Innovation 3.0: Machine Learning and Personalization

I first heard of Machine Learning while doing my Master thesis. At the time, I was helping a music education company build an intelligent algorithm that can “listen” to students’ piano playing, and give them real-time feedback if they played anything wrong. It was an interesting experience that resulted in successfully launching an app called “Singspiel” to the App Store.

Following this experience I then realized how powerful Machine Learning technology is. The general idea is to use well-filtered datasets to “train” the computer and teach it a set of general rules. After the training, the computer will be able to “think” as human beings and generate certain outputs according to the corresponding inputs.

For example, let’s say we want to teach a computer to recognize male/female faces. We would first feed in 1000 male pictures and 1000 female pictures for the computer to analyze. Then, the computer will be able to summarize a set of rules for these two categories (e.g. Male average nose height > 6 cm; 85% female have long hair). This is called “training,” which enables the computer to recognize certain patterns. Next, we feed in a new picture, without telling the computer which gender it represents. The computer will use previous rule-sets to come to its own conclusion. The result may not always be accurate, considering there are exceptions to every rule, and in some cases even humans aren’t able to determine gender from a photo. However, we can expect that the computer will make relatively reasonable guesses and produce a fairly accurate result (e.g. 80% likely male).

These kinds of technologies can be widely beneficial to our business, and are therefore a great addition to the Innovation Challenge technology roster. For example, how about a Boston Pizza app that analyzes your weekly orders and produces a suggested menu customized to your tastes? Or, what if we create an intelligent email service engine that can deploy an email to only the most interested users in order to increase click rates? Or, why can’t we produce a company tool that is more efficient than Google in finding the right internal resource and solution for you when you are facing an issue at work? It’s clear that the possibilities for machine learning are endless, and we’re eager to explore them.