Machine Learning Basics: What It Is and How It Can Work for You
What is machine learning?
Why is machine learning important for my business?
How does machine learning work?
Imagine it’s your first day at Marjory’s Muffins, a factory responsible for manufacturing thousands of muffins a day to be sold at supermarkets around town.
Your favorite part of the job, naturally, is taste testing and coming up with fancy new flavors, but you also have to inspect all muffins manufactured before they get shipped to supermarkets.
At Marjory’s Muffins, every muffin’s appearance needs to be magnificent, and any defects in the muffins make them unfit to be sold in supermarkets.
Are you ready to examine the muffins?
Machine learning uses math to take data, learn from it, and make decisions or accurately predict outcomes based on what it’s learned.
Machine learning can be a great way to smartly handle repetitive or dull tasks, freeing up valuable time for people to focus on the kind of tasks humans perform best, such as complicated decision-making not easily handled by a machine.
Imagine instead of asking an employee to spend the whole day on a time-consuming, taxing task like inspecting muffins, you could have them tasting muffins or interviewing supermarket customers about flavor preferences.
If Marjory’s Muffins collects images of muffins (both good and bad) as they roll off the production line, they could train a program to determine whether or not each muffin produced looks good enough to ship to stores.
Of course, machine learning – also called ML – isn’t just a fancy way to handle quality control. In fact, its uses are wide ranging.
From helping customers find the products they want more easily on your website to helping cars drive themselves, ML software such as TensorFlow, can be trained to do very different things.
While the world of machine learning involves a lot of vocabulary, including model, example, input, label, and, of course, algorithm, you only need to understand the basics to start using machine learning for your business.
First, there’s model – that’s what machine learning software is called once you train it.
To train a model, you show it examples. For instance, to train an ML model to catch defects in muffins before they’re shipped out to supermarkets, you’d show it many, many examples of muffins.
An example consists of an input and a label for that input. In your dataset of muffin images, the input would be muffin images, ones that look good and are labeled as good and defective ones labeled as bad.
After you train the model to know what’s a good muffin or a bad muffin, you can use it to predict the label (good or bad) of images of muffins it’s never seen before, specifically the muffins you produce every day at the factory.
LISTEN UP
ML works in this case because the model has learned from specific examples a general idea of what “good” looks like.
To train an ML model to perform this generalization, you need data, and lots of it – a large, labeled dataset is essential for successful ML.
For most tasks an ML model can be trained to perform, there is a standard algorithm it can use.
Think of a standard algorithm like a wrench: You can use a wrench to fix a leaking pipe, or you can use it to build a skateboard. Either way, it’s the same tool and the same function (gripping/turning bolts), but you’re using it differently.
The algorithm Marjory’s Muffins might use to monitor muffins is called an image classification network. An agriculture company could train that same algorithm to perform a totally different task, like detecting diseased leaves in photos of plants.
Other standard algorithms can help with other, non-image-based tasks. Luckily you don’t have to know how every algorithm works – just know that there are different tools to choose from in your machine learning toolbox.
Choosing the right standard algorithm may be like choosing the right tool, but unlike hardware tools, you need to train a standard algorithm in order to use it.
Let’s say the company training an algorithm to detect diseased leaves is called Terrific Trees. To perform their tasks, Terrific Trees and Marjory’s Muffins both start with the same standard algorithm used to evaluate images.
To use this algorithm for their own purposes, Marjory’s Muffins has to train the algorithm to understand photos of muffins. Meanwhile, Terrific Trees trains the algorithm to understand photos of leaves.
While both Marjory’s Muffins and Terrific Trees started off with the same algorithm, they trained them on different labeled examples and came up with 2 unique ML models – one used to classify muffins and one used to classify leaves.
Once you’ve trained a model, you can actually reuse the same code for other situations focused on the same kind of task.
Since identifying defective muffins involves classifying images, you might reuse the same model to identify bad ingredients used in the muffin making process.
ML takes time to train and will make mistakes. The more data and time you can give an algorithm to learn, the better the results will be. The best time for your business to use ML is when you need to make repeated decisions and have lots of data.
DO THIS NOW
Now that you’ve had an introduction to machine learning, let’s imagine a task you have and see if ML is the right way to get it done.
If you’re participating in the course, go to the next section to access your self assessment.
KEY TAKEAWAYS
Machine learning uses math to take data, learn from it, and make decisions or accurately predict outcomes based on what it's learned.
The key to making an ML model generalize is data, and lots and lots of it - a large, labeled dataset is essential for successful ML.
For most problems a machine learning model can be trained to solve, there is a "standard algorithm" it can use to solve it.