The goal of this article is to provide a more detailed overview of the main types of machine learning models. It is very helpful for those who are not familiar with machine learning and want to get an overview of machine learning.
Main types of machine learning models
Simply put, machine learning is used to make predictions about future data based on past and current data. This is used in many real cases, from recommending films to watch to identifying types of cancer cells. The process of building an ML model can be quite complex and must be done so that it works for the particular problem that it should be dealing with.
Where data is available, machine learning can be used to solve a problem, share knowledge, and for many other purposes. Note, however, that machine learning models cannot use a single type of algorithm for all cases. Different categories of algorithms are available, which have been developed for handling different data and solving different problems using different approaches. The input provided, the task performed and the expenditure delivered for each class differ greatly.
A distinction is made between the following main types of machine learning, which are dealt with in detail in this article:
- Supervised learning
- Unsupervised learning
- Empowering learning.
Supervised learning
In order to understand supervised learning algorithms, let us first consider a pictorial example:
There are two ships, Ship A and Ship B, and both carry a variety of cars. While all cars on ship A are labeled in a database, the cars on ship B contain no labels. Now imagine that someone is given the task of identifying each car on Ship B and labeling it in the database. He can do this by developing a model based on Ship A data and then applying it to Ship B data. For example, if he learns from Ship A that cars with four wheels and more than four seats are called SUVs, he instructs the model to identify such cars on Ship B as SUVs based on this information.
This is a brief, pictorial representation of the supervised learning model. It is a type of machine learning, in which a solution to the current problem is found on the basis of the data already available.
Let us now represent this technically. Supervised learning is a type of ML algorithm in which a model or a function is created on the basis of the paired input-output data in the training data and maps the input of the test data to their respective output. Training data or a training data record is understood to mean a data record with labeled data and test data or the test data record is a set of entries without labels. In short, supervised learning can be seen as a type of machine learning that requires a teacher, technically training data.
Supervised learning can be roughly divided into two types: regression and classification. While the model of supervised learning uses numerical (or continuous) values in regression, the classification uses categorical values, which are usually binary values (0 or 1).
There are two key problems with supervised learning:
- Since these models are created with distortions and variance parameters, there are some problems here. First, distortions can result in errors. This can be explained by the fact that they are not based on the data itself, but on assumptions about the data. Second, variance, especially high variance, can make models work well on training data and poorly on test data.
- Noise in the data: Errors in the data can lead to data inconsistencies.
Supervised learning is widespread and includes the following algorithms:
- Support vector machines
- Linear regression
- Logistic regression
- Naive Bayes
- discriminant function
- decision trees
- k-Nearest Neighbor algorithm
- Neural networks (multilayer perceptron)
- Similarity learning.
Unsupervised learning
To understand unsupervised learning algorithms, let’s consider another example:
Suppose that instead of two ships, Ship A and Ship B, there is only one ship. And instead of cars, it transports several types of fruit and you do not know which types of fruit they are. To solve this problem, you should divide all fruits on the ship into groups according to their physical properties. For example, you can first divide by size, then by color, then by shape, and so on and so on. This is a visual representation of how the unsupervised learning models work.
Now let’s try to represent this technically. Unsupervised learning is a type of machine learning in which conclusions are drawn from a data set without labels. One of the most popular categories of unsupervised learning is clustering. It can be viewed as a process in which the process of exploratory data analysis is performed to find hidden patterns or groupings in data. In other words, unsupervised learning algorithms are models that learn from the test data themselves, as opposed to models of supervised learning that learn from a teacher, ie training data.
The most common algorithms used by unsupervised learning models are:
Clustering models
- Hierarchical clustering
- K-means
- mixing models
- DBSCAN ( Density-Based Spatial Clustering of Applications with Noise, for example: density-based spatial cluster analysis with noise)
- OPTICS algorithm (Ordering Points To Identify the Clustering Structure, for example: order points to identify the cluster structure)
Anomaly Detection Models
- LOF: local outlier factor ( about: Local outlier factor )
Neural network models
- car encoder
- Deep Belief Networks
- Hebrew learning
- GAN: Generative Adversarial Networks ( about: generating opposing networks) .
Empowering learning
Models of reinforcing learning can be compared to training a dog. That is, the way a dog is trained and you are always rewarded with a treat when you complete a target task. In these algorithms, the dog acts as an agent and dog treats as the reward. Note that there is an action or a target task to be performed. In addition, the environment and the interpreter are among the two important elements of these models. Accordingly, they can be understood in the place where the dog performs the target tasks and the person who commands the dog and rewards him with goodies.
How can algorithms of reinforcing learning be technically explained? They are the type of ML model in which tasks are performed by an agent in a specific environment. In this model, the agent is given either a reward or a penalty for each task performed. As the name suggests, it is about making continuous improvements based on certain rules. The reinforcing learning algorithm differs from other machine learning approaches in that it does not tell it how a task should be performed, but must learn by itself.
Because the reinforcement learning models require a lot of data, they work where there is readily available data. The typical use cases are gameplay and robot scenarios. For example, reinforcing learning was used in board games such as backgammon, checkers and chess. To test the results of the empowering learning models, an agent’s teacher is tested against a human.
The main algorithms for reinforcing learning include:
- Monte Carlo
- Q-Learning
- State Action Reward State Action (SARSA)
- Deep Q Network (DQN)
- Asynchronous Actor-Critic Agent (A3C)
- Deep Deterministic Policy Gradient (DDPG)
- Normalized Advantage Functions ( NAF ).
Conclusions
Machine learning can be divided into three main types: supervised learning, unsupervised learning, and empowering learning. For supervised learning models, the label of test data can be predicted by training a model based on the label of training data. Depending on the type of data used, supervised learning models can be divided into regression and classification models. In unsupervised learning models, hidden patterns and groupings are found in the data without the help of training data. Finally, reinforcement learning models can be seen as the models in which an involved agent performs the tasks in an environment based on the rewards.
If you would like to learn more about the most important types of machine learning, you can contact Tech By Light team by email or Q&A .
Leave a Reply
You must be logged in to post a comment.