Deep Learning for Dummies: An Overview of Deep Learning and Neural Networks for Beginners

by Limor Wainstein

Deep learning and neural networks have been making a lot of headlines in the technology sphere lately, and with good reason too. The growth of deep learning is an important aspect of what some sources are calling “the A.I. revolution”. Artificial intelligence is nothing new, however, its potential is now being realized more than ever due to the availability of cheap, vast computational power and enormous stores of data.

The projected growth of deep learning over the next several years is staggering. A report by Grand View Research expects the global market to be worth $10.2 billion from 2025, having been worth just $270 million in 2016.

Dedicated A.I. platforms and frameworks, including TensorFlow and MissingLink, are becoming more common in the enterprise as companies look to leverage the power of deep learning and neural networks.

While the terms “deep learning” and “neural networks” are often used synonymously, they are not the same thing, and a semantic distinction is important if you want to properly understand this family of artificial intelligence. Read on to get an overview of deep learning and neural networks for beginners.

Neural Networks: Overview

A neural network, or more precisely, an artificial neural network, is a type of computing architecture designed to somewhat mimic the biological neural networks that make up human (and other animals’) brains.

Oliver Selfridge, who was a pioneer of artificial intelligence research, released his famous Pandemonium model in 1957, and this formed the early basis for modeling cognitive processes using computers. Neural networks can learn things using data alone, by mapping inputs to outputs.

Modern neural networks are more complicated than the above example, but the premise is similar. The architecture of a neural network is a layered design with each layer composed of many neurons. These neurons are living cells in the context of biology, however, in computer systems, they are simply mathematical functions that make decisions based on input data.

A neural network excels at learning how to recognize patterns autonomously, but you need to feed it the right amount of data for it to become good at this.

For example, take a neural network that learns how to recognize images of dogs. In the beginning, you feed images into the system as data points, e.g. dog (1) or not dog(-1)). But at this stage, each neuron outputs random data (dog or not a dog), and it is no more than guesswork. The output is random because the system initially assigns random weights to each input it receives.

But here’s where the magic happens. The weights are adaptable, and the way they adapt is by using an algorithm that compares the output of the network with what the output should have been (dog or not a dog). From this comparison, the neural network calculates its error, and the error is sent back through all layers of the system via the algorithm.

The weights are then adjusted based on this error, and the system becomes better over time at recognizing what is an image of a dog and what isn’t. The improved learning occurs by feeding the network large numbers of images to more precisely adjust the weights, or connections between neurons.

Neural networks require what is known as supervised learning. Someone, typically a data scientist, will feed the system with labeled data to help the network learn how to classify things and recognize patterns.

Deep Learning: Overview

Deep learning networks are designed based on neural networks. However, the main differences are that deep learning networks have many more hidden layers, and crucially, deep learning networks can perform unsupervised learning in addition to supervised learning.

Advancements in processing power have catered for deep neural networks and made unsupervised learning possible. This type of learning occurs when the deep learning network learns on its own how to categorize and classify data that has not previously been labelled.

Use Cases for neural networks and deep learning:
  • Speech recognition has come on leaps and bounds, and the popular Amazon virtual assistant Alexa uses a neural network.
  • The enhanced abilities of computers to recognize images promises to positively shape the future of medical image diagnostics. Deep learning networks have underpinned this evolution in medical image analysis.
  • Google has used deep learning and neural networks to improve the fluency and accuracy of its translation services.
  • Deep learning is being used to improve security on the internet by detecting network intrusions and malware quicker and more accurately.
Wrap Up

Hopefully this article has made clear what neural networks and deep learning networks entail without getting too much bogged down in complicated math and science. This subfield of artificial intelligence has enormous potential, as is evident in its sample uses cases. And some of the most exciting use cases for deep learning and neural networks are surely yet to be realized, so watch this space.