By Tech Brew Staff
less than 3 min read
Definition:
Machine learning is any type of system that can perform tasks without explicit instructions via an algorithm shaped by data. Almost all AI commonly discussed today is fundamentally based on machine learning. Deep learning is a subfield, and a neural network is a model of machine learning.
Rudimentary beginnings
In the early 1960s, early AI researcher Donald Michie created a mechanical “computer” of sorts using about 300 matchboxes. Each matchbox represented a different tic-tac-toe board configuration and contained colored beads, with each color representing a possible move. To play, a human would select the appropriate matchbox for the current board state, draw a bead at random, and make the corresponding move. After each game, the player removed beads that led to losing moves, or added extra beads for winning moves. Over many, many games, the collection of matchboxes (aided by its human player) thus “learned” to play tic-tac-toe by increasing the probability of a winning strategy.
That’s an early and simplistic example of one type of machine learning—reinforcement learning, where learning is induced by way of “rewards” for a desirable outcome, like a tic-tac-toe win. Machine learning algorithms are generally made up of variables that are endlessly tweaked based on relationships within training data. The more training data these equations are fed, the more accurate they become.
Supervised learning
A simple shape classification algorithm might be trained on the numeric pixel values of squares and triangles and their corresponding labels. It will adjust the variables that make up its algorithm based on this data until the system can accurately identify what shape is in a picture. This is called supervised learning, because it involves labeled datasets.
Imagine doing something similar, if much more complicated, with credit card transaction data. Feed the algorithm a huge dataset with the items, amounts, and various circumstances of card purchases and whether or not they were fraudulent. The algorithm will eventually learn to score transactions on the likelihood of fraud. Major credit card companies have been using systems like these for years.
Machine learning algorithms are used in image classification, facial recognition, recommendation engines, large language models, predictive analytics, self-driving cars, search engines, and many other modern technologies.
