Privacy-Preserving Active Learning for circular manufacturing supply chains for extreme data sparsity scenarios

Description and use

A prototype system combining differential privacy, active learning, and federated aggregation extracts signal from extremely sparse, private data in circular manufacturing supply chains.

Problem

Sparse, highly sensitive labeled data across a distributed network of recyclers and remanufacturers makes federated learning ineffective, as local models overfit and aggregated updates degrade performance.

Outcome

The integrated system successfully extracted signal from extremely sparse, private data, demonstrating that DP‑protected active learning can operate in federated settings where naive averaging fails.

Workflow

• Central orchestrator maintains global model and aggregates DP‑noised updates; broadcasts acquisition scores to clients; each client runs local DP‑SGD on labeled data, computes uncertainty scores on unlabeled pool; DP‑protects acquisition scores with a smaller privacy budget; clients send DP‑noised updates and scores back; orchestrator aggregates updates and updates global model; repeat until convergence or budget exhausted.

Evidence

While exploring the intersection of federated learning and circular economy systems, I stumbled upon a problem that genuinely kept me up at night.

Source

The fix isn't more data—it's smarter querying of the data you already have.

Source

acquisition scores themselves leak information. If a client reports "sample #47 is highly uncertain," that reveals something about the local data distribution.

Source