/* ---- Google Analytics Code Below */

Wednesday, June 24, 2020

Is Slow Neuroinformatics Private?

I have worked with companies which used machine learning image analysis to determine, for example, if a person was 'of age' in various locations.  To match with local regulatory provisions.    Could be done quite accurately, about 95%, but not perfectly.  Can that be privately done?

The Benefits of Slowness
Ruhr-Universitat Bochum
Meike Drießen
June 15, 2020

Neuroinformatics engineers at Ruhr-Universitat Bochum's Institute for Neural Computation in Germany have developed an algorithm that estimates an individual’s age and ethnic origin with greater than human-level accuracy. The team fed the algorithm several thousand photos of faces of different ages, sorted by age. The system disregarded features that varied between images, and only considered features that slowly changed over time. In calculating the age of the people in the photos, the algorithm outperformed even human assessment. The algorithm also estimated the correct ethnic origin of the subjects in the photos with greater than 99% probability, even though the images' average brightness was standardized, making skin color an insignificant marker for recognition. ... " 

No comments: