What Is Deep Learning and Neural Networks?
This is part 5 of a 9 part series on Machine Learning. Our goal is to provide you with a thorough understanding of Machine Learning, different ways it can be applied to your business, and how to begin implementations of Machine Learning within your organization through the assistance of Untitled. This series is not by any means limited to only those with a technical pedigree. The objective is to provide a volume of content that will be informative and practical for a wide array of readers. We hope you enjoy and please do not hesitate to reach out with any questions. To start from part 1, please click here.
What Is Deep Learning?
So what is deep learning and neural networks? Deep Learning is a part of the broader category of Machine Learning methods that mimics the learning structures of biological intelligence such as the human nervous system. Within the field of Machine Learning, a great deal of new buzz for the category surrounds the potential of Deep Learning. Quite a few of the most impressive emergent technologies such as self-driving cars, our personal home assistants such as Amazon’s Alexa and Google Home, and natural language processing are driven by Deep Learning algorithms.
I have held off on discussing Deep Learning throughout my series thus far, because it can leverage all three styles of learning: supervised learning, unsupervised learning, and reinforcement learning. With that said, Deep Learning really does deserve a post of its own, and could in many ways be thought of as its own respective field.
What is a Neural Network?
A Neural Network is one of the many types of Deep Learning structures available to us. Neural networks were first invented in the early 1940’s by Walter McCulloch and Walter Pitts. They used a straightforward mathematical method called threshold logic, which later paved the road for binary classification neural networks. Threshold logic was a system originally designed for electrical circuitry and how currents would pass through the circuit board using binary units of logic at different gateways, which fit well given the way electrical signals proceed through the human brain.
Neural networks come in a variety of shapes, forms and stylizations but a few important terms to take note of are inputs, hidden layers, outputs and computational nodes.
Inputs are relatively simple to understand, as this is the data we send into the neural network. Think of it as the raw data or stimulation that we are sending through the network, with the goal of using the network to classify the data in the output. Hidden layers are all of the computational layers we put into the neural network to assist in the classification process.
A rule of thumb is that the more layers you add to a neural network, you’ll get increased accuracy in classification. However, with increased layers, you sacrifice speed and computational resources. Outputs are self explanatory, they’re essentially the target variables we want the neural network to predict. Finally there are computational nodes, which act as the individual neurons in the layers of the neural network.
To give an example: if I wanted to build a neural network that could determine if an image contained a dog, it would look a bit like this:
Inputs during training: Labeled images of dogs
Inputs post training: images of animals, some dogs, some not dogs
Hidden layers: Computational layers that deconstruct and reconstruct the likeness of a dog, in search of feature commonality
Output: A classification of an image, and whether it contains a dog, or doesn’t contain a dog, with a degree of confidence such as 95% chance the image contains a dog, 5% chance it contains a sea otter.
A Neural Network is designed to mimic the way neurons behave. Through constructing a multitude of layers within a Neural Network, we step closer to modeling the way human brains think and respond to information stimuli in the form of computer information systems. The closer we get to deconstructing the pathology of the brain and consciousness, the closer we step towards true artificial intelligence.
How Do Neurons Work?
A neuron is the fundamental informational building block of the human nervous system. Neurons pass information back and forth to each other, as well as to cells and muscles. When neurons send and receive information, they are essentially transmitting electrical impulses to one another. In a very similar fashion, the neural network is transmitting synthetic information (e.g. data) back and forth amongst its nodes and computational layers to derive meaning from its inputs, then assign value to its outputs.
The Implications for Deep Learning as a Field
At the moment, there are a few ambitious Deep Learning projects that have the goal to reverse engineer human consciousness, ultimately to derive the data needed to build a truly conscious machine. Deep Learning brings mankind at a very peculiar intersection of neuroscience, computer science, advanced mathematics, law and ethics. Without the incredible work that has been done in neuroscience to date, we would not be able to build the Deep Learning systems that are driving important parts of consumerism, culture and health. However, neuroscientists themselves would tell you we are far from understanding how the human brain really works.
The Black Box Paradox
One of the strongest implications of Deep Learning is what technologists refer to as the black box paradox. Essentially, the idea behind the paradox is that we know the building blocks of the intelligent algorithms that we create, however, the outputs and derivations of what the algorithms produce can be perplexing.
It is quite difficult for us to reverse engineer Deep Learning algorithms to understand how it arrived at certain outputs or beliefs. This has immense implications for the human race given the fact that these algorithms will and do drive many of the key factors of our lives. If such algorithms develop biases that are negatively disposed towards race, religions, sex, gender, wealth distribution and others, these can potentially become even more systemic through the permutations of Deep Learning algorithms.
In Summary
We hoped you enjoyed this post and will continue on to part 6 The Difference Between AI and ML. The next post will investigate the origins of AI and clarify the differences between Machine Learning and Artificial Intelligence.
If you are interested in starting on a Machine Learning project today or would like to learn more about how Untitled can assist your company with data analytics strategies, please reach out to us through the contact form.
Check out part six of this series.
