Tech5 commits to ‘consent-based’ images for facial recognition algorithm training databases
Geneva-based digital identity management firm Tech5 is committing itself to the “conscious choice” to keep investing in using images gathered with consent for its databases used for training AI algorithms.
The firms says it will continue to invest in millions of facial images given by consent rather than approaches used by other firms such as academic databases, social media network website and scraping. “As a result, some companies are taking advantage of this easy access to train their AI-based algorithms,” notes the announcement.
Tech5 told Biometric Update that its methods for gathering millions of facial images is a “trade secret”, and that those whose images are collected can withdraw consent at a later date. All images are depersonalized data and no personally identifiable information is collected or used.
The firm also told us that multiple images of each person is required for the training of their neural networks, and that the images are used expressly for that purpose. Then for products conducting authentication, verification, deduplication and so on, the algorithm does not remember the faces used in training. The firm is ensuring it gathers images from “various ethnicities and races to make sure that the final algorithm is not biased”.
“In keeping with our company’s mission of creating technology that allows citizens to retain sole control over their identities, we use consent-based, depersonalized data for the training of our neural networks,” comments Rahul Parthe, Co-founder, CTO and chairman of Tech5.
“We have made the conscious choice to continue investing in this area, obtaining the necessary rights for millions of images to train and benchmark our algorithms using these data. Access to supersized datasets will ensure that our technologies are not only highly accurate and robust, but also inclusive of people from all walks of life.”
“One of our goals as a company is to contribute to ethical use of biometrics,” comments Rob Haslam, strategic advisor of Tech5. “We believe that only implementation of ethically trained algorithms can help to develop a transparent and secure biometric market.”
In recent weeks, the Ethiopian government began a trial of Tech5 biometric technologies for registrations into a MOSIP-based identity management scheme and Chile began using its face biometrics to assist in COVID-19 health status efforts, involving biometric tracing.
accuracy | algorithms | biometrics | dataset | ethics | face photo | facial recognition | neural networks | research and development | TECH5 | training