Andrzej Janusz
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Andrzej Janusz has a total of 42 co-authors. Here are the top co-authors by number of shared publications:
- Dominik Ślęzak
University of Warsaw, Warsaw, Poland
9 shared publications - Marek Grzegorowski
Institute of Informatics, University of Warsaw, Warsaw, Poland
7 shared publications - Daniel Kałuża
University of Warsaw
4 shared publications - Łukasz Marcinowski
FitFood Solskiego 11/28 31-216 Cracow Poland
3 shared publications - Dominik Ślęzak2 shared publications
- Grzegorz Śliwa
FitFood Solskiego 11/28 31-216 Cracow Poland
2 shared publications
Latest Publications
- Open Access
- 30 September 2026
- Published by Sungkyunkwan University School of Medicine in Precision and Future Medicine
- Vol. 10 (3) , 150-159
Generative methods, validation frameworks, and equity considerations of synthetic data in medical artificial intelligence
Synthetic data is emerging as a strategic solution to persistent barriers in medical artificial intelligence (AI) development, including data scarcity, privacy constraints, and dataset bias. This perspective examines the landscape of synthetic multimodal data in healthcare, including its mechanisms, demonstrated benefits, technical and ethical challenges, and a proposed validation framework linking fidelity to clinical utility. We integrate evidence from recent empirical studies, the generative AI literature, and regulatory developments. We discuss how synthetic data is generated across four healthcare domains: medical imaging, electronic health records, clinical text, and genomics. We also review methods such as generative adversarial networks, diffusion models, large language models, and rule-based simulators. Synthetic multimodal data demonstrably improves AI model performance, addresses class imbalance, enables privacy- preserving data-sharing, and supports regulatory evaluation pathways. Illustrative case studies suggest potential gains in model performance, although systematic quantitative evidence across modalities remains limited. Key challenges include re-identification risks, the perpetuation of bias, computational inequity, and the absence of cross-modal coherence standards. A seven-dimensional validation scorecard is proposed, covering statistical fidelity, privacy risk, clinical utility, robustness, cross-modal coherence, and governance. Synthetic multimodal data can catalyze medical AI innovation, but only if paired with rigorous validation, transparent governance, and a firm commitment to equity. Priority actions include international benchmark consortia, regulatory sandboxes, investment in federated learning, and mandatory intersectional bias auditing.
- 30 September 2026
- Published by Sungkyunkwan University School of Medicine in Precision and Future Medicine
- Vol. 10 (3) , 150-159
Open AccessGenerative methods, validation frameworks, and equity considerations of synthetic data in medical artificial intelligence
Synthetic data is emerging as a strategic solution to persistent barriers in medical artificial intelligence (AI) development, including data scarcity, privacy constraints, and dataset bias. This perspective examines the landscape of synthetic multimodal data in healthcare, including its mechanisms, demonstrated benefits, technical and ethical challenges, and a proposed validation framework linking fidelity to clinical utility. We integrate evidence from recent empirical studies, the generative AI literature, and regulatory developments. We discuss how synthetic data is generated across four healthcare domains: medical imaging, electronic health records, clinical text, and genomics. We also review methods such as generative adversarial networks, diffusion models, large language models, and rule-based simulators. Synthetic multimodal data demonstrably improves AI model performance, addresses class imbalance, enables privacy- preserving data-sharing, and supports regulatory evaluation pathways. Illustrative case studies suggest potential gains in model performance, although systematic quantitative evidence across modalities remains limited. Key challenges include re-identification risks, the perpetuation of bias, computational inequity, and the absence of cross-modal coherence standards. A seven-dimensional validation scorecard is proposed, covering statistical fidelity, privacy risk, clinical utility, robustness, cross-modal coherence, and governance. Synthetic multimodal data can catalyze medical AI innovation, but only if paired with rigorous validation, transparent governance, and a firm commitment to equity. Priority actions include international benchmark consortia, regulatory sandboxes, investment in federated learning, and mandatory intersectional bias auditing.
