas; Nielsen, Jacob; Pichna, Jacob; Bedej, Juraj; Saarensilta, Eemeli; Dupouy, Julie; Barmina, Gianluca; Blasi Núñez, Andrea; Schneider-Kamp, Peter; Košťál, Kristian; Ries, Michal; Galke Poech, Lukas

Read paper

2026 SAGE

International Journal of Qualitative Methods

Qualitative AI Multimodal NLP

Drawing out What They Struggle to say: AI-Augmented Analysis of Projective Techniques in Qualitative Health Research

Beltoft, Stine Lyngsø; Nielsen, Jacob; Askegaard, Søren; Schneider-Kamp, Anna

Read paper

2026 arXiv

arXiv preprint

AI Safety Memorization Evaluation

LLMs Can Leak Training Data But Do They Want To? A Propensity-Aware Evaluation of Memorization in LLMs

Barmina, Gianluca; Schneider-Kamp, Peter; Poech, Lukas Galke

Read paper

2026 arXiv

arXiv preprint

AI Safety Agents

The Moltbook Files: A Harmless Slopocalypse or Humanity's Last Experiment

Brach, William; Torrielli, Federico; Beltoft, Stine Lyngso; Pirchert, Annemette Brok; Schneider-Kamp, Peter; Poech, Lukas Galke

Read paper

2026 arXiv

arXiv preprint

Interpretability AI Safety

Confidence and Calibration of Activation Oracles for Reliable Interpretation of Language Model Internals

Torrielli, Federico; Schneider-Kamp, Peter; Poech, Lukas Galke

Read paper

2026 arXiv

arXiv preprint

AI Safety Agents

Emergent Languages in Populations of Language Model Agents: From Token Efficiency to Oversight Evasion

Beltoft, Stine Lyngso; Brach, William; Torrielli, Federico; Nielsen, Jacob; Pirchert, Annemette Brok; Tonini, Filippo; Schneider-Kamp, Peter; Poech, Lukas Galke

Read paper

2026 arXiv

arXiv preprint

Efficient NLP Federated Learning

FlexMoRE: A Flexible Mixture of Rank-heterogeneous Experts for Efficient Federatedly-trained Large Language Models

Pirchert, Annemette Brok; Nielsen, Jacob; From, Mogens Henrik; Poech, Lukas Galke; Schneider-Kamp, Peter

Read paper

2025 Springer

Data Mining and Knowledge Discovery

Synthetic Data Evaluation Privacy

Syntheval: a framework for detailed utility and privacy evaluation of tabular synthetic data

Lautrup, Anton D.; Hyrup, Tobias; Zimek, Arthur; Schneider-Kamp, Peter

Read paper

2025 SDU Research Portal

Ph.D. dissertation, University of Southern Denmark

Synthetic Data Evaluation

Generation and Evaluation of Realistic Tabular Synthetic Data

Lautrup, Anton Danholt

Read paper

2025 ACL Anthology

RANLP 2025

Data Multimodal NLP

MLDataForge: Accelerating Large-Scale Dataset Preprocessing and Access for Multimodal Foundation Model Training

