Synaptiq.ai

Our Blog

Our Blog

Teaching Machine Learning
 Image source: https://xkcd.com/1838

Image source: https://xkcd.com/1838

It’s impossible for business executives nowadays to avoid hearing about how some machine learning implementation or the other has catapulted a business and shaken its competitors to the core. There’s much hand wringing about robots replacing white collar jobs. There’s alarm about of how systems with “generalized artificial intelligence” will, in the near enough future, run amok.

How is a business manager to react? How are they supposed to evaluate the potential of machine learning in their own companies or business units? How can they make the most of the data science teams that are becoming part of their organizations?

In the fall semester of 2017, together with a friend and former colleague, Mukul Kumar, I taught a course on machine learning at the Harvard Extension School. The goal of the course was to give business managers and executives “practical knowledge and tools to think creatively about using data and machine learning – in collaboration with their data science teams -- to advance their business goals.”

Machine learning is a theory-rich field. Learning its conceptual and mathematical underpinning is easily a Ph.D.-style pursuit that usually takes 3 to 5 years of intensive postgraduate study. How then did we teach this subject to newcomers in 13 weeks and teach it in a way that would enable them to collaborate productively with the data science teams in their organizations?

The approach we took is one that the famous physicist Richard Feynman describes in his book QED: The Strange Theory of Light and Matter, a popularization of the theory of quantum electrodynamics. In the first few pages of that book, Feynman describes how he’s going to explain a complex, highly-mathematical theory to a lay audience. He writes:

“To understand how subtraction works -- as long as you don't have to actually carry it out -- is really not so difficult. That's my position: I'm going to explain to you what the physicists are doing when they are predicting how Nature will behave, but I'm not going to teach you any tricks so you can do it efficiently. You will discover that in order to make any reasonable predictions with this new scheme of quantum electrodynamics, you would have to make an awful lot of little arrows on a piece of paper. It takes seven years -- four undergraduate and three graduate -- to train our physics students to do that in a tricky, efficient way. That's where we are going to skip seven years of education in physics: By explaining quantum electrodynamics to you in terms of what we are really doing, I hope you will be able to understand it better than do some of the students!"

In our machine learning course we taught students the equivalent of making lots of little arrows on a piece of paper -- a scheme for calculation, but not necessarily the most efficient scheme of calculation. But efficiency doesn’t matter as much as learning the concepts; and the concepts can be made clear without any mathematics or buzzword mumbo jumbo.

It is a minor miracle that such a scheme is even possible for machine learning. While machine learning is highly theoretical, it is also immensely practical. You can implement a machine learning system without much fuss if you understand the conceptual nuts and bolts. You don’t need to be familiar with the intense mathematical work that underpins these concepts. Think of it like this. A detailed knowledge of chemistry might make you a better cook, but there are plenty of great cooks who know no chemistry.

If you have some ability to code, there are a mind-blowing number of excellent tutorials that can get you started. A weekend is all it takes! But what if you don’t want to code? There are platforms like Orange that enable you to implement machine learning models without writing a line of code. This sounds too good to be true, but it is true. By the end of the course students were building sophisticated models in Orange to solve real-world problems using real-world data.

If you’d like a simple explanation of how machine learning works and what you can do with it at your company, reach out! We’d be delighted to give you and your team a leg up on machine learning!