Ahmed Ullah Qureshi

Title: Ahmed Ullah Qureshi and His Patent for Protecting Machine Learning Models.

Introduction

Ahmed Ullah Qureshi is an inventor associated with Los Altos, California. His work includes a patented method designed to make trained machine learning models more resistant to extraction attacks.

Latest Patents

Qureshi is credited with 1 patent: “Method for Hardening a Machine Learning Model Against Extraction.” The method assesses whether a query submitted to a model may be an attempted attack. If it identifies an attack, it rearranges the confidence levels associated with prediction results, then outputs a prediction with the highest confidence level assigned through that rearrangement. This approach is intended to make it more difficult for an attacker to extract the model by analyzing its responses.

Career Highlights

Qureshi is associated with NXP B.V. His patented work addresses a security challenge in machine learning: limiting the information an attacker can infer from a model’s predictions and confidence levels.

Collaborations

Marc Joye is identified as Qureshi’s coworker. Their shared professional association connects them through work at NXP B.V.

Conclusion

Ahmed Ullah Qureshi’s patent presents a technique for strengthening machine learning model security by altering prediction confidence information when a query appears to be an extraction attempt.