What is Machine Learning?
I decided to start this series on Machine Learning and AI because I believe that Machine Learning has been gaining incredible traction, and these algorithms will become so advanced that they may be able to make critical decisions such as in business, government and medicine that are lifechanging. I have taken several courses in Machine Learning and Data Science, and I would like to share my knowledge of Machine Learning.
Machine Learning is the science of getting computers to learn and act as humans do. Imagine a computer algorithm that constantly improves based on the data that it analyzes, and the interactions that a user has with the results that this algorithm generates.
Traditional computer algorithms are deterministic, i.e. given a series of inputs, they generate the same set of outputs. These algorithms don't need to be trained with massive amounts of data. Simply put, deterministic algorithms usually do not predict an output based on inputs, but run a set of commands (like if-statements or loops) to calculate an output.
Machine learning algorithms, on the other hand, are probabilistic, i.e. given a series of inputs, they generate outputs that the algorithm "thinks" are best based on the training that the algorithm got. Machine Learning algorithms are trained using historical data. The chances of a Machine Learning algorithm generating correct outputs usually goes up with the amount of historic data that is used to train the algorithm.
However, these two kinds of algorithms can sometimes be used hand-in-hand, where traditional algorithms can be used for preprocessing, then ML models can use the processed data for their training. This is extremely common in Computer Vision such as OpenCV, where raw video streams can have their exposure, saturation, and RGB values modified, then directly fed into the Machine Learning model.
There are 3 kinds of Machine Learning models: Supervised model, Unsupervised model, and Reinforcement model. Each algorithm has its own appropriate uses.
In a Supervised model, first an algorithm is "trained" using a series of labeled input. During the training process, the algorithm compares the output that it generates with the labels on the input. The algorithm makes appropriate modifications to itself during this learning process to ensure that the model fits appropriately to the training data. Once the algorithm is trained, it can then be used to process new input data to predict an output with some degree of confidence.
Supervised models are used in Classification problems. For example, take a toddler who is trying to sort all of his toys into two “classes”: a Teddy Bear pile and a Toy Train pile. After every decision that the toddler makes, his older, and more experienced, brother tells him whether he is right or not. The toddler will eventually learn how to effectively differentiate between the two kinds of toys. The supervised model for Classification works in a similar manner. The training data acts like the older, experienced brother that trains the algorithm to take the right classification decision.
Another type of Supervised model is Regression. Regression models are used to predict a continuous value. For example, predicting prices of a house given the features of the house like size, price etc. I will cover this in more detail in a future write-up.
An unsupervised model is the one in which the model captures patterns in an unlabeled dataset on its own. These algorithms discover hidden patterns or data groupings without the need for analyzing the training output. I will cover unsupervised models in more detail in a future article.
