Adnan Siraj Rakin

Title: Adnan Siraj Rakin and Neural Network Architecture Protection.

Introduction

Adnan Siraj Rakin is an inventor based in Tempe, Arizona, and is associated with Arizona State University. His work includes methods for protecting neural network designs from architecture theft while preserving their functionality.

Latest Patents

Rakin is named on 1 patent: “Systems and Methods for a Full-Stack Obfuscation Framework to Mitigate Neural Network Architecture Theft.” The invention describes a framework that obscures a neural network’s architecture while maintaining its functionality with limited performance overhead. Its obfuscation techniques, or “knobs,” include layer branching, layer widening, selective fusion, and schedule pruning. These techniques alter operator counts, latency, and cache and DRAM accesses. A genetic algorithm coordinates combinations of the techniques to obscure layer sequences and dimensions, making the architecture more difficult to extract.

Career Highlights

Rakin’s listed patent focuses on safeguarding neural network architecture through coordinated, full-stack obfuscation. The approach aims to balance stronger protection with limited impact on model performance.

Collaborations

The patent information identifies Jingtao Li and Chaitali Chakrabarti as Rakin’s coworkers. Their names are associated with the collaborative research context surrounding the invention.

Conclusion

Adnan Siraj Rakin’s patent presents a framework for making neural network architectures harder to steal while preserving their practical functionality.