1Q25 Federated Computing Update

Federated Computing (FC) is emerging as the new paradigm for cross-organization data collaborations. As FC’s ascendance continues, we want to help the Rhino community keep up to date on developments they may find relevant. Each quarter, we’ll review the most important news in Federated Learning (FL) frameworks and FL-based products, applications of Federated Computing, and news from important adjacent technologies such as edge computing, confidential computing, and more. 

Interested in solving one of the biggest challenges in AI and data by seamlessly connecting siloed data? Reach out to schedule a demo of the Rhino Federated Computing Platform (Rhino FCP) today!

Major FL Framework and Federated Computing Product Updates

• Rhino Federated Computing: Most personal - we rebranded from Rhino Health to Rhino Federated Computing this quarter, recognizing sustained demand for Federated Computing from diverse industries. We continue to partner closely with our original healthcare & life sciences customers, but look forward to further evangelizing Federated Computing to an even broader audience. Read more about why HERE.
• NVIDIA FLARE:  At GTC 2025, Rhino partner NVIDIA announced several exciting additions planned for release NV FLARE due out later this year, including native support for confidential computing and cross-device Federated Learning. By encrypting data during processing, confidential computing offers a critical enhancement to data security strategies, particularly in sensitive or regulated environments. Cross-device federated learning unlocks new applications for Rhino's enterprise users, including real-time personalization on edge devices, decentralized predictive maintenance, and privacy-preserving health monitoring across diverse endpoints.
• Rhino + Flower: Rhino also announced a partnership with Flower Labs, taking advantage of NVFLARE 2.5’s integration with the Flower framework to offer Flower users the ability to run their projects using Flower’s framework via Rhino’s leading enterprise-ready platform, allowing these Flower-based experiments to be securely deployed in production. Find an example HERE. 
• OpenFL: The LF & AI Data Foundation released OpenFL 1.7, which  introduced a FederatedRuntime for seamless transition from local simulation to distributed training via Jupyter notebooks. It also added features like federated evaluation support and XGBoost integration.
• Google: Google introduced Parfait, an open-source GitHub organization with several tools & frameworks relevant to federated learning and analytics. The suite is being used across Google products such as Gboard and Google Maps. Google Research also announced Confidential Federated Analytics, a novel technique that leverages confidential computing to perform predefined analyses on user data directly on their devices. 

Applications of Federated Computing

1Q25 saw a large number of publications across several areas - including healthcare & life sciences, cyber security, mobile devices, and even space exploration. 

Mobile Devices

• Apple: Apple published “Private Federated Learning In Real World Application – A Case Study”. This case study demonstrates the implementation of Private Federated Learning (PFL) on edge devices, ensuring user data remains on-device with only essential model updates transmitted securely to a central server. PFL enhances privacy while improving model accuracy by adapting to user behavior over time in a device-first approach. 

Life Sciences

• Nature Machine Intelligence published “Data-driven federated learning in drug discovery with knowledge distillation” The paper introduces FLuID, a federated distillation framework that prioritizes data-centric information distillation over federated averaging, enhancing institutional collaboration, data privacy, and model performance in decentralized drug discovery.

Cyber Security

Scientific Reports published several articles using FL for cyber security applications, including intrusion detection and prediction of backdoor attacks:

• “Intelligent deep federated learning model for enhancing security in internet of things enabled edge computing environment” in which researchers introduced the Intelligent Deep Federated Learning Model for Enhancing Security (IDFLM-ES), designed to bolster intrusion detection in IoT-enabled edge computing environments. By integrating a federated hybrid deep belief network with advanced optimization techniques, the model achieved a notable accuracy of 98.24% on benchmark datasets.
• “An optimal federated learning-based intrusion detection for IoT environment,” in which researchers used an FL-based intrusion detection system (IDS) for IoT networks, using the Chimp optimization algorithm for optimal feature selection. This approach achieved a detection accuracy of 95.59%.
• “Practical Implementation of Federated Learning for Detecting Backdoor Attacks in a Next-word Prediction Model”, which presents a next-word prediction model using federated learning, incorporating a detection mechanism for backdoor attacks. The study demonstrates that this mechanism effectively reduces model bias introduced by compromised devices, especially when their proportion is low.
• “Advanced artificial intelligence with federated learning framework for privacy-preserving cyberthreat detection in IoT-assisted sustainable smart cities” introduces the Advanced Artificial Intelligence with Federated Learning Framework for Privacy-Preserving Cyberthreat Detection (AAIFLF-PPCD) in IoT-assisted smart cities. This framework combines Harris Hawk optimization for feature selection, a stacked sparse auto-encoder for threat detection, and the walrus optimization algorithm for hyperparameter tuning.  

Healthcare

• npj Digital Medicine published “Real world federated learning with a knowledge distilled transformer for cardiac CT imaging”, which reports on the largest federated cardiac CT analysis to date, involving 8,104 scans from eight hospitals. Researchers developed a two-step, semi-supervised strategy that distills knowledge from task-specific convolutional neural networks into a transformer model. This approach improved predictive accuracy and enabled simultaneous learning from partially labeled datasets across institutions. The transformer model outperformed UNet-based models in generalizability on downstream tasks.
• Communication Engineering published “Distributed training of foundation models for ophthalmic diagnosis”, which introduces a distributed deep learning framework that combines self-supervised learning with domain-adaptive federated learning to improve the detection of eye diseases from optical coherence tomography images. The study reports that this approach enhanced the area under the curve by at least 10% compared to local models. 
• Frontiers in Computer Science published “A reliable and privacy-preserved federated learning framework for real-time smoking prediction in healthcare”. This study is another proofpoint of the applicability of FL to public health research, allowing healthcare leaders in provider settings or public health agencies to deeply understand trends in population health in order to better target their strategies. 
• Scientific Reports published “A privacy-preserving dependable deep federated learning model for identifying new infections from genome sequences”. The authors introduce a privacy-preserving and dependable deep federated learning model aimed at identifying new infections in healthcare settings. The model utilizes federated learning to enable collaborative training across multiple institutions without compromising patient data privacy. In experiments, the model achieved an accuracy of 98.5% in detecting emerging infections.
• Scientific Reports also published “Applying YOLOv6 as an ensemble federated learning framework to classify breast cancer pathology images”, which presents an ensemble federated learning framework utilizing a pruned YOLOv6 model for classifying breast cancer pathology images. By combining federated learning with homomorphic encryption, the approach enables collaborative training across multiple institutions while preserving patient data privacy. The model achieved validation accuracies of 98% on the BreakHis dataset and 97% on the BUSI dataset, outperforming traditional centralized models like VGG-19, ResNet-50, and InceptionV3.

Space Exploration

• Researchers from NASA published “Bridging Earth and Space: A Flexible and Resilient Federated Learning Framework Deployed on the International Space Station” to bioRxiv. This preprint preprint introduces a flexible and resilient federated learning framework designed for secure transmission of model updates between Earth and the International Space Station. This approach enables collaborative model training across terrestrial and space-based systems while preserving data privacy of astronauts and preserving precious bandwidth. 

‍