What is a Machine Learning Model – Understanding the Basics

Machine learning model icon on a green gradient background
In simple terms, a machine learning model is a mathematical model that is used to predict outcomes from data. The model is trained on a dataset, which is a collection of data that includes both the input data (such as features) and the output data (such as labels). The model is then able to make predictions on new data, which has not been seen before.

We need to know some definitions before we really grasp what a machine learning model is. A machine learning model is the working part of machine learning, which is a huge field that needs several articles in itself.

  • Artificial intelligence: it is a field of computer science; it is about making computers and electronics exhibit intelligent behaviors like taking sensory inputs, recognizing pattern, decision making, … etc.
  • Machine learning: it is yet a subfield of AI focusing on enabling the computers to “learn” intelligence; without explicitly hard-coding such behaviors. The term machine learning model is often used interchangeably with machine learning algorithms, but there are some nuances that need to be emphasized to use these terms right.
  • Machine learning algorithm: it is a way to find patterns in datasets mathematically; that means it is a kind of an abstract representation for the “learning” process.
  • Machine learning models: They are the actual implementations of ML algorithms as software that is able to recognize patterns and sometime make decisions (or predictions) based on the datasets.
An oversimplified analogy is ML algorithms are the definitions of functions, and ML models are the actual functions that find patterns.

ML models are two types, based on the training:

  1. Untrained models: they are the predecessor of working models; in our analogy, they are the forms of functions that their parameters are not known (say a quadratic function without knowing its coefficients).
  2. Trained models: they are the actual models that classify, predict or make decisions when you feed an unseen input; all based on the model’s knowledge gained from the learning data, in our analogy, they are functions with known coefficients, the functions are in the same form as untrained models, the coefficients are the “knowledge” of the model.

The process of finding the coefficients is called training the model, in formal terms, it is the process of finding the parameters that optimize the outcome of the model not only for the training data, but also for unseen data.

Machine Learning Model

First, a machine learning specialist looks at the data that need to be classified (or make predictions based on), then they decide what is the best form that the data need to be translated to before we feed it into the model; because the model only works in numbers, for example we can encode text with mathematical representations such as ASCII.

Then, the specialist chooses the best approach (an algorithm) to structure the model – based on the nature of the data, now we have the pretrained model.

Before training we should choose a uniform sample of the data, so don’t be unbiased, for example if we have a model that classifies images as a dog or a cat, or sample should contain different breeds of both, so our model can classify all breeds not just one.

Now the training takes place upon our training data, to get to our trained model, it should be able to classify images that it hasn’t seen before.

Based on the structure of the model; it can be categorized to many categories:

Supervised Machine Learning Models

The training data here is labeled, for example, in our dog-cat classifications example the pictures we feed in to the model should be labeled as either a dog or a cat, the labels work as a performance indicator of our model, and in some scenarios, it is used to adjust the parameters of our model.

word image 11728 1

Unsupervised Machine Learning Models

The training data here is unlabeled, so the model’s job is to find relations between the input data.

word image 11728 2

Reinforcement Learning Models

The best example of reinforcement models is models that play games, the model is set to play the game without telling the model the rules, then the model ventures the game’s world by itself, and it can evaluate how well it takes actions is based on the model’s score on the game, for example if it does something that adds to the score it keeps doing that thing, if it does something that lowers the score it avoids that.

word image 11728 3

Every one of the categories of models have many subcategories, and every subcategory is well suited for a specific application.

Conclusion

In this article, we learned what is artificial intelligence, machine learning, and machine learning models. We also we walked through machine learning model types and how a machine learning model models works.

0 Shares:
Subscribe
Notify of
guest
Receive notifications when your comment receives a reply. (Optional)
Your username will link to your website. (Optional)

0 Comments
Inline Feedbacks
View all comments
You May Also Like