Patents

US 12748945 B2: Systems and methods for a full-stack obfuscation framework to mitigate neural network architecture theft

Granted
2026-09-29
Filed
2023-06-09 application 18208027
Type
Utility · 16 claims
Elsewhere
Google Patents ↗ · USPTO patent document ↗ · USPTO Patent Center record ↗

Abstract

A full-stack neural network obfuscation framework obfuscates a neural network architecture while preserving its functionality with very limited performance overhead. The framework includes obfuscating parameters or “knobs”, including layer branching, layer widening, selective fusion and schedule pruning, that increase the number of operators, reduce/increase the latency, and number of cache and DRAM accesses. In addition, a genetic algorithm-based approach is adopted to orchestrate the combination of obfuscating knobs to achieve the best obfuscating effect on the layer sequence and dimension parameters so that the architecture information cannot be successfully extracted.

Source: PTBLXML, file of 2026-09-29 (ipgb20260929.xml).

Assignees

AssigneeTypePlaceAlso on furp.orgSource
Arizona Board of Regents on Behalf of Arizona State University 1,571 patentsUS company or organizationTempe, AZPTBLXML, file of 2026-09-29

Inventors

InventorPlacePatentsPerson pageSource
Jingtao LiTempe, AZPTBLXML, file of 2026-09-29
Chaitali ChakrabartiTempe, AZ3PTBLXML, file of 2026-09-29
Deliang FanTempe, AZPTBLXML, file of 2026-09-29
Adnan Siraj RakinTempe, AZPTBLXML, file of 2026-09-29

CPC categories

SubclassCategoryFirst groupSource
G06NComputing arrangements based on specific computational modelsG06N3/04PTBLXML, file of 2026-09-29

Cooperative Patent Classification, as currently assigned (one row per subclass).

Assignments and security interests

No recorded assignments or security interests on file for this patent.

USPTO bulk data: PatentsView granted patents, disambiguated (PVGPATDIS) for patents, assignees, inventors, places and CPC categories; the USPTO Patent Assignment Dataset (ECORSEXC) for assignments and security interests. Patents filed in the last 20 years. No claims or full text. Links to other records are by name (and state): EXACT means the identical full name and state; FUZZY means a looser name match, with the confidence shown.