Convolutional neural net face recognition works in non-human-like ways

Peter J B Hancock,Viktoria R Mileva,Rosyl S Somai

doi:10.1098/rsos.200595

Peter J B Hancock, Viktoria R Mileva + Show 1 more

Open Access

https://doi.org/10.1098/rsos.200595

Copy DOI

Journal: Royal Society open science	Publication Date: Oct 1, 2020
Citations: 11	License type: CC BY 4.0

Affiliation: University of Stirling

Abstract

Convolutional neural networks (CNNs) give the state-of-the-art performance in many pattern recognition problems but can be fooled by carefully crafted patterns of noise. We report that CNN face recognition systems also make surprising ‘errors'. We tested six commercial face recognition CNNs and found that they outperform typical human participants on standard face-matching tasks. However, they also declare matches that humans would not, where one image from the pair has been transformed to appear a different sex or race. This is not due to poor performance; the best CNNs perform almost perfectly on the human face-matching tasks, but also declare the most matches for faces of a different apparent race or sex. Although differing on the salience of sex and race, humans and computer systems are not working in completely different ways. They tend to find the same pairs of images difficult, suggesting some agreement about the underlying similarity space.

Highlights

Convolutional neural networks (CNNs) have transformed pattern recognition, achieving the state-of-the-art performance in many applications, including automated face recognition (AFR) [1]
The current controversies around the public use of AFR and lack of clear legislation have resulted in a ban on its use in some places
CNNs 1 and 4 show near-perfect performance, while others would improve with a different decision threshold—the decision boundary between match and mismatch

Summary

Introduction

Convolutional neural networks (CNNs) have transformed pattern recognition, achieving the state-of-the-art performance in many applications, including automated face recognition (AFR) [1]. They can be deceived by noise patterns, either on their own or added to another image [2]. The current controversies around the public use of AFR and lack of clear legislation have resulted in a ban on its use in some places. What can you say about the people depicted in figure 1? The second image figure 1b is a composite made of several male actors. This work originated with the observation that a state-of-the-art CNN face (a)

Objectives

Methods

Results

Conclusion