Maryam Amir Haeri
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Maryam Amir Haeri has a total of 49 co-authors. Here are the top co-authors by number of shared publications:
- Gianluigi Folino
ICAR-CNR, Rende, Italy
4 shared publications - Mohammad Mehdi Ebadzadeh
Amirkabir University of Technology, Tehran, Iran
4 shared publications - Michael J. A. M. van Putten
Leiden University Medical Center 1 Department of Human Genetics 2333 ZC Leiden
2 shared publications - Pantelis M. Papadopoulos
Department of Learning, Data Analytics, and Technology, University of Twente, Enschede, The Netherlands
2 shared publications - Radin Hamidi Rad
Department of Computer Engineering and Information Technology Amirkabir University of Technology Tehran Iran
2 shared publications - Seyed Amin Fadaee
Department of Computer Science and Information Technology Amirkabir University of Technology Iran
2 shared publications
Latest Publications
- Open Access
- 30 October 2026
- Published by Elsevier in SSRN Electronic Journal
Privacy-preserving federated tensor decomposition of single-cell immune data: recovering multicellular programs across institutions
**Background and Objectives:** Tensor decomposition of single-cell data identifies multicellular programs but needs pooled data, which consent and data-protection rules often prohibit. We develop a federated estimator and evaluate its approximation to the pooled analysis and the leakage of released objects.**Methods:** Sites centre their data on a federated global mean and send a truncated SVD summary of the local program subspace; a coordinator takes the top singular vectors of the stacked summaries. The merge is approximate, with error set by the omitted local singular components. When sites profile different cell types, donor Gram matrices are summed. A membership-inference attack measures leakage. Seven public atlases were analysed.**Results:** In a lupus cohort, federated recovery closely approximates the pooled result (subspace correlation 0.967; AUC difference 0.000, 95% CI [-0.004, +0.012]), also across multi-ancestry splits and the three original COVID-19 collection sites (correlation 0.989). When no site observes all cell types, donor Gram matrices recover the full program (correlation 1.000, exact by construction). Membership-inference AUC is 0.906 for site-specific subspaces, 0.609 for the merged global subspace and 1.000 for the donor Gram. The program out-predicts the best single cell type in lung (AUC 0.959 vs 0.910) and liver and is comparable elsewhere.**Conclusions:** Cell-type-incomplete sites can be federated where fixed-feature federated PCA cannot. The stacked merge is not an additive aggregation, and the released program carries no formal privacy guarantee. The predictive advantage of multicellularity is tissue-dependent. The code and the public data are released.
- 30 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPrivacy-preserving federated tensor decomposition of single-cell immune data: recovering multicellular programs across institutions
**Background and Objectives:** Tensor decomposition of single-cell data identifies multicellular programs but needs pooled data, which consent and data-protection rules often prohibit. We develop a federated estimator and evaluate its approximation to the pooled analysis and the leakage of released objects.**Methods:** Sites centre their data on a federated global mean and send a truncated SVD summary of the local program subspace; a coordinator takes the top singular vectors of the stacked summaries. The merge is approximate, with error set by the omitted local singular components. When sites profile different cell types, donor Gram matrices are summed. A membership-inference attack measures leakage. Seven public atlases were analysed.**Results:** In a lupus cohort, federated recovery closely approximates the pooled result (subspace correlation 0.967; AUC difference 0.000, 95% CI [-0.004, +0.012]), also across multi-ancestry splits and the three original COVID-19 collection sites (correlation 0.989). When no site observes all cell types, donor Gram matrices recover the full program (correlation 1.000, exact by construction). Membership-inference AUC is 0.906 for site-specific subspaces, 0.609 for the merged global subspace and 1.000 for the donor Gram. The program out-predicts the best single cell type in lung (AUC 0.959 vs 0.910) and liver and is comparable elsewhere.**Conclusions:** Cell-type-incomplete sites can be federated where fixed-feature federated PCA cannot. The stacked merge is not an additive aggregation, and the released program carries no formal privacy guarantee. The predictive advantage of multicellularity is tissue-dependent. The code and the public data are released.
