arXiv:2605.27892v1 Announce Type: new 
Abstract: Synthetic Electronic Health Record (EHR) generation provides a promising avenue for data augmentation and cross-hospital modeling in privacy-constrained healthcare settings. However, most existing EHR generative models are centralized and require pooling data across hospitals, which is often infeasible when real-world data sharing is restricted. While federated EHR generation offers a natural solution, direct federated modeling often collapses or diverges due to the high dimensionality, sparsity, and cross-hospital heterogeneity of EHR data. In this work, we propose FedEHR-Gen, the first federated framework for synthetic time-series EHR generation across distributed hospitals. FedEHR-Gen uses a two-stage learning paradigm. First, we introduce a federated autoencoder that projects high-dimensional and sparse EHR features onto a compact latent space. To ensure semantic consistency across hospitals, we develop a layer-wise matching aggregation mechanism that aligns local encoders into a unified global latent space. Second, operating on this aligned latent space, we train a federated temporal conditional variational autoencoder (TCVAE) with distribution-aware aggregation, enabling stable temporal generative modeling under severe cross-hospital heterogeneity. Extensive experiments on the eICU and MIMIC-III datasets demonstrate that FedEHR-Gen achieves generation fidelity, downstream utility, and privacy risk comparable to centralized training, while consistently outperforming the standard federated baseline.

تم تقديم FedEHR-Gen كإطار فدرالي رائد لتوليد سجلات الصحة الإلكترونية (EHR) الاصطناعية عبر المستشفيات الموزعة، مما يعالج التحديات المتعلقة بالأبعاد العالية وخصوصية البيانات في بيئات الرعاية الصحية. تستخدم هذه الطريقة المبتكرة نموذج تعلم من مرحلتين، يتضمن مشفرًا تلقائيًا فدراليًا لتمثيل البيانات بشكل فعال.

FedEHR-Gen se ha presentado como un marco federado pionero para la generación de Registros Electrónicos de Salud (EHR) sintéticos en series temporales a través de hospitales distribuidos, abordando los desafíos de alta dimensionalidad y privacidad de datos en entornos de atención médica. Este enfoque innovador utiliza un paradigma de aprendizaje en dos etapas, incorporando un autoencoder federado para una representación eficaz de los datos.

FedEHR-Gen a été introduit comme un cadre fédéré novateur pour la génération de dossiers de santé électroniques (DSE) synthétiques en série temporelle à travers des hôpitaux distribués, répondant aux défis de la haute dimensionnalité et de la confidentialité des données dans les soins de santé. Cette approche innovante utilise un paradigme d'apprentissage en deux étapes, incorporant un autoencodeur fédéré pour une représentation efficace des données.

FedEHR-Gen has been introduced as a pioneering federated framework for generating synthetic time-series Electronic Health Records (EHRs) across distributed hospitals, addressing the challenges of high dimensionality and data privacy in healthcare settings. This innovative approach utilizes a two-stage learning paradigm, incorporating a federated autoencoder for effective data representation.

FedEHR-Gen: Federated Synthetic Time-Series EHR Generation via Latent Space Alignment and Distribution-Aware Aggregation

arXiv:2606.07141v1 Announce Type: new 
Abstract: Language models trained for clinical disease inference are trained on patient data, which may include sensitive and private information, and data owners may request the removal of their data from a trained model due to privacy or copyright concerns. However, exactly unlearning patient-specific data is intractable, and retraining with minor data removal is resource-intensive. While there exists several machine unlearning methods that can be used, their utility is generally restricted to non-medical domains. Moreover, the existing benchmarks for evaluating such unlearning methods primarily utilize synthetically curated datasets, which are not truly representative of real-world systems. Hence, the effectiveness of these unlearning methods in the medical domain is largely unclear. To this end, we introduce REMEDI, an extensive benchmark for machine unlearning tailored to multi-label and multiclass clinical disease inference, where label correlations, longitudinal structure, and safety constraints make unlearning particularly challenging. Unlike the existing benchmarks, REMEDI considers: (1) a relevant application domain (medical), (2) comprehensive unlearning setups involving diverse sets of forget instances, (3) challenging unlearning scenarios including multi-label and multi-class classification tasks, and (4) evaluation metrics involving performance both in terms of utility and extent of unlearning achieved. REMEDI is developed using the MIMIC-III clinical database that contains comprehensive clinical data of patients. Experiments with existing unlearning methods indicate that there exists a trade-off between utility and unlearning performance. They are also largely unsuited to multi-label classification tasks. To facilitate reproducibility, we make our benchmark publicly available.

تم تقديم REMEDI كمعيار لتقييم طرق التعلم غير المرغوب فيه في استنتاج الأمراض السريرية متعددة التسميات، حيث يتناول التحديات المتعلقة بإزالة بيانات المرضى المحددة من نماذج اللغة المدربة على معلومات طبية حساسة. يهدف هذا المعيار إلى تقديم تقييم أكثر واقعية لتقنيات التعلم غير المرغوب فيه في المجال الطبي، الذي لم يتم استكشافه إلى حد كبير.

REMEDI se ha introducido como un estándar para evaluar los métodos de des-aprendizaje en la inferencia clínica multi-etiqueta, abordando los desafíos de eliminar datos específicos de pacientes de modelos de lenguaje entrenados con información médica sensible. Este estándar busca proporcionar una evaluación más realista de las técnicas de des-aprendizaje en el ámbito médico, que ha sido en gran medida inexplorado.

REMEDI a été introduit comme une référence pour évaluer les méthodes de désapprentissage dans l'inférence clinique multi-label, abordant les défis liés à la suppression des données spécifiques aux patients des modèles de langage formés sur des informations médicales sensibles. Cette référence vise à fournir une évaluation plus réaliste des techniques de désapprentissage dans le domaine médical, qui a été largement inexploré.

REMEDI has been introduced as a benchmark for evaluating machine unlearning methods in multi-label clinical disease inference, addressing the challenges of unlearning patient-specific data from language models trained on sensitive medical information. This benchmark aims to provide a more realistic assessment of unlearning techniques in the medical domain, which has been largely unexplored.

FedEHR-Gen: Federated Synthetic Time-Series EHR Generation via Latent Space Alignment and Distribution-Aware Aggregation

Was this article worth reading? Share it

Ready to build your own newsroom?