- Researchers at the retail giant are working on machine-learning systems that could potentially shape fashion trends of the future.
- The system relies on a tool called GAN (generative adversarial network).
- It consists of two neural networks that learn from looking at raw data.
- One generates fake images based on the Amazon data set while the other uses that data to deermine if the image is real.
- Using GANs on the ‘Other items you might like’ section on the website could help retailers work out what consumers might like.
- They internalise specific properties of a certain style by looking at lots of examples of it being worn.
- They then work out what looks good and can apply this information to specific items of clothing.
In a new paper, researchers from the University of California and Adobe trained a neural network to work out how someone likes to dress. Using data from Amazon, they created six categories of items – shoes, tops and trousers for both men and women, writes Technology Review. The system relies on a tool called GAN (generative adversarial network). GAN which uses input to ‘teach’ an algorithm about a particular subject by feeding it massive amounts of information.It consists of two neural networks that learn from looking at raw data. One generates fake images based on the Amazon data set while the other uses that data to determine if the image is real. Using GANs on the ‘other items you might like’ section on retail websites could help companies work out what consumers might like. Experts say the model is still basic – for example if a shopper likes blue shirts, GAN creates more blue shirts. However, in the future these two-dimensional computer images could be turned into 3D renderings that could make actual pieces of clothing. This could be useful retail companies by helping them track trends better and make recommendations based on products popping up on people’s Facebook and Instagram posts.
In separate research, a team at Amazon’s Sunnyvale-based Lab126 has also been developing algorithms that learn about fashion from images and create their own styles from scratch.
The retail giant is currently testing the new service for Prime members called Prime Wardrobe which allows them to try on the latest styles before they buy at no upfront charge
‘There’s been a whole move from companies like Amazon trying to understand how fashion develops in the world,’ Kavita Bala, professor at Cornell University told Technology Review in August. ‘This is completely changing the industry’, said Dr Bala who took part in an Amazon workshop on machine learning last week where they revealed these projects. This algorithmic fashion finder is just Amazon’s latest thrust into fashion and many fashion retailers already use social media to track and respond to changing fashions. Earlier this year the company launched its Echo Look smart camera. The app lets users create a personal lookbook, browse your outfits, and use computer vision-based background blur to make your outfits pop. The feature, called Style Check, uses advanced machine learning algorithms and advice from fashion specialists to let users know whether their choice in clothing was a do or a don’t. The research would make these algorithms even more sophisticated.
For example by analysing labels it could provide feedback or recommend adjustments. The retail giant is currently testing the new service for Prime members called Prime Wardrobe which allows them to try on the latest styles before they buy at no upfront charge. Customers have seven days to decide what they like and only pay for what they keep.
SOURCE