Machine Learning Algorithms
The Machine Learning algorithm is the set of hypotheses taken at the start before the practice starts with information from the real world. When we say the Linear Regression model, it means a set of functions that define similar characteristics as defined by Linear Regression, and from those function sets we will choose one function that best fits the training data.
Overview of Machine Learning Algorithms
Broadly, there are four types of Machine Learning is existing which is Supervised ML, the second one is Unsupervised Machine Learning, Sami Supervised Machine Learning, and finally, Reinforcement ML, which is used in different types of models.
In include, can say models that are methods or procedures used to accomplish a mission or to solve a problem, While Models are well-defined calculations resulting from a model that takes some value or set of values as an input and produces some value or set of values as an output. We will introduce some of them;
Supervised Machine Learning Algorithm (SML A)
Supervised Machine Learning also is known as Supervised Learning, because the process of a model learning all from a training dataset can be said, which is a teacher or leader supervening the process of learning.
It is the search for models that give rise to general hypotheses from externally supplied instances, which then predict future instances. Supervised classification is one of the tasks that intelligent systems most frequently perform.
Machine learning models are grouped into a taxonomy centered on the optimal performance of the model. Supervised training produces a method that maps inputs to the appropriate outputs.
The classification problem is one basic formulation of the supervised learning task, the learner should learn (to approximate the action of) a function that converts a vector into one of many categories by looking at several input-output function examples. Process of learning a set of orders from an instance of the dataset which used Inductive machine learning, or can be said in general propose, create a classifier that can be used from new instances to generalize. The figure describes the process of applying supervised ML to a real-world problem.
Overview of Supervised Machine Learning Process
Every SML classification needs a structure like this figure when faced with a problem. When wants a good result for our classification, it is necessary to follow these rules step by step. The base case is when the dataset train then needs to select a model, which should careful to select the models to work properly.
Now we can have a brief introduction about the Supervised Machine Learning model with its Model. The following includes supervised machine learning models dealing more with classification such as (Linear Classifiers, Naive Bayes Classifier, Logistic Regression, Perceptron, Boosting, Neural networks, Support Vector Machine, Quadratic Classifiers, Random Forest (RF), K-Means Clustering, Decision Tree, Bayesian Networks and so on).
Some models of SML are accurate but a litter slow, such as Random Forest (RF), Gradient Boosting Tree (it is Random Forest-like, but easier to overfit), neural network (needs a lot of data points), and so on. And some of them act faster but are not accurate such as linear regression, logistic regression, Decision Tree, and such models.
Unsupervised Machine Learning Algorithm (UML A)
Unsupervised Machine Learning also is known as self-organization, because it is a type of self-organized hidden learning which is helped previews unknown patterns without a labeled dataset. Unsupervised learning is the training of the models using a dataset that is neither labeled and neither classified, and allows the model to act on Datasets without any guidance.
A UML system may group unsorted information according to similarities and differences in unsupervised learning, although no categories are available. Unsupervised Learning models can perform processing tasks that are more complex than the Supervised Learning Model.
Generative learning models are often associated with unsupervised learning, although they may also use a retrieval-based approach (most commonly associated with supervised learning). Among the systems that can use either supervised or unsupervised learning, approaches are facial recognition programs, chatbots, self-driving cars, robots, expert systems, and so on.
Unsupervised learning models divide the data into an unlabeled data set based on the hidden features underlying the data. There is no way to evaluate the result (a key difference in supervised learning models) because there are no tags.
Overview of Unsupervised Machine Learning Process.
UML takes the dataset as input then interprets the dataset after interpretation applied some model’s role to take the best result and process the role, finally output the result of UML. However, which is, the set of labels is missing from the input dataset. That is, the dataset being existed without any labels.
UML used different models to solve a problem in the best case, such as clustering, hierarchical clustering, k-means, Deep Belief Nets, mixture models, OPTICS model, Anomaly detection, Self-organizing map, Local Outlier Factor, Neural Networks, Generative Adversarial Networks, and such models.
Semi-Supervised Machine Learning Algorithm (SSML A)
Semi-Supervised Learning is another aspect of the Machine Learning model, which also allow machine learning to train labeled and unlabeled dataset. Usually, a few numbers of the labeled and large amounts of unlabeled Datasets can train by semi-supervised Machine learning.
SSML is halfway between supervised and unsupervised learning, we can say the combination of supervised learning and unsupervised learning is called semi-supervised learning. Because the model of semi-supervised learning examined in this research has the same goal of generating a predictor as supervised learning, it is generally applicable to any problems where supervised learning can be applied.
It seems that SSML actually began when the problem of estimating the Fisher linear discriminatory policy with unlabeled data was considered. More specifically, the setting was in the case that each class-conditional density is Gaussian with an identical covariance matrix.
The model's probability is then maximized by using the labeled and unlabeled records with the support of an iterative model like the expectation-maximization (EM) model. Maybe with a mixture of Gaussian, the use of a mixture of multinomial distributions estimated with labeled and unlabeled data has been investigated. Instead of a mixture of Gaussian, the use of a mixture of multinomial distributions estimated with labeled and unlabeled data has been investigated. Later, Miller and Uyar extended this one component per class setting to several components per class and further generalized it.
The main goal of SSML is to draw back the overcome of both SML and UML, which is SML always need a dataset which is has labeled other ways it cannot work properly and need a hang amount of training data to classify the test data, it can take more time and cast affection, but the Unsupervised Machine Learning does not need labels in the dataset, but it is a litter complex. The semi-Supervised Machine Learning Algorithm removes these problems, it can work properly with both typed datasets.
Further division of SSM L is Semi-Supervised Classification and the other one is Semi-Supervised clustering (SSC is like a supervised approach that required more training data to classify test data). However, SSML is better than SML and UML in case of preparation for creating a model to train the datasets.
Some models of SSM LA such as Self-Training (image categorization), Generative Models (generate models by itself as a required classifier), Semi-supervised support vector machine (S3VMs), Graph-Based models (text classification). Also, has been applied Semi-Supervised Learning in Real-time network traffic classification, natural language processing (machine translation, word sense disambiguation, named entity classification, document categorization, sentiment analysis), bioinformatics (protein function prediction), cognitive psychology, computer vision (image segmentation, object recognition), and also has another usage.
About Reinforcement Machine Learning (RML) I will write in my next article, if you want more about Machine Learning you should subscribe.
For more Details:
Email: [email protected]
GitHub: https://github.com/aketman
You must be logged in to post a comment.