Machine learning vs. deep learning
The hottest topics in the current technical field are machine learning and deep learning technology. You will detect these words in the context of Artificial Intelligence (AI), self-driving cars, humans hitting games, and other new technology advances. If you are new to the AI industry, you may be wondering about the differences between these two technologies.
To be clear: deep learning and machine learning are the parts of artificial intelligence. Furthermore, deep learning is a part of machine learning. Machine learning is an Artificial Intelligence strategy, and deep learning is a machine learning strategy.
Machine learning is one of the many AI strategies
In the early days of AI, the field relied on hard-coding rules and algorithms. Playing chess against AI is an exercise in rude computational power; the computer program looks at every possible move and selects the step with the best possible result. AI chatbots can have a “conversation” with you by searching for specific words and phrases provided by the user. Then it responds with the canned responses that a programmer imagined long before (modern virtual assistants still depend on this technique). While these practices may look intelligent, they depend on their projected intelligence – they lack the ability to learn with experience.
A machine learning algorithm is trained by providing real-world data instead of relying on hard-coding rules to solve problems. Machine learning then creates a model that looks for patterns between the data you submit and what you are trying to predict. That model can generate predictions for new things that have never been seen before. As the model is exposed to more training data, its accuracy is getting better and better.
One simple example is to imagine that you want to build a system that can predict the sale price of a home-based on the characteristics of that home. With things like home location, square feet, the number of bathrooms, and age, you can train the machine learning method using historical feeding data on home sale prices. The process of identifying how these different properties of home are affected creates a model that begins with an understanding of how the sale price and the final price of each property affect the home. For new homes on the market, this machine learning algorithm can be used as a model to automatically predict its sales price. As more and more home sales are delivered to the system over time, its accuracy gets better.
This machine learning system does not believe in a human-made rule; rather, it teaches them based on actual data.
Deep learning is one of many machine learning strategies
How might that house price system work? It’s really simple; you can plot a variety of properties, such as square feet, against the sales prices you train the system, apply a curve to each, and use those curves to predict the price of new houses hitting the market. This is called multiple regression.
Or, you can build a decision tree that learns a series of decision points that lead to an accurate price prediction. It can start with the price range for a particular area and can be further enriched by the house’s age, depending on the size of the house and until the final price is estimated. These are two of the many machine learning methods we can use, but they don’t call them “deep learning.”
Deep learning as convoluted artificial neural networks
Although deep learning is another machine learning strategy, it has drawn great admirations because it is so flexible – and is inspired by how our own human brain works.
Deep learning systems are composed of layers of virtual neurons. The task of each neuron is to collect the inputs and determine whether the output signal is emitted to the next layer of neurons above it.
Each neuron in one layer is connected to each neuron in the layers of the network above and below it. This neural network can solve a variety of problems by learning the optimal weights for each of these links. Although a neural network is a simple concept, the sheer number of connections between neurons indicates that they represent very intricate problems.
When it comes to real estate pricing, all the properties (location, size, etc.) in your training data are processed identically and delivered to the neurons in the lowest layer of your neural network. With many iterations, the neural network weighs the best between its connections, making an accurate price prediction at its highest layer output. The model can quickly begin to predict prices for new homes that have never been seen before this neural network is trained to weigh between neurons.
When the number of layers in a neural network is more than one, we say that it is a deep neural network. This is the meaning of the word deep learning. The deep learning model is a machine learning system operated by a deep neural network. This is not the case with machine learning and deep learning; Deep learning is a machine learning technique – and one of the most exciting! We only scratched its surface here; there is still much to learn.