There seems to be much confusion among the ranks of the untrained when it comes to an understanding of some basic IT concepts. This is especially true when looking at Artificial Intelligence (AI) and the misuse of nomenclature such as Machine Learning (ML), and Deep Learning (DL). In this article, I will try to brush away the mists of confusion, and present the case for each specific data related technologies.
First of all, you need to understand that all three are related and sit within a specific hierarchy.
To properly understand the differences between these three technologies, let’s first look at how they stack together in terms of hierarchy.
AI is the core concept or field of study that houses within it ML, and DL is a subset of ML that is used on neural networking.
As you can see, it’s quite simple.
Artificial Intelligence is a method or set of algorithms designed to enable a Turing Machine (computer) to reach decisions based on data without human intervention. How the data is collected by the machine and how it is assimilated and extrapolated is part of the AI algorithms. Essentially a true AI would be able to collect data including and beyond the five senses that humans have, and this data would also be analyzed using experience, as a subset of analysis.
Machine learning is a subset of AI. This is a software-driven process where the machine is given predictive and trend analysis models to use to enhance projection and forecasting, which can then be used to predict outcomes and tested when the situations for the outcomes occur.
Deep learning is targeted machine learning technique that manages big data; it grows in power as it consumed more information, and builds up hierarchical networks that create “mind maps” of the information. As such, DL is all about creating structure out of data.
DL was created to meet the need for neural pathway networking methodologies, in simple words, scientists needed a method that worked in a similar fashion to the human mind.
Now that we have a basic picture of the three different concepts let see how they mesh together.
Creating a Car from AI
Let’s consider that we live in a car-less world, and we are tasked to create a means of transport that is both efficient and viable. The basic approach would be to utilize trial by error approach and go through multiple theories, models, and tests before a final prototype could be reached.
AI works in the very same manner; AI starts off by testing itself, it is essentially a binary structure where success and failure are relegated to statistical probabilities., with experience from previous results recorded for future reference.
In other words, AI needs to build up a memory of data, much like humans do, to know what is right and what is not, it also creates probability outcomes, just like we do, where risk is calculated. The decision-making process of whether to perform a specific action or not is also algorithm based, and with some AI machines, control over these factors becomes a part of the whole cyclic, learn by mistake and adjust the process.
An example of this process is seen in SIRI and Alexa, as well as in Waymo and Uber AV AI software.
ML models are now being used to predict with a relevant degree of certainty the outcome of currency exchange rates. The larger the data asset (big data) and the more trading that feeds the data into the ML, enables ML to predict with a greater level of confidence.
Pattern recognition is a subset of ML, it is DL, and in DL the machine is taught to recognize patterns using hierarchical structures. DL basically compares data with an existing data set and as such, is basically a perfect pattern and trend recognition program.
To remove confusion, just remember that the hierarchy where AI contains ML that uses DL. If you would like a head-start and learn any of these technologies, check out Edureka’s AI and Deep Learning course and get started with these technologies.