A federated learning system for data protection-compliant data exchange and collaboration

Abstract

A federated learning system (100) for data protection in data sharing and collaboration, consisting of: a module for data acquisition and local preprocessing that is configured to clean, normalize and standardize local data sets at each participating node without transferring raw data externally; a local model training module configured to train a machine learning model on the pre-processed local dataset; a secure model update and encryption module configured to encrypt and secure model parameters or updates before transmission using privacy protection techniques; a federated aggregation and coordination module configured to aggregate encrypted updates from multiple participating nodes into a global model; a module for monitoring and ensuring data protection compliance, configured to enforce data protection budgets and audit protocols and to ensure compliance with data protection regulations; a performance optimization and resource management module configured to optimize communication, computation, and resource utilization across all nodes; and a module for global model delivery and feedback, configured to redistribute the aggregated global model to participants and integrate performance feedback for iterative improvements.

Description

[0001] The present invention relates to the field of federated learning and privacy-compliant data collaboration systems. More specifically, it relates to methods and architectures that enable multiple units to jointly train machine learning models without directly exchanging sensitive or raw data.

[0002] In modern digital ecosystems, organizations and institutions generate vast amounts of data, which holds immense potential for training advanced machine learning models. However, the direct sharing of such data between different entities is severely restricted due to concerns about data privacy, security, ownership rights, and regulatory compliance. Traditional centralized machine learning approaches require the aggregation of raw data in a single repository, which not only increases the risk of data breaches but also makes organizations hesitant to participate in collaborative AI initiatives. This limitation prevents the full realization of the collective value of distributed datasets.

[0003] Existing solutions such as anonymization, encryption, or secure data transmission mechanisms can only partially mitigate these challenges. Anonymization techniques often fail against sophisticated re-identification attacks, while secure transmission still requires sensitive data to be transferred across networks, introducing vulnerabilities. Furthermore, many companies are subject to strict data privacy laws such as the GDPR or HIPAA, which restrict direct data traffic or disclosure. As a result, industries such as healthcare, finance, and smart manufacturing struggle to collaborate effectively on AI-driven innovations without compromising confidentiality.

[0004] Federated learning has emerged as a promising paradigm for addressing these problems, as it enables collaborative model training without requiring direct data exchange. However, current federated systems face challenges such as communication overhead, model convergence, heterogeneous data distribution, and ensuring robust data protection during model updates. Therefore, there is a significant need for an improved federated learning system that not only enables data protection-compliant collaboration but also ensures efficiency, scalability, and trust among participants, while adhering to stringent regulatory frameworks.

[0005] One objective of the present disclosure is to provide a federated learning system that enables collaborative model training without requiring direct data sharing, thereby preserving privacy.

[0006] Another objective of the present disclosure is to offer secure mechanisms for encrypting and transmitting model updates in order to prevent the leakage of sensitive information.

[0007] Another objective of this disclosure is to ensure compliance with global and regional data protection regulations such as the GDPR and HIPAA during federated operations.

[0008] Another objective of this disclosure is to address the challenges of heterogeneous and unbalanced data across multiple organizations while maintaining model accuracy.

[0009] Another objective of this disclosure is to reduce communication overhead and optimize resource allocation in order to improve the scalability of the federated learning process.

[0010] Another objective of this disclosure is to create trust and transparency among the participants through audits and compliance monitoring.

[0011] Another objective of the present disclosure is to facilitate the real-time provision of aggregated global models to participants with version control.

[0012] Another objective of the present disclosure is to enable continuous improvement of global models through the integration of feedback and iterative training cycles.

[0013] The present invention relates to a federated learning system for collaboration while maintaining privacy, which enables multiple organizations to jointly train machine learning models without disclosing raw data, thereby improving security and compliance.

[0014] Another embodiment of the present invention is the module for data acquisition and local preprocessing, which standardizes heterogeneous data sets at each participant node and ensures that only refined and usable data are processed locally without transmission.

[0015] Another embodiment of the present invention is the local model training module, which enables each organization to train models independently on its local data in order to gain unique insights while ensuring that the raw data remains protected.

