What You Need to Know About Machine Learning in 2016

I was recently listening to a podcast featuring Kathryn Hume and Nick Vermeer of Fast Forward Labs, a company that works on cutting edge technology to help organizations sort out and use their data with the most up to date (even futuristic) machine intelligence capabilities. The topic of discussion was machine learning for the most part, with tangents including favorite Sunday activities and ways to stay creative in the tech industry. 

In addition to being an awesome show, I felt as though I learned a bit about machine learning and its applications. SO I thought today we could dive into the basics of machine learning and some cool ways it's being used in the fashion world - as learnt from this particular episode of Fashion is Your Business

Note, there is a facet of AI and Deep Learning I won't get into here, to be reviewed at a later date. Stay tuned. 

Machine Learning - What is it?

Machine learning evolved from finding patterns in human behavior that could effectively be called 'rules' for a machine to 'learn.' An example might be in pleasantries. When my waiter pours me a glass water, I say 'thank you.' You say 'How are you?' I respond, 'Good, how are you?' Much like a child learns her manners by observing patterns and repetition, algorithms can find set patterns within mass data sets and even make predictions on future data.

An easy example the Fast Forward group presented was in real estate. A good real estate listing generally follows a certain pattern. (A magnificent sea view may be more than 45 degrees. A spacious living room is over a certain square footage.) Given a large sample of listings to learn from and a data set on homes for sale in a particular market, machine learning can do the work of converting your excel doc to listings in no time. Real estate companies now can input an excel with standard headers: location, square footage, amenities, price, etc, and an algorithm can effectively put words around the data to create a perfectly executed housing listing faster than a human could do one.

A little bit on Artificial Intelligence

Artificial Intelligence (AI) is the 'science fiction' side of machine learning. Though technically the same thing, you can use the general term AI when referencing machines able to function as a being. For example, conversational bots are now being built that use machine learning to analyze patterns of data to sustain a conversation with me on those blue jeans I've been coveting.

Applications

As far as research goes, Fast Forward is deep into AI. However, real world application is a slow and steady process. Below are some companies making interesting headway.

Pictograph.us uses deep learning to analyze what you generally take pictures of on Instagram. To accomplish this, an enormous data set of pictures identifying objects would have taught an algorithm what exactly a cheeseburger looks like versus a baby. The evolution of this might be the ability to analyze that cheeseburger's calorie content through that same photo.

The Prisma app represents burgeoning new media art, applying an AI technique dubbed 'style transfer' to your insta photos. Machine learning has been able to abstract out patterns in famous works of art and then transfer the style to your sexy selfie. So that Fourth of July fireworks photo can now be made to look like Vincent Van Gogh's Stary Night.

Machine Learning in Fashion

In the world of fashion, a basic level of machine learning is being used in CRM systems to support the ever increasing focus on customer loyalty. Within their CRM system, Louis Vuitton (for example) can actually ask questions such as 'do you prefer whiskey or wine?' to their most VIP clients. They will then use this information to provide the ultimate in luxury experiences, reinforcing clients' loyalty to their brand. 

In e-commerce, collaborative filtering has long been used to make recommendations based on previous habits or the habits of 'someone like you.' Collaborative filtering is limited however based on the fact that trends move quickly - what I was searching on Amazon Fashion six months ago is likely not what I want today. What's new in this space is the ability to analyze images to find trends in user preferences, whereas previously, it was very difficult to collect data on image based searching. So now, those recommendations can be much more tailored to what you're viewing today.

Stitch Fix is paving the way forward for machine learning in recommending products with their online stylist. The website is a simply fun and seamless application of serious technology. 

Brick and Mortar surprisingly has perhaps the most radical changes. Using satellite date, in store conversion is easier to calculate as in store traffic is finally measurable. My favorite machine learning use is in probabilistic programming, which allows you to improve your confidence in the answer to a question. i.e. "Where should I open my next store? Chicago is doing well. Boston is hitting it over the Green Monster. Miami is a flop." A probabilistic programming model will take your (much more robust) dataset and compute your inference, give you a reason why you should move to New York City, and yield a percent confidence in the decision.

Finally, the Creepy Facts

Privacy Laws have not developed as fast as technology because of the fact that consumers simply haven't pushed for further security. This is likely because most of us don't know what stores are capable of doing. 

Stores can actually fingerprint your open wifi to follow you around stores and then predict future behavior. Even if you're not actively using your wifi, it just needs to be turned on for stores to track you. Great for stores, somewhat scary to little old me wandering around the lingerie section of Bergdorf's.

If your clothes have the RFID security tags still attached, which also allow retailers to have real time inventory management, the stores can also actually track you wherever you go. Please don't do anything naughty and try to escape while wearing your trackable device.