Katy consults on the impacts of machine learning on small to medium size engineering projects.
The terms "machine learning" and "deep learning" have turned into buzzwords around AI (artificial intelligence). But they don't mean the same thing.
A beginner can understand the difference by learning how they both support artificial intelligence.
What is Machine Learning?
Let's start by defining machine learning: it's a field covering all methods used to autonomously teach a computer.
You read that right! Computers can learn without being explicitly programmed. This is possible through machine learning (ML) algorithms. Machine learning gives software a problem and points it to a large amount of data to teach itself how to solve it.
This is similar to how humans learn. We have experiences, recognize patterns in the real world and then draw conclusions. To learn "cat" you saw a few images of the animal and heard the word. From that point on any feline you saw on TV, in books or in real life you knew was a cat. Computers need more examples than humans but can learn with a similar process.
They read in large amounts of data about the world. The software draws its own conclusions to create a model. It can then apply that model to new data to provide answers.
Does computers teaching themselves sound like futuristic AI? Yes, machine learning is an important aspect of Artificial Intelligence, or AI.
What is Deep Learning?
Now that we understand machine learning, what is deep learning? Deep learning is a subset of machine learning. It is one type of machine learning method for teaching computers.
Machine learning can either be accomplished through shallow learning or deep learning. Shallow learning is a set of algorithms
Linear regression and logistic regression are two examples of shallow learning algorithms.
Software needs deep learning when the task is too complex for shallow learning. Problems that use more than one input or output or multiple layers need deep learning.
They use "neural networks" of shallow learning algorithms to accomplish this. Neural networks are an important part of understanding deep learning so let's dig into that.
Deep learning uses a "neural network" to tackle these complex problems. Like neurons in the brain these models have many nodes. Each neuron or node is made up of a single shallow learning algorithm like linear regression. Each one has inputs and outputs that feed to the joining nodes. The layers of nodes progress until it reaches the final answer.
It's the job of deep learning to decide what that neural network needs to do to get to the final answer. It practices on data set after data set until it refines the neural network and is ready for the real world.
One of the most fascinating parts of deep learning is that the humans never need to program the inner layers of a neural network. Often, programmers don't even know what is going on in the "black box" of a neural network once it's complete.
Machine Learning vs Deep Learning
The terms "machine learning" and "deep learning" are sometimes used interchangeably. This is incorrect but even people familiar with the concepts will do it. So when interacting in the AI community it's important to understand the difference.
Machine Learning Terms
When people use "Machine Learning" in conversation it can have different meanings.
Field of Study: Machine learning is a field of study. While there's not an explicit Machine Learning degree in the US it's considered a subset of Computer Science.
Industry: Machine learning represents an emerging industry. Those concerned with business usually talk about AI and machine learning in this context.
Technical Concept: the term "machine learning" also represents the technical concept. It is an approach to solving large software problems with big data.
More About Machine Learning
Machine learning will be used by more and more industries to improve our lives. It's important to understand more basics about the process.
Smarter than a Human
With conventional programming computers are only as smart as the people who program them. But machine learning methods allow computers to see patterns on their own. This means they make connections that humans can't even imagine.
Rise of Machine Learning
Why are we hearing more and more about ML and deep learning recently? That's because the necessary processing power and data has only recently become available.
Something else that enables machines to learn is the shear amount of data available. Software needs to see a lot of data to build a reliable model. The data produced from the Internet and smart phones gives computers insight into how to help humans.
In the past, computers weren't able to consume the large amount of data they need to draw connections. Now, they can crunch all that data in a reasonable time.
One of the draws of ML algorithms is that the software continues to learn as it encounters more data. So a team can allow software to learn enough to be helpful and then deploy the system. As it encounters more real world tasks it continues to learn. It will continue to refine its rules as it finds new patterns.
© 2018 Katy Medium