Dataset Information

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

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.

SUBMITTER: Hancock PJB

PROVIDER: S-EPMC7657890 | biostudies-literature | 2020 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

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

Hancock Peter J B PJB Somai Rosyl S RS Mileva Viktoria R VR

Royal Society open science 20201007 10

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 appea ...[more]

PMID: 33204449

Similar Datasets

Project description:Cassava roots are complex structures comprising several distinct types of root. The number and size of the storage roots are two potential phenotypic traits reflecting crop yield and quality. Counting and measuring the size of cassava storage roots are usually done manually, or semi-automatically by first segmenting cassava root images. However, occlusion of both storage and fibrous roots makes the process both time-consuming and error-prone. While Convolutional Neural Nets have shown performance above the state-of-the-art in many image processing and analysis tasks, there are currently a limited number of Convolutional Neural Net-based methods for counting plant features. This is due to the limited availability of data, annotated by expert plant biologists, which represents all possible measurement outcomes. Existing works in this area either learn a direct image-to-count regressor model by regressing to a count value, or perform a count after segmenting the image. We, however, address the problem using a direct image-to-count prediction model. This is made possible by generating synthetic images, using a conditional Generative Adversarial Network (GAN), to provide training data for missing classes. We automatically form cassava storage root masks for any missing classes using existing ground-truth masks, and input them as a condition to our GAN model to generate synthetic root images. We combine the resulting synthetic images with real images to learn a direct image-to-count prediction model capable of counting the number of storage roots in real cassava images taken from a low cost aeroponic growth system. These models are used to develop a system that counts cassava storage roots in real images. Our system first predicts age group ('young' and 'old' roots; pertinent to our image capture regime) in a given image, and then, based on this prediction, selects an appropriate model to predict the number of storage roots. We achieve 91% accuracy on predicting ages of storage roots, and 86% and 71% overall percentage agreement on counting 'old' and 'young' storage roots respectively. Thus we are able to demonstrate that synthetically generated cassava root images can be used to supplement missing root classes, turning the counting problem into a direct image-to-count prediction task.

Project description:ImportanceDetection of cutaneous cancer on the face using deep-learning algorithms has been challenging because various anatomic structures create curves and shades that confuse the algorithm and can potentially lead to false-positive results.ObjectiveTo evaluate whether an algorithm can automatically locate suspected areas and predict the probability of a lesion being malignant.Design, setting, and participantsRegion-based convolutional neural network technology was used to create 924 538 possible lesions by extracting nodular benign lesions from 182 348 clinical photographs. After manually or automatically annotating these possible lesions based on image findings, convolutional neural networks were trained with 1 106 886 image crops to locate and diagnose cancer. Validation data sets (2844 images from 673 patients; mean [SD] age, 58.2 [19.9] years; 308 men [45.8%]; 185 patients with malignant tumors, 305 with benign tumors, and 183 free of tumor) were obtained from 3 hospitals between January 1, 2010, and September 30, 2018.Main outcomes and measuresThe area under the receiver operating characteristic curve, F1 score (mean of precision and recall; range, 0.000-1.000), and Youden index score (sensitivity + specificity -1; 0%-100%) were used to compare the performance of the algorithm with that of the participants.ResultsThe algorithm analyzed a mean (SD) of 4.2 (2.4) photographs per patient and reported the malignancy score according to the highest malignancy output. The area under the receiver operating characteristic curve for the validation data set (673 patients) was 0.910. At a high-sensitivity cutoff threshold, the sensitivity and specificity of the model with the 673 patients were 76.8% and 90.6%, respectively. With the test partition (325 images; 80 patients), the performance of the algorithm was compared with the performance of 13 board-certified dermatologists, 34 dermatology residents, 20 nondermatologic physicians, and 52 members of the general public with no medical background. When the disease screening performance was evaluated at high sensitivity areas using the F1 score and Youden index score, the algorithm showed a higher F1 score (0.831 vs 0.653 [0.126], P < .001) and Youden index score (0.675 vs 0.417 [0.124], P < .001) than that of nondermatologic physicians. The accuracy of the algorithm was comparable with that of dermatologists (F1 score, 0.831 vs 0.835 [0.040]; Youden index score, 0.675 vs 0.671 [0.100]).Conclusions and relevanceThe results of the study suggest that the algorithm could localize and diagnose skin cancer without preselection of suspicious lesions by dermatologists.

Dataset Information

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

Publications

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

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets