CoCo-Fed: A Unified Framework for Memory- and Communication-Efficient Federated Learning at the Wireless Edge

2026-07-11 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Internet of Things (IoT) & Connected Devices, Cloud Computing & IT Infrastructure · Depth: Expert, medium

Summary

CoCo-Fed is a novel Compression and Combination-based Federated Learning framework designed to enhance memory and communication efficiency for large-scale neural network deployments within Open Radio Access Network (O-RAN) architectures. Addressing critical bottlenecks like high local memory footprints on resource-constrained gNBs and backhaul link saturation from high-dimensional model updates, CoCo-Fed employs a dual-stage reduction mechanism. Locally, it uses a double-dimension down-projection of gradients, allowing the optimizer to operate on low-rank structures without adding inference parameters or latency. Globally, it introduces an orthogonal subspace superposition protocol, projecting and superimposing layer-wise updates into a single consolidated matrix per gNB, drastically reducing backhaul traffic. The framework includes a rigorous theoretical foundation proving convergence, even for unsupervised learning. Extensive simulations on an Angle-of-Arrival estimation task show CoCo-Fed significantly outperforms state-of-the-art baselines, including LoRA and GaLore-based FL, in both memory and communication efficiency while maintaining robust convergence under non-IID settings.

Key takeaway

For AI Architects or ML Engineers deploying large neural networks on O-RAN infrastructure, CoCo-Fed offers a critical solution to overcome memory and communication bottlenecks. You should consider integrating its dual-stage compression to enable efficient local training on gNBs and drastically reduce backhaul traffic. This framework allows you to deploy complex models for tasks like wireless sensing without compromising convergence or incurring prohibitive latency, making advanced edge intelligence feasible.

Key insights

CoCo-Fed unifies local memory efficiency and global communication reduction in federated learning for wireless edge networks.

Principles

• Exploit gradient low-rank structures for local memory reduction.
• Orthogonal subspace superposition drastically compresses global model updates.
• Achieve FL convergence even in unsupervised edge learning scenarios.

Method

Perform double-dimension down-projection of gradients locally, adapting the optimizer to low-rank structures. Globally, project and superimpose layer-wise updates into a single consolidated matrix using orthogonal subspaces.

In practice

• Enable large-scale NN deployment on resource-constrained gNBs.
• Mitigate backhaul link saturation during FL model aggregation.
• Implement unsupervised learning for wireless sensing tasks.

Topics

• Federated Learning
• O-RAN Architecture
• Edge AI
• Gradient Compression
• Low-Rank Optimization
• Wireless Sensing

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Architect

Related on AIssential

• Entropy-Guided Tensor Compression for Multimodal Federated Learning on Edge Devices — Spectral entropy guides adaptive compression rank allocation in multimodal federated learning, improving efficiency and accuracy on heterogeneous edge devices.
• Federated Learning of Binary Neural Networks: Enabling Low-Cost Inference — Federated Learning (FL) preserves privacy by distributing training across devices, but deploying computationally intensive Deep Neural Networks (DNNs) for inference on low-powered edge devices present…
• Hierarchical Federated Learning for Networked AI: From Communication Saving to Architecture-Aware Design — HFL transforms FL into a family of distributed optimizers shaped by network architecture, not just a deeper protocol.
• Contrastive-Collapsed Loss for Flexible and Geometrically Optimal Embeddings and Faster Convergence — CoCo Loss optimizes embeddings for better class separation and faster training.
• Achieving Near-Linear Training Scalability for Pinterest’s Foundation Models — For embedding-heavy models, communication, not compute, dominates multi-node scaling, requiring direct optimization of data transfer.
• Collate: Collaborative Neural Network Learning for Latency-Critical Edge Systems — Collate enables collaborative learning of heterogeneous models for edge systems, optimizing accuracy under diverse latency constraints.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.