Machine learning and deep learning are both hot topics and buzzwords in the tech industry. You’ll hear these topics in the context of artificial intelligence (AI), self-driving cars, computers beating humans at games, and other newsworthy technology developments. If you’re new to the AI field, you might wonder what the difference is between the two.
Think of it this way: deep learning and machine learning are both subsets of artificial intelligence. And, deep learning is a subset of machine learning. Machine learning is an AI technique, and deep learning is a machine learning technique.
Machine Learning is One of Many AI Techniques
In the early days of AI, the field relied on hard-coded rules and algorithms. Playing chess against an AI is an exercise in brute computational force; the computer program looks ahead at every possible series of moves and chooses the move with the best outcome. AI chatbots can hold a “conversation” with you by looking for certain words and phrases provided by the user. It then replies with canned responses a programmer thought of ahead of time (modern virtual assistants still rely on this technique). While these systems may seem intelligent, they depend on their programmed intelligence – they have no ability to learn with experience on their own.
Machine learning flips that on its head. Instead of relying on hard-coded rules to solve problems, a machine learning algorithm is trained by feeding it real-world data. Machine learning then builds a model that looks for patterns between the data you give it and the thing you’re trying to predict. That model can make predictions for new things it’s never seen before. As the model is exposed to more and more training data, its accuracy gets better and better.
As a simple example, imagine you want to build a system that can predict the sale price of a house based on the attributes of that house. You might train a machine learning algorithm by feeding it historical data of house sale prices, together with things like the home’s location, square footage, number of bathrooms, age, etc. The algorithm would start to find how these different properties of a house affect its sales price and build up a model that understands how each attribute affects the ultimate price of the home. For new houses going on the market, this machine learning algorithm could use the model to predict its sales price automatically. And as more and more home sales are fed into the system over time, its accuracy will get better and better.
This machine learning system isn’t relying on ruled programmed by a human; rather, it is learning them based on real data.
Coming back to the real estate pricing example, all the attributes in your training data (location, size, etc.) are processed to be on a similar scale and fed into the neurons at the bottom-most layer of your neural network. Through multiple iterations, the neural network arrives at the best set of weights between its connections to produce an accurate price prediction at the output of its top-most layer. Once this neural network has been trained with the best weights between neurons, it can start to quickly predict prices for new houses the model hasn’t seen before.
When the number of layers in a neural network is more than one, we say it is a deep neural network. And this is what we mean by the term deep learning. A deep learning model is a machine learning system implemented by a deep neural network.It’s not a case of machine learning vs. deep learning; deep learning is a machine learning technique – and a very exciting one! We’ve only scratched the surface of it here; there’s much more to learn.