Machine learning and deep learning, sometimes used interchangeably with artificial intelligence, are terms we are increasingly exposed to but what is really the difference between the three terms?
Writing this post, I quickly realized that a consensus of the three terms’ definitions has not been established. Nevertheless, the following should give you a good breakdown of the three terms, artificial intelligence, machine learning and deep learning. Let’s get to it!
Artificial Intelligence: Human intelligence performed by machines
AI regards machines that can conduct tasks that characterize human intelligence.
Things that require “human intelligence” is quite unclear and general but it comes down to tasks such as planning, interpreting and translating languages, solving complex issues and the ability to recognize different sounds and objects – for example, the cup of coffee on your desktop.
AI is thus really about making it possible for machines to learn from experience and improve its ability to conduct human tasks as new inputs are given.
Machine Learning: An approach to achieve AI
ML is “the ability to learn without being explicitly programmed”(Samuel, A. L. 1959).
Elaborating a bit on the brief definition provided by Samuel (1959), machine learning is the science of getting machines to learn and act as we humans do such that the machines can autonomously learn and improve over time.
Machine Learning is thus a way of achieving AI. With ML you “train” an algorithm to learn by itself by providing huge amounts of data for the algorithm to process and improve upon – hence the term machine learning.
To exemplify, let us go back to the coffee cup on your desktop. First, we add thousands (or millions) of pictures into the dataset. Next, we ask humans be that you, me, or a click-farm somewhere around the world, to answer whether there is a coffee cup in each picture. After doing so, the algorithm will try to build a model such that the machine can accurately recognize a coffee cup in pictures when it sees one.
It is important to note that you can achieve artificial intelligence WITHOUT using machine learning, BUT that requires much much much more code of complex decision-trees and rules.
Deep learning: A technique to implement ML
Deep learning is an approach to machine learning that teach machines to do something that comes naturally to us humans; learning by example.
The concept of deep learning is inspired by the function and structure of our complex brains. Hence, deep learning is based on ANNs (Artificial Neural Networks), which essentially is algorithms that imitate the biological structure of our brains.
Artificial Neural Networks consist of “neurons” which have discrete layers and connections to other “neurons” within the ANN. Hence, the depth in “deep learning” usually refers to the number of hidden layers in the neural network, where deep networks have up towards 150 layers vs. traditional neural networks of 2-3 layers.
Some of the key use-cases of deep learning are for autonomous cars, medical equipment, industrial automation, defense, aerospace and in electronics such as the voice-control/assistants we all know: Siri, Alexa and the Google Assistant playing Stevie Wonder in my apartment as I am writing this.
In regards to deep learning and its relation to machine learning, it is important to note that deep learning is just one of many approaches to machine learning. Among others, machine learning approaches include inductive logic programming, clustering, Bayesian networks, decision-tree learning and reinforcement learning.
But this shit is old?
Some of today’s biggest buzzwords are NOT new concepts but they are now much more relevant than before due to the increased data volumes gathered in the digitalized and interconnected world (IoT).
Moreover, we have seen advancements in algorithms and huge improvements in computing power and storage in recent years.
But as with so many other emerging and booming technologies, corporations are slow to adapt paving the way for tech start-ups to invest in.