An Introduction to Machine Learning for Business People
If you’re a business person, there’s a lot you need to know about machine learning (ML) . Why? Because it can drive unparalleled efficiency and innovation to your company. Still not sold? Back in the day, neither was Xerox.
I’m a believer in the phrase “you can never learn too much”. That being said as a business person you don’t need to know everything about ML. For the most part you just need to know what it can do, and what it can’t do.
What can Machine Learning do? What can’t it do?
Machine learning denotes the ability for a system to learn on its own via experience, without it needing to be directly programmed. You can feed in large volumes of information (ie. training data) to have it perform a task, and then have it get better as it performs more tasks.
Consequently, ML’s potential is enormous. Here are a few applications for a business person to consider:
Computer Vision — Teaching a computer to understand what is happening in a video or image
Natural Language Processing (NLP) — Enabling a computer to understand what is contained within texts, read, extract, and make relevant implications
Deep Learning — Allows for complex, multi-variate prediction across a number of different dimensions, using layered neural networks
Some business people like to think of ML as analytics on steroids. That isn’t a particularly useful heuristic, however, as it doesn’t tell you much. Instead I would distinguish it more meaningfully in the following way:
Analytics enables you to get specific insights you were looking for. ML enables you to perform specific actions you were looking for, and then potentially gain insights, including some that you may not have intended.
Sometimes you can have ML perform an action without understanding how it did it — because the volume of data it used was so vast.
To better understand this, let’s close with a more technical view of data modelling, and data science (the umbrella under which both analytics and ML sit).
The technical lines I would draw between ML and analytics is a regression, either linear or logistic.
A linear regression is an equation with a series of inputs (and sometimes weights are attached to these inputs) that will tell you the linear relationship between a dependent variable, and an independent variable. Similarly, a logistic regression will do the same, but within a specified relevant categorization (ie. yes/no, ranking order, etc).
These can enable an insight, or an action, so fit under both ML and analytics. More complicated models as I discuss next, however, fall under ML.
A neural network is an equation with weights that fire when the outcome passes a threshold parameter. If the answer is greater than 1.0, for example, the neuron fires. They can be simple for linear requirements, or multi-layered for more complex ones. A decision tree functions similarly, but operates along a number of different binary paths, that leads to more disparate outcomes. Random forest is a data model that is made up of a number of decision trees, all perturbed along different binary points in order to get a complete picture for prediction.
These different models come with their challenges. An inference attacks is when the information you’ve provided on a model like random forest reveals too much information about your private data set. If, for example, a factor is given a very high weights only in specific circumstances, that can provide re-identifying information on your data. In addition, data scientists have to deal with issues like overfitting, where your model is trained too specifically for the data set that you’ve provided, and is unable to give you meaningful implications for the real world.
The world of machine learning is growing, and you can never learn enough. The information I’ve provided above, however, should enable you to have a conversation — and that is often where the best learning starts.
