What is machine learning?
In the modern world, you can often hear about machine learning, but what is machine learning?
Machine learning is one of the most dynamically developing areas in the field of information technology. Every year, significant changes take place in it and new technologies appear that change the way people work and their daily lives. In this article, we will look at several recent trends in machine learning that will affect the future of this field.
Thanks to machine learning, the programmer does not need to create instructions that take into account all possible problems and solutions. Instead, an algorithm is embedded in a computer or program that independently finds solutions using a comprehensive analysis of statistical data, identifying patterns and making predictions based on them.
The origins of machine learning technology based on data analysis go back to the 1950s, when the first programs for playing checkers were developed. Over the past decades, the basic principle has remained unchanged. However, due to the rapid growth of computing power of computers, the complexity of the identified patterns and the accuracy of forecasts have increased significantly, and the range of tasks solved using machine learning has significantly expanded.
To start the machine learning process, you first need to upload a dataset to your computer — a set of source data on which the algorithm will learn. For example, you can use photos of dogs and cats with labels indicating which class they belong to. After training, the program will be able to independently recognize objects in new images without labels. The training continues even after the forecasts are issued: the more data is analyzed, the more accurately the program recognizes the images.
With the help of machine learning, computers can recognize not only faces in photographs and drawings, but also landscapes, objects, text and numbers. As for the text, machine learning is actively used to check grammar in text editors and on phones. It takes into account not only the correct spelling of words, but also the context, shades of meaning and other linguistic aspects. Moreover, there is already software that can automatically write news articles on economic topics and sports, without human intervention.
Types of machine learning
Machine learning (machine learning, abbreviated ml) it is no longer an obscure and inaccessible technology. Today, it is in the spotlight and is used in many areas of life. But what new opportunities have appeared recently?
All tasks solved using machine learning (ML) can be divided into the following categories:
1. The task of regression is forecasting based on a sample of objects with different characteristics. The result should be a real number (for example, 2, 35, 76.454, etc.), as in the case of determining the price of an apartment, the value of a security in six months, the expected income of the store for the next month, or the quality of wine during blind testing.
2. The task of classification is to obtain a categorical answer based on a set of features. This is a finite number of answers, often in the "yes" or "no" format: for example, the presence of a cat in a photo, the recognition of a human face in an image, or the diagnosis of cancer in a patient.
3. The task of clustering is the distribution of data into groups. Examples include segmentation of mobile operator customers by solvency level or classification of space objects as planets, stars, black holes, etc.
4. The task of dimensionality reduction is to reduce a large number of features to a smaller number (usually 2-3) for ease of visualization (for example, data compression).
5. The task of detecting anomalies is to separate anomalies from standard cases. At first glance, it is similar to the task of classification, but there is an important difference: anomalies are rare phenomena, and there are either very few or no training examples to identify them, so classification methods do not work here. In practice, such a task is, for example, the detection of fraudulent activities with bank cards.
What tasks can machine learning be used for?
The purpose of machine learning (ml) is partial or complete automation of complex professional tasks in various fields of human activity.
Machine learning has applications in a wide range of fields:
- Speech recognition
- Image analysis
- Handwriting recognition
- Technical diagnostics
- Medical diagnostics
- Time series forecasting
- Bioinformatics
- Fraud detection
- Spam detection
- Classification of documents
- Technical analysis on the stock exchange
- Financial control
- Credit scoring
- Forecasting customer care
The scope of machine learning is constantly expanding. Widespread digitalization contributes to the accumulation of huge amounts of data in science, industry, business, transport, and healthcare. The tasks of forecasting, management, and decision-making that arise in this case are often reduced to case-based learning in order to train computer systems based on data to make predictions, make decisions, and complete tasks. In the past, when such data did not exist, these tasks were either not set at all, or were solved using fundamentally different methods.
Machine learning methods
Since machine learning was formed, on the one hand, from the science of neural networks, which was divided into network learning methods and various topologies of their architecture, and on the other hand, absorbed the methods of mathematical statistics, the following machine learning methods come from neural networks. The main types of neural networks, such as perceptron and multilayer perceptron (and their modifications), can be trained with a teacher, without a teacher and actively. However, some neural networks and most statistical methods can only be attributed to one of the learning methods. Therefore, classifying machine learning methods according to the method of learning, with regard to neural networks, it is more correct to talk about the classification of their learning algorithms, rather than belonging to a certain type.
Learning with a teacher — for each example, a pair of "situation, required solution" is set:
- Error correction method
- The method of error back propagation
Unsupervised learning — for each example, only a "situation" is set, and it is required to group objects into clusters using data on pairwise similarity of objects:
- Alpha system
- Gamma system
- The nearest neighbor method
Reinforcement learning — for each example there is a pair of "situation, decision":
- Genetic algorithm
Active learning differs in that the trained algorithm independently selects the next situation under study, for which the correct answer will be known.
What type of machine learning doesn't happen?
Any machine learning method must be based on algorithmic and mathematical principles, so there is no type of machine learning that does not have a theoretical basis. However, machine learning methods may vary in their effectiveness and versatility depending on the task and the source data. For example, it is impossible to create a universal algorithm capable of coping with any machine learning tasks without configuring or preprocessing data. Also, there is no single machine learning algorithm that would work equally effectively with different types of data, such as text, images, sound, etc.
Training and test samples
Understanding the concepts of training and test samples is critical to successfully train machine learning models and make accurate predictions.:
A training sample is a set of data on which the model is trained. It adjusts its weights and determines the dependencies between the input data and the target variable based on this data.
A test sample, on the other hand, is an independent set of data on which the model has not been trained, but which is used to assess its quality and accuracy of predictions.
The separation of data into training and test samples is necessary to assess the ability of the model to generalize to new data that was not present in the training sample. If the model is trained and tested on the same data, it can demonstrate high accuracy of predictions, but it does not cope well with new data. This phenomenon is called overfitting and leads to the inability of the model to effectively solve real problems with a variety of data.
To avoid overfitting and to test the generalization ability of the model, it is necessary to use a test sample to verify the accuracy of predictions on new data. The test sample should be representative of the entire dataset, but should not overlap with the training sample.
Cross-validation is based on the idea of dividing the data into several folds, training the model on several of them, and using the rest of the data to test the model. This procedure is repeated several times with different combinations of folds, and as a result, an estimate of the accuracy of the model is obtained for the entire data set.
Cross-validation allows you to evaluate the accuracy of a model on independent data and compare the quality of different models on the same data. This method also helps to reduce the influence of random factors, such as splitting into training and test samples, and improve the stability of the model.
