Company news & product updates

Python vs C++ clients in FEDn: Python uses less idle memory but C++ is more efficient during training with better memory management and faster task completion. FEDn supports both simultaneously.

Scaleout Systems secured 35 MSEK with Fairpoint Capital joining existing investors to enhance its FEDn framework for secure cloud-edge AI deployment. The funding will help address data processing challenges in industrial IoT, automotive and defense sectors while maintaining data sovereignty.

Federated learning represents a shift in AI architecture where models travel to and train on distributed data sources, eliminating the need to centralize sensitive data. This approach enables organizations and devices to learn collaboratively while maintaining data privacy and reducing latency for real-time edge processing.

The future of AI is in the cloud-edge continuum, allowing organizations to use both centralized machine learning and edge computing. As data increases, federated learning becomes important, letting models learn from local data while keeping privacy and optimizing resources across devices.

Scaleout's FEDAIR project for NATO's DIANA programme enables secure, decentralized ML model updates in conflict zones, allowing adaptation without compromising sensitive data.

An article on using FEDn, a framework that enables federated learning by distributing training across multiple clients. It demonstrates how to package a PyTorch MNIST model for federated training using FEDn's "compute package" system.

Federated Learning (FL) with Differential Privacy (DP) is a privacy-focused machine learning approach. This post covers two main areas: the basics of Differential Privacy, including the application of Gaussian noise for enhanced privacy, and privacy measures in Federated Learning, addressing both record-level and client-level protections.

YOLO object detection models can be integrated with federated learning to enable privacy-preserving machine learning across distributed devices. This combination allows for real-time object detection while keeping sensitive data local, making it ideal for applications like autonomous vehicles, medical imaging, and manufacturing quality control.

This post explains using Optuna to optimize the server-side learning rate of FedAdam in FEDn. It covers defining an objective function, tuning hyperparameters, and automating the process to improve model performance.

Mycronic is enhancing its semiconductor Pick and Place (PnP) machines with AI, but the effectiveness of these algorithms relies on access to large, sensitive data sets owned by their customers.

Federated Multi-task Learning (FMTL) combines federated and multi-task learning. Clients train local models and share only necessary parameters to maintain privacy. This approach handles data and system heterogeneity, ideal for scenarios like hospitals collaborating on different medical tasks while keeping data private.

Trusted Execution Environments (TEEs) provide secure hardware-based protection for code and data, ensuring confidentiality and integrity. They are used to secure ML model serving, federated learning, and attestation processes. Performance benchmarks show Intel SGX excels with smaller applications, while AMD SEV handles larger workloads better.

Our project uses the Hugging Face 'Transformers' library in FEDn to fine-tune a BERT-tiny model for spam detection on the Enron email dataset. Federated learning ensures data privacy by splitting the dataset between two clients. Training results in high accuracy (~99%) after a few rounds.

Federated self-supervised learning trains AI on autonomous vehicles using their sensor data without centralizing or labeling it. This method handles privacy, compliance, and data volume challenges by updating models directly on vehicles and training with unlabeled data.

About using JWT authentication in FEDn Studio to secure API access for clients and admins. It explains the JWT structure, authentication process, and its application in securing API endpoints, ensuring that only authenticated users can access and manage federated projects effectively.

Explaining how ArgoCD enhances FEDn Studio for managing federated projects on Kubernetes, using a custom helm chart for automation and consistency. This integration improves deployment efficiency, scalability, and oversight across multiple projects.

Learn more about backdoor attacks in federated learning, detailing experiments with the MNIST dataset that demonstrate the challenge of detecting such attacks when a minority of clients insert hidden triggers into the data. It also discusses potential mitigation strategies from recent research, emphasizing the ongoing security challenges in federated learning environments.

Scaleout's FEDn offers SaaS and self-managed deployment options for federated learning, providing flexibility, security, and scalability to meet evolving user needs

We're announcing a strategic collaboration with Flower which enables developers to run Flower projects on FEDn, providing access to enterprise-grade security, scalability, and monitoring capabilities.

Examination of the effects of label-flipping attacks on federated machine learning. Experiments show these attacks have a limited impact on the global model's accuracy compared to centralized training, due to the federated averaging process limiting malicious clients' influence.

Using FEDn, this post demonstrates cross-device federated learning with intermittently connected clients. It sets up a local dev environment where clients randomly connect, train a model briefly, then disconnect - repeating the cycle. Results show FEDn can robustly train models under intermittent conditions, bridging research to production deployments.

We're integrating federated learning to create an innovative intrusion detection system that enhances privacy and threat detection. This approach promises a secure, privacy-focused IoT, leveraging decentralized data without compromise. More details in the post and follow for updates.

Diving into the world of federated learning, this text introduces and explores Scaleout's innovative software suite, specifically its key components, FEDn and Studio, and how they collaboratively deliver a powerful solution for federated machine learning initiatives.

Federated machine learning provides a solution to stringent data privacy regulations by allowing model training without centralizing data, turning developers into smart service providers.

Scaleout Systems, tackling data security and privacy in AI, secured seed funding.

An overview of the Vinnova-sponsored project: Validating a System Development Kit for Edge Federated Learning.

The Scaleout Studio platform update includes a redesigned user interface for easier management of Federated Learning Networks, a streamlined process for adding clients, improved training session management, and better monitoring of session progress and model outcomes. The release also introduces a new REST-API, APIClient, and extensively updated documentation.

The FEDn version 0.7.0 introduces significant enhancements including the integration of Python's logging framework for client-side logging, compatibility improvements for Windows users, and substantial refactoring of the gRPC server for better scalability and efficiency. This update brings a more robust and user-friendly experience to our users, particularly in managing client-server interactions and data handling.

As the field of machine learning evolves, the need for enhanced data privacy and security has given rise to federated machine learning. This approach decentralizes the data, addressing key privacy concerns and providing innovative solutions to traditional machine learning challenges. Explore how federated machine learning works, its benefits, potential risks, and the measures taken to fortify its security and privacy in the following discussion.

Real-world federated learning software must be robust, resilient, and scalable, as well as being capable of handling large numbers of edge-clients and model sizes. Scaleout is designed to meet these demands.