DG-FedReuse: Proxy-Gradient-Gated Cached-Update Reuse with Matched Sparse Uplink Accounting

DG-FedReuse operates at the simulator level, allowing selected clients to contribute age-decayed cached updates when a stochastic head-gradient discrepancy proxy stays below a round-dependent threshold. The design enforces a hard cache-age limit and a minimum quota of fresh clients, while fresh updates use an adaptive per-tensor Top-K numerical-field representation. This reduces the need for repeated local optimization steps and full model transmission, addressing key bottlenecks in federated learning. The paper's relevance lies in its potential to lower communication overhead and speed up federated training, especially in scenarios with limited bandwidth. Future work would need to validate the approach on standard benchmarks and non-IID data distributions.