[0016] Another embodiment of the present invention is the module for secure model updating and encryption, which uses cryptographic and data protection methods to secure model updates and prevent the leakage of sensitive patterns.

[0017] Another embodiment of the present invention is the federated aggregation and coordination module, which combines encrypted updates using algorithms such as federated averaging into a unified global model to achieve collaborative intelligence.

[0018] Another embodiment of the present invention is the module for monitoring and complying with data protection, which ensures compliance with legal, regulatory and ethical requirements by monitoring data protection budgets and maintaining audit trails.

[0019] Another embodiment of the present invention is the performance optimization and resource management module, which reduces communication overhead, balances resource usage, and ensures scalability in different environments.

[0020] Another embodiment of the present invention is the global model provision and feedback module, which returns the final global model to the participants, integrates feedback, and supports continuous improvement through iterative training cycles.

[0021] The present invention relates to a federated learning system (100) for collaborative, privacy-preserving learning that enables organizations to jointly train models without disclosing raw data. The system comprises a data ingestion and local preprocessing module for preparing heterogeneous datasets and a local model training module for independently training models at each node. A secure model update and encryption module ensures that model updates are protected before transmission, while a federated aggregation and coordination module merges them into a unified global model. Furthermore, a data privacy monitoring and compliance module ensures adherence to legal regulations, and a performance optimization and resource management module improves scalability and efficiency.Finally, a global model delivery and feedback module distributes the global model back to the participants and integrates feedback for continuous improvement.

[0022] The modules of the federated learning system (100) for collaboration while respecting privacy are as follows: Module for data acquisition and local preprocessing

[0023] This module enables each participating institution, such as a hospital, bank, or company, to preprocess and prepare its local datasets for training without moving them off-site. The module includes tools for data cleansing, normalization, feature extraction, and format standardization, ensuring that heterogeneous datasets from all participants can be effectively used for federated learning. By preprocessing on the local node, the system ensures that sensitive raw data never leaves the owner's infrastructure, thus maintaining confidentiality. Module for local model training

[0024] Once the data is preprocessed, this module trains a machine learning model locally on each participating node using the prepared dataset. It supports multiple machine learning and deep learning frameworks, providing flexibility for diverse applications. The local training process updates model weights or parameters based on institution-specific data patterns while preventing raw data from being shared with external systems. This ensures that valuable insights are captured at the source while maintaining strict data privacy boundaries. Module for secure model updates and encryption

[0025] To share model progress with the central system, this module encrypts locally trained model updates using advanced cryptographic techniques such as homomorphic encryption, secure multi-party computation, and differential privacy. This encryption ensures that model updates do not reveal sensitive information that could indirectly expose properties of the underlying dataset. By securing communication channels and employing privacy protection mechanisms, this module guarantees that the data of individual participants remains protected even during model aggregation. Module for federated aggregation and coordination

[0026] The core of the system is the aggregation module, which receives encrypted or masked model updates from multiple participants. It combines these updates into a global model using algorithms such as federated averaging (FedAvg) or weighted aggregation methods. The module also addresses challenges such as non-independent and identically distributed (non-IID) data, unbalanced datasets, and communication efficiency. By coordinating training rounds and synchronizing updates between participants, this module ensures that collaborative learning results in a high-performing, shared model. Module for data protection monitoring and compliance

[0027] This module continuously monitors all processes to ensure compliance with data protection regulations such as GDPR, HIPAA, or regional data protection laws. It enforces predefined data protection budgets, manages parameters for differentiated data protection, and reviews communication protocols for transparency. By ensuring compliance, the module builds trust among participants and guarantees that federated learning processes are aligned with ethical and legal frameworks, thereby promoting cross-institutional collaboration without legal risks. Module for performance optimization and resource management

[0028] To ensure scalability and efficiency, this module optimizes communication bandwidth, computational resources, and model convergence rates. It supports adaptive client selection, dynamic adjustments to the learning rate, and parallel training mechanisms. By reducing unnecessary communication overhead and distributing the computational load across heterogeneous devices, this module ensures that the federated learning process remains cost-effective and practical even in large environments with limited resources. Module for global model deployment and feedback

[0029] Once the aggregated global model has reached a satisfactory performance level, this module enables its deployment to participating organizations or external users. It supports versioning, continuous updates, and model validation on unknown local datasets. The module also gathers performance feedback from participants, which is integrated into future training cycles, thus enabling continuous improvement of the federated model. This ensures that the final model is not only accurate but also adaptable to changing data distributions and real-world use cases.

