Privacy‑Preserving Active Learning for Circular Manufacturing Supply Chains under Real‑Time Policy Constraints

Описание и применение

Внедрена система активного обучения в федеративном режиме с дифференциальной приватностью и ре‑таймовой политикой, позволяющая обучать модели по данным, разбросанным по цепочке поставок, без раскрытия сырья и соблюдая меняющиеся регуляции.

Задача

Circular manufacturing supply chains need predictive models across siloed, proprietary data, but data owners will not share raw data and regulatory policies can change during training, making labeling expensive and policy compliance difficult.

Результат

The integrated loop allows continuous, privacy‑preserving model improvement while automatically adapting to evolving regulations, preventing data leakage and ensuring compliance across the circular supply chain.

Рабочий процесс

• 1. Each client (OEM, recycler, logistics) keeps local data and runs local active learning to select uncertain samples.
• 2. Clients compute DP‑noised gradient updates (DP‑SGD) and send only these updates to the aggregator.
• 3. Aggregator performs FedAvg, applies DP noise, and projects updates to satisfy current policy constraints via a policy engine.
• 4. Policy engine emits real‑time constraints (feature bans, output limits, data‑sharing budgets, temporal updates).
• 5. Aggregator broadcasts updated global model and constraints back to clients.
• 6. Clients update local models and repeat the loop, respecting privacy budgets and policy changes.

Доказательства

The single most important lesson was that privacy budget is a resource you spend, not a checkbox.

Источник

Naive uncertainty sampling fails badly when clients are heterogeneous.

Источник