🤔 AI thinks is Miranda Priestly now

Happy Tuesday!

"I Had Hope. Anyway, You Ended Up Disappointing Me More Than Any Of The Other Silly Girls.” —Miranda Priestly

Will Fashion tech disappoint Miranda? The jury is out.

In today’s newsletter:

• AI thinks is Miranda Priestly now — And it may just give her a run for her money.

• Papers

• New Feature — Size matters. Especially when it comes to a 24.4% Avg. Return Rate.

• Learn

Penses

AI thinks is Miranda Priestly now

The now cult classic The Devil Wears Prada has done much to solidify workplace trauma in the minds of the American psyche. Miranda does some pretty mean things for the sake of keeping her job. But even she isn’t prepared for this thing called AI. The data scientists are coming for her job.I’ll personally be betting on Miranda Priestly all day, every day. But she has found some stiff competition in recommendation models.

The IMDB movie dataset has been a stable in data science portfolio projects for years. Movies have been the foundational representation of recommendation models for a long time. Projects have looked at the:

• similarities between movies from in a genre,

• runtime of the movie,

• reviews,

• actors ,

• and more.

All to get you to spend more time watching movies. Well now, Fashion Scientists trying to bring AI to fashion styling. But they don’t think that approach will work for them. Fashion changes, its subjective, their textures, brands, and the list goes on.

A rom-com is a rom-com in movie world but a cerulean belt isn’t a cerulean belt in the fashion world.

Many e-commerce store websites we see recommendation model. It's that box that has ' these products are often bought to together' section. They use the metadata of the product to make recommendations. But not the image or how compatible the item is to other products. That matters in fashion. They outfit needs to match.

Researchers have started looking into neural networks to build image based recommendation models. They hope to perform in-depth fashion analysis from clothing to makeup and so on. To do it, they created a unique fashion ontology.

Some researchers have called the categories:

• the previous clothes purchased that we know about,

• the clothes being purchased,

• and the purpose of the clothes.

This brings us into the categories of recommendations which include:

• item retrieval,

• complimentary item

• recommendation, whole outfit recommendation,

• and capsule wardrobe recommendation.

Watch the movie here.

Paper

Current techniques being used include taken from Study of AI Driven Fashion Recommenders:

• “A Dual Attribute aware Ranking Network (DARN) to represent in-depth features via attribute-guided learning. DARN modeled domain dis-

• “…Built three separate algorithms for recovering the same fashion item in a real-world photograph from an online shop. Two deep learning baseline techniques were included in the three approaches, and one method learned the similarity between two street and store domains”

• “researchers introduced a deep bi-directional cross-triplet embedding approach to describe the similarity of cross-domain pictures, which enhanced the one-way problem, street-to-shop retrieval job. They also extended this technology to fetch several related accessories to go along

• “A Graph Reasoning Network (GRN) was introduced to construct the similarity pyramid to improve on existing retrieval task methods that only considered global feature vectors and represented the similarity between a query and a clothing inventory by considering global and local representations.”

• “An early work was introduced on the computer-generated design of fashion garments as a part of applications of Generative Adversarial Networks.”

  The list is even longer than the first few I mention here. The future is going to be a fairly interesting place. Don’t you think?

Read more on AI, Fashion , and recommendation engines here , here , and here.

Classic reading on recommendation systems here.

New Features

Amazon is trying to reduce it’s loss from returns by helping you find better fit clothing. It’s doing this by using generative AI models to read the comments section of products. It then summarize customer reviews on the items. Read the full article on TechCrunch.

Learn

For information of the type of models mentioned here are some good videos.

One more here