AI is currently wreathed in excitement, but it is still lacking sophistication in important areas like recognising diversity in facial recognition. Google is setting out to remedy this problem through enhanced smile detection.
Google is setting out to enhance smile detection across demographic groups, noting that facial recognition technology has failed in the past to successfully identify people based on factors including ethnicity and gender.
Twofold transfer learning is being used by the team behind the project for the complex task of modelling facial diversity, an advanced form of machine learning that allows stored knowledge to be applied to two problems at once.
So far it appears that significant progress has been made in pursuit of the ultimate goal. In a paper on the initiative submitted by Google engineers Hee Jung Ryu, Margaret Mitchell and Hartwig Adam, the team said: “Our best-performing model defines a new state-of-the-art for smiling detection, reaching 91% on the Faces of the World dataset.”
The limitations of AI image recognition were also recently realised in research conducted by Su Jiawei and a team from Kyushu University. The team found the technology could often be fooled by changing a single pixel, causing it to recognise a turtle as a rifle and a taxi as a dog.
Ryu, Mitchell and Adam also said in the paper: “Recent progress in deep learning has been accompanied by a growing concern for whether models are fair for users, with equally good performance across different demographics. In computer vision research, such questions are relevant to face detection and the related task of face attribute detection, among others.”
Facial recognition has become mainstream with the release of the latest iPhone X, raising awareness to the new technology and increasing expectations of its functionality and widespread availability in the near future.
“We measure race and gender inclusion in the context of smiling detection, and introduce a method for improving smiling detection across demographic groups. Our method introduces several modifications over existing detection methods, leveraging twofold transfer learning to better model facial diversity. Results show that this technique improves accuracy against strong baselines for most demographic groups as well as overall,” said the Google team.