[0030] The invention is explained again below with reference to the figure. This shows: Fig. : a federated learning system (100) for data protection in data sharing and collaboration.

[0031] Fig.Figure 100 illustrates a federated learning system for privacy-compliant data exchange and collaboration. The system's operation begins with each participating organization feeding its local datasets into the data collection and local preprocessing module, where the data is cleaned, normalized, and standardized without leaving the institution's infrastructure. The refined data is then forwarded to the local model training module, which independently builds and updates machine learning models at each node to ensure insights are gained without exposing raw data. The resulting model parameters are secured using the Secure Model Update and Encryption module, which employs cryptographic techniques and privacy-protecting methods before updates are transmitted to the central system.These encrypted updates are collected by the Federated Aggregation and Coordination module, where they are combined into a unified global model using algorithms such as Federated Averaging, while addressing challenges like unbalanced and heterogeneous data. Throughout this process, the Data Privacy Monitoring and Compliance module ensures adherence to legal standards by managing data privacy budgets and audit activities to guarantee lawful and ethical collaboration. To improve scalability and efficiency, the Performance Optimization and Resource Management module dynamically manages communication, computation, and resource allocation, thereby reducing overhead and accelerating convergence.Finally, the trained global model is distributed via the global model deployment and feedback module, enabling participating companies to use the improved model in real-world applications while simultaneously providing feedback on performance results to guide future training cycles. This seamless orchestration of modules ensures privacy-compliant, efficient, and regulatory-compliant federated learning across multiple organizations.

Claims

[1] A federated learning system (100) for data protection in data sharing and collaboration, consisting of: a module for data acquisition and local preprocessing that is configured to clean, normalize and standardize local data sets at each participating node without transferring raw data externally; a local model training module configured to train a machine learning model on the pre-processed local dataset; a secure model update and encryption module configured to encrypt and secure model parameters or updates before transmission using privacy protection techniques; a federated aggregation and coordination module configured to aggregate encrypted updates from multiple participating nodes into a global model; a module for monitoring and ensuring data protection compliance, configured to enforce data protection budgets and audit protocols and to ensure compliance with data protection regulations; a performance optimization and resource management module configured to optimize communication, computation, and resource utilization across all nodes; and a module for global model delivery and feedback, configured to redistribute the aggregated global model to participants and integrate performance feedback for iterative improvements. [2] System (100) according to claim 1, wherein the module for safe model updating and -encryption uses cryptographic methods selected from the group consisting of homomorphic encryption, secure multi-party computation, and differential privacy. [3] System (100) according to claim 1, wherein the module for federated aggregation and coordination uses federated averaging (FedAvg) or weighted aggregation algorithms to generate the global model. [4] System (100) according to claim 1, wherein the data protection monitoring and compliance module is configured to automatically enforce regional regulations or other jurisdiction-specific data protection frameworks. [5] System (100) according to claim 1, wherein the performance optimization and resource management module dynamically selects participant nodes based on computing capacity, bandwidth availability or data quality. [6] System (100) according to claim 1, wherein the module for providing and providing feedback of global models supports version control and continuous integration of updated global models across all participants. [7] System (100) according to claim 1, wherein the module for data acquisition and local preprocessing performs feature extraction, dimensionality reduction and anomaly filtering prior to training. [8] System (100) according to claim 1, wherein the local model training module supports heterogeneous machine learning frameworks, including neural networks, decision trees or regression models. [9] System (100) according to claim 1, wherein the module for monitoring and compliance with data protection regulations provides audit reports and compliance dashboards for verifying participants and building trust. [10] System (100) according to claim 1, wherein the performance optimization and resource management module reduces the communication overhead by applying gradient compression or sparsification update techniques.

Images