Blasi Nunez, Andrea; Galke, Lukas Paul Achatius; Schneider-Kamp, Peter

Read paper

2025 arXiv

arXiv preprint

AI Safety

Not Everything That Counts Can Be Counted: A Case for Safe Qualitative AI

Beltoft, Stine; Galke, Lukas

Read paper

2025 arXiv

arXiv preprint

Summarization

Chain of Summaries: Summarization Through Iterative Questioning

Brach, William; Kostal, Kristian; Poech, Lukas Galke

Read paper

2025 ACL Anthology

Findings of IJCNLP-AACL 2025

Interpretability Multilingual NLP

Isolating Culture Neurons in Multilingual Large Language Models

Namazifard, Danial; Poech, Lukas Galke

Read paper

2025 AAAI

AAAI 2026

Efficient NLP Distributed Training

DeToNATION: Decoupled Torch Network-Aware Training on Interlinked Online Nodes

From, Mogens Henrik; Nielsen, Jacob; Poech, Lukas Galke; Schneider-Kamp, Peter

Read paper

2025 arXiv

arXiv preprint

Tool Use Reasoning

Training Language Models to Use Prolog as a Tool

Mellgren, Niklas; Schneider-Kamp, Peter; Poech, Lukas Galke

Read paper

2025 ACL Anthology

Findings of ACL 2025

Efficient NLP Quantization

Continual Quantization-Aware Pre-Training: When to transition from 16-bit to 1.58-bit pre-training for BitNet language models?

Nielsen, Jacob; Schneider-Kamp, Peter; Poech, Lukas Galke

Read paper

2025 arXiv

FAIEMA 2025

Agents

Super-additive Cooperation in Language Model Agents

Tonini, Filippo; Galke, Lukas

Read paper

2025 arXiv

LREC 2026

Danish NLP Evaluation

DaLA: Danish Linguistic Acceptability Evaluation Guided by Real World Errors

Barmina, Gianluca; Norman, Nathalie Carmen Hau; Schneider-Kamp, Peter; Poech, Lukas Galke

Read paper

2024 arXiv

arXiv preprint

Efficient NLP Quantization

When are 1.58 bits enough? A Bottom-up Exploration of BitNet Quantization

Nielsen, Jacob; Poech, Lukas Galke; Schneider-Kamp, Peter

Read paper

2024 Springer

DeLTA 2024

Efficient NLP Quantization

BitNet B1.58 Reloaded: State-of-the-Art Performance Also on Smaller Networks

Nielsen, Jacob; Schneider-Kamp, Peter

Read paper

Filter by topic

Disjoint Generation of Synthetic Data

BrainSurgery: Reproducible and Reliable Declarative Weight Manipulations for Model Editing and Upcycling

PsychoSafe: Eliciting Psychologically-Informed Refusals in Large Language Models

SDUs DAISY: A Benchmark for Danish Culture

Dynaword: From One-shot to Continuously Developed Datasets

DFM Mimir v1: An Open HRM Delivering Frontier Performance at 1B Parameters Using Only Permissible Post-Training Data

SommBench: Assessing Sommelier Expertise of Language Models

Drawing out What They Struggle to say: AI-Augmented Analysis of Projective Techniques in Qualitative Health Research

LLMs Can Leak Training Data But Do They Want To? A Propensity-Aware Evaluation of Memorization in LLMs

The Moltbook Files: A Harmless Slopocalypse or Humanity's Last Experiment

Confidence and Calibration of Activation Oracles for Reliable Interpretation of Language Model Internals

Emergent Languages in Populations of Language Model Agents: From Token Efficiency to Oversight Evasion

FlexMoRE: A Flexible Mixture of Rank-heterogeneous Experts for Efficient Federatedly-trained Large Language Models

Syntheval: a framework for detailed utility and privacy evaluation of tabular synthetic data

Generation and Evaluation of Realistic Tabular Synthetic Data

MLDataForge: Accelerating Large-Scale Dataset Preprocessing and Access for Multimodal Foundation Model Training

Not Everything That Counts Can Be Counted: A Case for Safe Qualitative AI

Chain of Summaries: Summarization Through Iterative Questioning

Isolating Culture Neurons in Multilingual Large Language Models

DeToNATION: Decoupled Torch Network-Aware Training on Interlinked Online Nodes

Training Language Models to Use Prolog as a Tool

Continual Quantization-Aware Pre-Training: When to transition from 16-bit to 1.58-bit pre-training for BitNet language models?

Super-additive Cooperation in Language Model Agents

DaLA: Danish Linguistic Acceptability Evaluation Guided by Real World Errors

When are 1.58 bits enough? A Bottom-up Exploration of BitNet Quantization

BitNet B1.58 Reloaded: State-of-the-Art Performance Also on Smaller Networks