- 9 June 2026
- Published by Springer Nature
- p. 387-404
DupliMend: Online Detection and Refinement of Imprecise Activity Labels
- 9 June 2026
- Published by Springer Nature
- p. 387-404
DupliMend: Online Detection and Refinement of Imprecise Activity Labels
- 15 April 2026
- Published by Elsevier in Applied Soft Computing
- Vol. 198
Fast approximate loss change for active learning on imbalanced data
- 15 April 2026
- Published by Elsevier in Applied Soft Computing
- Vol. 198
Fast approximate loss change for active learning on imbalanced data
- 15 June 2025
- Published by Springer Nature
- p. 205-222
ThreatTrace: Cyber-Attack Detection Through Trace Abstraction and Soft Clustering
- 15 June 2025
- Published by Springer Nature
- p. 205-222
ThreatTrace: Cyber-Attack Detection Through Trace Abstraction and Soft Clustering
- 14 June 2025
- Published by Springer Nature
- p. 129-141
Suspicious Activity Detection Using Blockchain Process Mining
- 14 June 2025
- Published by Springer Nature
- p. 129-141
Suspicious Activity Detection Using Blockchain Process Mining
- Open Access
- 1 January 2025
- Published by University of Zielona Góra, Poland in International Journal of Applied Mathematics and Computer Science
- Vol. 35 (1) , 33-43
Evidence-theoretical modeling of uncertainty induced by posterior probability distributions
We discuss how the posterior probability distributions produced by machine learning models for analyzed...
- 1 January 2025
- Published by University of Zielona Góra, Poland in International Journal of Applied Mathematics and Computer Science
- Vol. 35 (1) , 33-43
Open AccessEvidence-theoretical modeling of uncertainty induced by posterior probability distributions
We discuss how the posterior probability distributions produced by machine learning models for analyzed...
- Open Access
- 1 January 2025
- Published by University of Zielona Góra, Poland in International Journal of Applied Mathematics and Computer Science
- Vol. 35 (1) , 19-31
On explainability of cluster prototypes with rough sets: A case study in the FMCG market
Despite the growing popularity of machine learning (ML), such solutions are often incomprehensible to...
- 1 January 2025
- Published by University of Zielona Góra, Poland in International Journal of Applied Mathematics and Computer Science
- Vol. 35 (1) , 19-31
Open AccessOn explainability of cluster prototypes with rough sets: A case study in the FMCG market
Despite the growing popularity of machine learning (ML), such solutions are often incomprehensible to...
- 31 December 2023
- Published by Springer Nature
- p. 371-386
Towards ML Explainability with Rough Sets, Clustering, and Dimensionality Reduction
- 31 December 2023
- Published by Springer Nature
- p. 371-386
Towards ML Explainability with Rough Sets, Clustering, and Dimensionality Reduction
- 31 December 2023
- Published by Springer Nature
- p. 479-494
On Several New Dempster-Shafer-Inspired Uncertainty Measures Applicable for Active Learning
- 31 December 2023
- Published by Springer Nature
- p. 479-494
On Several New Dempster-Shafer-Inspired Uncertainty Measures Applicable for Active Learning
- Open Access
- 28 September 2023
- Published by IOS Press
- p. 1207-1214
Robust Assignment of Labels for Active Learning with Sparse and Noisy Annotations
The performance of classification models depends on the quality of training data labels. Unfortunately, acquiring good-quality annotations for many tasks is infeasible or too expensive. To address this challenge, active learning algorithms are commonly employed to select only the most relevant data for labeling. However, this is possible only when the quality and quantity of labels acquired from experts are sufficient. In many applications, a trade-off between annotating individual samples by multiple annotators to increase label quality vs. annotating new samples to increase the total number of labeled instances is necessary. In this paper, we address the issue of faulty data annotations in the context of active learning. In particular, we propose two novel annotation unification algorithms that utilize unlabeled parts of the sample space. The proposed methods require little to no intersection between samples annotated by different experts. Our experiments on four public datasets indicate the robustness and superiority of the proposed methods in both, the estimation of the annotator's reliability, and the assignment of actual labels, against the state-of-the-art algorithms and the simple majority voting.
- 28 September 2023
- Published by IOS Press
- p. 1207-1214
Open AccessRobust Assignment of Labels for Active Learning with Sparse and Noisy Annotations
The performance of classification models depends on the quality of training data labels. Unfortunately, acquiring good-quality annotations for many tasks is infeasible or too expensive. To address this challenge, active learning algorithms are commonly employed to select only the most relevant data for labeling. However, this is possible only when the quality and quantity of labels acquired from experts are sufficient. In many applications, a trade-off between annotating individual samples by multiple annotators to increase label quality vs. annotating new samples to increase the total number of labeled instances is necessary. In this paper, we address the issue of faulty data annotations in the context of active learning. In particular, we propose two novel annotation unification algorithms that utilize unlabeled parts of the sample space. The proposed methods require little to no intersection between samples annotated by different experts. Our experiments on four public datasets indicate the robustness and superiority of the proposed methods in both, the estimation of the annotator's reliability, and the assignment of actual labels, against the state-of-the-art algorithms and the simple majority voting.