When it comes to choice, more is not always better. The all too familiar conundrum of choosing between dozens of chocolate biscuits or hundreds of toothpastes can sometimes induce a kind of decision paralysis, a phenomenon that psychologists call the paradox of choice.
Skin care conundrum
It’s a problem often seen in the confused faces of customers surrounded by hundreds of products in the skin care section of department stores.
A 2013 survey by consumer product giant Procter & Gamble found one third of women were unable to find what they were looking for in facial skin care aisles. Almost two-thirds say they have unused facial products at home.
“There’s been an explosion of skin care brands and products in the past 10 years or so,” says Dr Frauke Neuser, principal scientist for P&G brand Olay. “One result is that women are shopping in places where they can get advice. However, that can be intimidating for those who might not want to buy a £150 skin cream recommended for them.”
P&G says it has the solution. Its web-based Olay Skin Advisor analyses make-up-free selfies uploaded by users to estimate skin age and make personalised product suggestions. It is believed to be the first artificial intelligence-based skin care advice tool.
It’s the result of decades of endeavour. P&G scientists have been studying skin for over 60 years and developing improved image capture and analysis technology for almost three decades (see “A history in imaging“). That includes a 2015 clinical study comparing the facial skin of 330 women of different ethnicities. These included a subset judged to look at least 10 years younger than they actually are and who proved to have a common pattern of gene expressions.
Inspired by this study, P&G scientists wondered whether artificial intelligence could replace the subjective human judgements of age with something more objectives.
The idea of machines that can perceive the world, learn from examples and understand humans has been around since the 1950s. It is however only in recent years that major strides in building such machines have been made thanks to key advances.
These include the rapid growth in large datasets that include images and the ability to store such vast amounts of data. Also important has been a shift to using computer chips known as graphics processing units (GPUs) originally developed for video gaming. These offer enhanced power over normal chips for certain tasks. “There’s a natural fit between the workloads of deep learning models and gaming. Both involve doing very large batches of computations, and doing so in parallel,” says Thomas Bradley, Director of Developer Technology at NVIDIA, which created the GPUs used to train and run the Olay Skin Advisor deep learning algorithm.
That algorithm, the result of experimenting with image recognition software, is the outcome. It was developed by the P&G bioinformatics group with artificial neural networks, computer simulations inspired by the way human brains process information. They allow so-called deep learning by a machine.
Deep learning models
“Human brains learn to estimate people’s ages by building a model that associates features we see in faces with the ages we know those people are,” says Dr Jun Xu, a P&G computational biologist and lead developer of the Olay Skin Advisor’s VizID™ algorithm. “Deep learning mimics this process.”
Deep learning models are made up of hierarchies of filters called “layers” designed to detect the presence or absence of features.
You must be logged in to post a comment.