Partnership with Intel®: Improving brighter AI’s High-Performance Anonymization Solution

6. December 2022

brighter AI is a member of Intel® Partner Alliance.

Driving innovation in the space of smart analytics and artificial intelligence requires enormous amounts of data. Video analytics is one of the key technologies to fuel new digital solutions and improve the customer experience in various industries. However, this often goes against privacy which is an important social topic as it’s closely linked with surveillance. As a result, data protection regulations, such as CCPA in the US, PIPL in China, and, above all, GDPR in the EU are becoming tougher and more restrictive. Companies that are not compliant will face, if not already have faced, millions of euros of fines, reputational damage, and loss in revenue. 

There are a few choices: not use data, make uninformed decisions and lose revenue, use their data without being compliant and be fined, or anonymize their data. Anonymization solutions have to be more efficient in order to satisfy the requirements of data protection and be able to handle large amounts of data in a short time. Therefore, we used our partner Intel®’s C++ Compiler and Math Kernel Library to identify and resolve performance bottlenecks. We also tested Intel Neural Compressor and Intel VTune Profiler, and would like to further explore their use cases soon.

We improved our anonymization solutions’ performance with Intel®’s technologies. The results were great. We achieved speedups of more than 60% on both of our solutions, Deep Natural Anonymization (DNAT) and Precision Blur. This consists of an improvement in processor time reduction of 37% on Precision Blur, and 40% on DNAT after compiling parts with Intel C++ Compiler and Intel Math Kernel Library. The other 20% is due to configuration improvements. 

The time and effort invested in this project by Intel® proved to be very fruitful. They involved multiple experts in the calls, took the time to understand the challenges, and suggested ways of solving optimization tasks. We appreciate the improvements achieved with their support, and look forward to more collaboration in the future.