- Open Access
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-6
The Glitch Between Data Doubles and Lived Experiences: A Duoethnographic Study of Engagement with Commercial Health Apps
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-6
Open AccessThe Glitch Between Data Doubles and Lived Experiences: A Duoethnographic Study of Engagement with Commercial Health Apps
- Open Access
- 9 April 2026
- Published by IOP Publishing in Environmental Research Communications
- Vol. 8 (4)
A Geospatial Approach to Neighbourhood Prioritisation for Urban Greening Under Climate–Health Risk
Urban vulnerability arises from the interaction of environmental exposures, socio-demographic inequalities, and public-health stressors. Urban green space is part of the solution, but its adaptive value depends on vegetation type and spatial distribution. We developed a spatial, indicator-based framework for Enschede, the Netherlands, to support neighbourhood prioritisation for urban greening under climate-health risk. We integrated indicators of heat exposure, flood exposure, mental health, physical health, and socio-demographic status into a composite score to rank neighbourhoods by relative need, and modelled associations between tree canopy and grass cover and each indicator while accounting for socio-demographic status. Tree cover showed the strongest and most consistent associations with lower climate exposure and better mental health. Grass cover was also associated with enhancing heat and reducing flood exposure, but associations were smaller and less consistent. Physical health was primarily associated with socio-demographic status. A key spatial finding is that climate exposures follow an inverse distribution: flood risk concentrates in the dense urban core while heat stress is more pronounced in peri-urban neighbourhoods with lower canopy density when using the 41°C threshold. Spatial analysis further confirmed that areas with Physical Equivalent Temperature below 41°C correspond almost exclusively to locations with tree canopy or forest cover, underscoring the specific cooling function of trees relative to other vegetation types. These results support spatially differentiated, municipality-level prioritisation of greening, with emphasis on protecting and expanding tree canopy in high-need areas and complementing canopy interventions with permeable surfaces where flood risk dominates. Although cross-sectional and not causal, the framework is transparent, reproducible, and suitable for participatory use, providing a practical starting point for locally responsive and evidence-informed urban resilience planning.
- 9 April 2026
- Published by IOP Publishing in Environmental Research Communications
- Vol. 8 (4)
Open AccessA Geospatial Approach to Neighbourhood Prioritisation for Urban Greening Under Climate–Health Risk
Urban vulnerability arises from the interaction of environmental exposures, socio-demographic inequalities, and public-health stressors. Urban green space is part of the solution, but its adaptive value depends on vegetation type and spatial distribution. We developed a spatial, indicator-based framework for Enschede, the Netherlands, to support neighbourhood prioritisation for urban greening under climate-health risk. We integrated indicators of heat exposure, flood exposure, mental health, physical health, and socio-demographic status into a composite score to rank neighbourhoods by relative need, and modelled associations between tree canopy and grass cover and each indicator while accounting for socio-demographic status. Tree cover showed the strongest and most consistent associations with lower climate exposure and better mental health. Grass cover was also associated with enhancing heat and reducing flood exposure, but associations were smaller and less consistent. Physical health was primarily associated with socio-demographic status. A key spatial finding is that climate exposures follow an inverse distribution: flood risk concentrates in the dense urban core while heat stress is more pronounced in peri-urban neighbourhoods with lower canopy density when using the 41°C threshold. Spatial analysis further confirmed that areas with Physical Equivalent Temperature below 41°C correspond almost exclusively to locations with tree canopy or forest cover, underscoring the specific cooling function of trees relative to other vegetation types. These results support spatially differentiated, municipality-level prioritisation of greening, with emphasis on protecting and expanding tree canopy in high-need areas and complementing canopy interventions with permeable surfaces where flood risk dominates. Although cross-sectional and not causal, the framework is transparent, reproducible, and suitable for participatory use, providing a practical starting point for locally responsive and evidence-informed urban resilience planning.
