Sparse Federated Representation Learning for Circular Manufacturing Supply Chains

The article explores a novel approach to modeling circular manufacturing supply chains using sparse federated representation learning, which preserves the inherent sparsity patterns in the data and enables collaborative learning without sharing raw data.

Details

The article describes the author's journey in exploring privacy-preserving machine learning for industrial IoT systems, which led to the discovery of a new paradigm for modeling circular manufacturing supply chains. The key insight was that the inherent sparsity patterns in supply chain data could be exploited using sparse representation learning techniques, while a federated architecture could enable collaborative learning without sharing raw data. The author experimented with different neural network architectures and aggregation methods, finding that traditional federated averaging was destroying the sparse structure of the learned representations. This led to the exploration of sparse federated optimization techniques that could preserve and exploit the inherent structure of manufacturing data across distributed nodes, enabling a system with zero-trust governance guarantees.