Machine learning involves the input of data that creates an algorithm (“a process or set of rules to be followed in calculations or other problem-solving operations, especially by a computer”) that it uses to predict the outcome. Data for machine learning is comparable to experiences for humans. The more data a machine gets, the more it learns. Machine learning is our best attempt at mimicking what the brain does.
AI, ML, and DL
Often, Machine Learning is used interchangeably with Artificial Intelligence or Deep Learning, yet there is a subtle difference that should be clear:
The outside layer:
Artificial Intelligence (AI) is a broad term for a program that mimics human intelligence. This includes hardcoded systems that do exactly as they are programmed (the system is not learning on its own). The system does not veer from the path, it simply follows the code written by the developer to accomplish a task.
The inside layer:
Machine Learning (ML) is a subcategory of AI. It uses structured/labeled data to make decisions. It’s the developers’ responsibility to determine the relevant features to feed the algorithm to optimize its accuracy. ML then utilizes past experiences to gain a better understanding of future output, which may require human intervention. ML can make decisions that are not explicitly programmed.
Take the game of checkers for example: Recall that AI is the broad term and AI is mimicking human thought. You can program to simply play the game of checkers (i.e. “If I get to this spot on the checkerboard, I can then move left, right, or jump”). ML takes the process a step further where the program actually learns off the moves it’s made because it’s picking up on the data & establishes which move produces the greatest chance of winning. AI = action (“I’m making moves based on the rules of the game”), ML = reaction (“I’m learning to do things better even though I wasn’t explicitly taught…I’ve learned from experience that I can do better”).
The innermost layer:
Deep Learning (DL) is a subcategory of ML. It is more complex and requires less human intervention to learn from its errors. The algorithm is fed raw data and determines the relevant features needed to accurately predict the outcome. Deep learning requires less human interaction and more data, so deeper learning can occur (i.e. face recognition).
The human mind doesn’t see a person’s face and immediately recognize it; it’s only after you see a person’s face at least once that you can recognize it at a later time. Once you’ve had multiple exposures to that face, you can then begin to identify and recall its unique details. You eventually have enough built up experience about that person that you can adapt to the face, even if a small factor is changed (i.e. the person grows a beard, gets a new pair of glasses, or wears a different color of lipstick). Based on all the other features and experiences in the past, the brain can still recognize that person (i.e. my friend David has a beard now, but I still know it’s my friend David’s face).
Use cases
Gaming:
Keep gamers engaged longer by equipping computer characters with the ability to remember playing styles and learning from the gained experiences.
Security/Fraud:
Match a selfie with a government-issued ID to reassure your clients that their personal data is secure.
Forecasting:
Make more informed financial decisions based on the latest current events with minimal human intervention.
Marketing/Recommendations:
Customize your marketing and product recommendations based on the specific features of your customers.
Healthcare:
Increase the accuracy of your patient’s diagnosis.
Robotics:
Smart Cars