Intellectual Property India Publishes Patent Application for System And Method For Hardware-Rooted, Query-Bound, Composition-Aware Differential Privacy Authorization Of Shared On-Device Machine Learning Inference With Cost- Function Integrity Binding, Timing-Invariant Atomic Debit, And Verifiable Adaptive Composition Accounting

MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641127064 A) filed by Vellore Institute Of Technology on October 25, 2026, for System And Method For Hardware-Rooted, Query-Bound, Composition-Aware Differential Privacy Authorization Of Shared On-Device Machine Learning Inference With Cost- Function Integrity Binding, Timing-Invariant Atomic Debit, And Verifiable Adaptive Composition Accounting.

Inventors include Dr. Gunasekaran G; Suresh Krishna K; Harish Y; Vishva Sa; and Dhanush T.

The application for the patent was published on October 30, 2026, under issue no. 44/2026.

Abstract: A hardware-rooted differential privacy authorization system for shared on-device machine-learning inference is disclosed. The system comprises a trusted execution environment, a cost-function integrity module, a composition accountant, hierarchical privacy-budget stores, a Privacy Authorization Transaction engine, a zero- knowledge sufficiency prover, a hardware debit barrier controller, and a model execution gate. The Privacy Authorization Transaction cryptographically binds an application identity, an inference query, a model version, a verified privacy-cost artifact, and a preceding composition state to a specific authorization context. The composition accountant maintains a tamper-evident record of privacy expenditure, while the authorization controller validates compliance with application-level, model-level, and device-level privacy budgets. The hardware debit barrier coordinates provisional budget deductions and commits them only upon successful completion within a predetermined duration, invalidating incomplete transactions. Zero-knowledge proofs establish budget sufficiency without disclosing private budget balances. An adaptive order selector supports privacy-accounting optimization, and a conservation-verified reallocation engine manages unused budget allocations. The model execution gate permits inference only upon verification of the corresponding authorization receipt, thereby strengthening accounting integrity, rollback resistance, and privacy-budget enforcement.

Disclaimer: Curated by HT Syndication.