- Open Access
- 1 April 2026
- Published by Elsevier in Studies in Educational Evaluation
- Vol. 89
Concept map analysis of reading comprehension assessment: Tracing situation models in multiple-choice and constructed-response items
- 1 April 2026
- Published by Elsevier in Studies in Educational Evaluation
- Vol. 89
Open AccessConcept map analysis of reading comprehension assessment: Tracing situation models in multiple-choice and constructed-response items
- 2 January 2026
- Published by Springer Nature in Universal Access in the Information Society
- Vol. 25 (1) , 18
What if conversational agents’ answers appear correct but are judgmental? Piloting the potential effect on people’s experience
- 2 January 2026
- Published by Springer Nature in Universal Access in the Information Society
- Vol. 25 (1) , 18
What if conversational agents’ answers appear correct but are judgmental? Piloting the potential effect on people’s experience
- 21 July 2025
- Published by Springer Nature
- p. 156-163
Searching for Transactivity in a Collaborative Chat: Analysis of Novelty and Reference with GenAI
- 21 July 2025
- Published by Springer Nature
- p. 156-163
Searching for Transactivity in a Collaborative Chat: Analysis of Novelty and Reference with GenAI
- Open Access
- 18 July 2025
- Published by Elsevier in Seizure: European Journal of Epilepsy
- Vol. 131, 219-225
Impact of seizure detectors on treatment and costs of epilepsy
- 18 July 2025
- Published by Elsevier in Seizure: European Journal of Epilepsy
- Vol. 131, 219-225
Open AccessImpact of seizure detectors on treatment and costs of epilepsy
- Open Access
Impact of Seizure Detectors on Treatment and Costs of Epilepsy
Introduction: Epilepsy management relies heavily on the accurate detection of seizures. Unreliable seizure readouts can lead to missed seizures and erroneous physician visits due to false positives. These inaccuracies can also result in delayed treatment or inappropriate medication adjustments, ultimately increasing healthcare costs. Here, we introduce a model to simulate the effects of an imperfect seizure readout on the (perceived) efficacy of treatment, the number of physician visits and associated health-care costs. Methods: To assess the effects of the reliability of seizure detection systems on treatment, we introduce a simple model that generates seizures with a probability P , which can be altered by medication. In addition, we simulate different degrees of reliability of seizure detectors, where changes in medication are based on the perceived seizure frequency . We quantify the resulting costs in terms of remaining true seizures, physician visits, and the number of medication changes as a function of detector reliability. Results: We show that the degree of unreliability in seizure readouts significantly affects overall healthcare costs. Costs increased up to three to five times more for unreliable systems. Moreover, unreliable systems may result in delayed or even inappropriate change of medication, further complicating effective treatment. Conclusion: Reliable seizure detection systems can significantly reduce health-care costs and improve epilepsy treatment. Key Findings: In a perfect detection system, costs arise primarily from true seizures, which include consultations with a neurologist and costs associated with hospital admissions or loss of productivity. As the unreliability of detectors increases, costs from missed seizures become more prominent, leading to untreated epilepsy episodes and limiting optimal patient care. Erroneous visits, resulting from unreliable seizure detection systems, may also contribute to suboptimal treatment and increase healthcare costs.
Open AccessImpact of Seizure Detectors on Treatment and Costs of Epilepsy
Introduction: Epilepsy management relies heavily on the accurate detection of seizures. Unreliable seizure readouts can lead to missed seizures and erroneous physician visits due to false positives. These inaccuracies can also result in delayed treatment or inappropriate medication adjustments, ultimately increasing healthcare costs. Here, we introduce a model to simulate the effects of an imperfect seizure readout on the (perceived) efficacy of treatment, the number of physician visits and associated health-care costs. Methods: To assess the effects of the reliability of seizure detection systems on treatment, we introduce a simple model that generates seizures with a probability P , which can be altered by medication. In addition, we simulate different degrees of reliability of seizure detectors, where changes in medication are based on the perceived seizure frequency . We quantify the resulting costs in terms of remaining true seizures, physician visits, and the number of medication changes as a function of detector reliability. Results: We show that the degree of unreliability in seizure readouts significantly affects overall healthcare costs. Costs increased up to three to five times more for unreliable systems. Moreover, unreliable systems may result in delayed or even inappropriate change of medication, further complicating effective treatment. Conclusion: Reliable seizure detection systems can significantly reduce health-care costs and improve epilepsy treatment. Key Findings: In a perfect detection system, costs arise primarily from true seizures, which include consultations with a neurologist and costs associated with hospital admissions or loss of productivity. As the unreliability of detectors increases, costs from missed seizures become more prominent, leading to untreated epilepsy episodes and limiting optimal patient care. Erroneous visits, resulting from unreliable seizure detection systems, may also contribute to suboptimal treatment and increase healthcare costs.
- 5 November 2024
- Published by Springer Nature
- p. 526-545
IterSHAP: An XAI-Based Feature Selection Method for Small High-Dimensional Datasets
- 5 November 2024
- Published by Springer Nature
- p. 526-545
IterSHAP: An XAI-Based Feature Selection Method for Small High-Dimensional Datasets