Research
Publications
Peer-reviewed papers and preprints by the three of us and our co-authors: generative models for synthetic data and how to evaluate them, deep learning on customer transaction data, and large-scale computing for brain imaging.
Papers
TabularARGN: A Flexible and Efficient Auto-Regressive Framework for Generating High-Fidelity Synthetic Data
Introduces TabularARGN, an auto-regressive network for mixed-type and sequential tables that can generate data conditioned on any subset of columns. The benchmarks in the paper show state-of-the-art quality with much shorter training and generation times.
Tiwald, P., Krchova, I., Sidorenko, A., Vargas Vieyra, M., Scriminaci, M., Platzer, M. · arXiv:2501.12012 · 2025
TabularARGN: An Auto-Regressive Generative Network for Tabular Data Generation
The peer-reviewed short version of the TabularARGN paper (four pages), focused on fidelity and efficiency across established benchmarks.
Sidorenko, A., Krchova, I., Vargas Vieyra, M., Tiwald, P., Scriminaci, M., Platzer, M. · EurIPS 2025 Workshop: AI for Tabular Data · 2025
Privacy-Preserving Tabular Synthetic Data Generation Using TabularARGN
Tests how well TabularARGN output protects privacy, using systematic membership-inference attacks, and compares statistical similarity and machine-learning utility with other generators.
Sidorenko, A., Tiwald, P. · arXiv:2508.06647 · 2025
Benchmarking Synthetic Tabular Data: A Multi-Dimensional Evaluation Framework
A holdout-based method for measuring whether synthetic data is both accurate and private, using distribution comparisons, embedding similarity and nearest-neighbor distances. The code is open source.
Sidorenko, A., Platzer, M., Scriminaci, M., Tiwald, P. · arXiv:2504.01908 · 2025
A Note on Statistically Accurate Tabular Data Generation Using Large Language Models
Existing LLM-based methods often lose the dependencies between features, especially categorical ones. This note has the LLM estimate conditional probability distributions instead, and the results show this can improve statistical fidelity.
Sidorenko, A. · arXiv:2505.02659 · 2025
Democratizing Tabular Data Access with an Open-Source Synthetic Data SDK
Presents the open-source Python SDK built on TabularARGN: multi-table and sequential data, differential privacy, fairness-aware generation and automated quality checks in one toolkit.
Krchova, I., Vargas Vieyra, M., Scriminaci, M., Sidorenko, A. · arXiv:2508.00718 · 2025
Deep Generative Models for Synthetic Data: A Survey
A review of deep generative models for sequential synthetic data, with an evaluation framework built on five criteria: representativeness, novelty, realism, diversity and coherence. It shows which criteria matter most in which domains, from language and audio to healthcare and mobility data.
Eigenschink, P., Reutterer, T., Vamosi, S., Vamosi, R., Sun, C., Kalcher, K. · IEEE Access 11 · 2023
Customer Base Analysis with Recurrent Neural Networks
Uses LSTM networks on customer transaction histories to forecast future purchases when churn is never directly observed. Across eight real datasets the model is more accurate per customer and less biased per cohort than established models, and it captures seasonal patterns. An open-source implementation is available.
Valendin, J., Reutterer, T., Platzer, M., Kalcher, K. · International Journal of Research in Marketing 39(4) · 2022
Big Data Approaches for the Analysis of Large-Scale fMRI Data Using Apache Spark and GPU Processing: A Demonstration on Resting-State fMRI Data from the Human Connectome Project
Software tools for analyzing very large fMRI datasets with Apache Spark and GPUs, including distributed loading of 4D brain-imaging files, demonstrated on resting-state data from the Human Connectome Project.
Boubela, R. N., Kalcher, K., Huf, W., Našel, C., Moser, E. · Frontiers in Neuroscience 9:492 · 2016