Machine learning.pdf

Translated from Arabic to English - www.onlinedoctranslator.com Artificial intelligence/third stage Dr.. Hanaa Fathi Mahmoud Machine Learning:...

Translated from Arabic to English - www.onlinedoctranslator.com Artificial intelligence/third stage Dr.. Hanaa Fathi Mahmoud Machine Learning: It is one of the branches of artificial intelligence that is concerned with designing and developing algorithms and techniques that allow computers to have the property of “learning”. In general there are two levels of learning: Inductive: Inductive deduces general rules and judgments from big data. Deductive: Deductive learning is based on general judgments that are applied in specific examples. The basic task of machine learning is to extract valuable information from data, and therefore it is very close to data mining, statistics, and theoretical informatics. Machine learning is used in many fields from engineering to medicine Key components of machine learning If we were to summarize all the goals behind the field of machine learning, the only goal would be to predict certain results based on the input data (i.e., learning from the input data). All machine learning tasks can be represented in this way. Any system that uses machine learning will need three main components: 1.data)(Data The greater the diversity of the data (sometimes calledWith samples) aggregated, the relatively easier the task of finding relevant patterns and predicting the outcome. For example, to detect spam emails, you must obtain spam. On samples of these letters For the purpose of predicting changes in stock prices? one should look for price records Stocks. For the purpose of knowing what are the user preferences? His activities on Facebook must be analyzed, the more diverse the data, the better the result. Sometimes tens of thousands of data records are the minimum to conclude a certain piece of information. In others, millions of samples are needed. There are two main ways to obtain data: - Manual method. 1 Al-Tariba College of Pure Sciences computer section  Artificial intelligence/third stage Dr.. Hanaa Fathi Mahmoud - Automated method. Manually collected data has much fewer errors than its automated counterpart, but it takes longer to compile and is generally more expensive. The automated method is cheaper, as all that has to be done is to collect everything that can be found and hope that the quality of this data is acceptable. A good set of data is usually difficult to obtain, which is calledData setThese collections are very important, and a high-quality data set is in fact a real treasure for its owner, to the point that companies can sometimes reveal their algorithms, but they rarely reveal their data sets. (Dataset) 2. Features Also known asTransactionsOr (Parameters)VariablesWhich can express the distance traveled by cars, the gender of the user, the stock price, or the frequency of a specific word in the text. In other words, these are the features that the machine should look at (Variables). When data is stored in tables, it's simple - the attributes are the column names. But what if you have 100 gigabytes of cat photos? Of course, we cannot consider every pixel a feature. This is why choosing the right features usually takes longer than any other step in building a machine learning- based system. This is also the main source of errors. Therefore human choices are always subjective. They only choose features that they like or that they find “most important.” 3. Algorithms This is the easiest and most obvious part. Any problem can be solved in different ways. However, the method you choose will affect accuracySampleThe final product, its performance and size. There is one important nuance: if the data is bad not even the best algorithm out there will help you. Sometimes it is referred to by the term “bad income will lead to bad results”. So don't care too much about the accuracy rate, and try to get more data to start with. Types of machine learning methods Machine learning algorithms are classified into several types: 2 Al-Tariba College of Pure Sciences computer section  Artificial intelligence/third stage Dr.. Hanaa Fathi Mahmoud Supervised Learning:It is one of the most popular types of learning Automation is based on the presence of correct data at the time of learning, so that this data constitutes real examples from which the model can learn. Where the machine is trained using inputs whose outputs are known in advance. During supervised learning training, systems are exposed to large amounts of previously labeled data, for example images of handwritten numbers explaining the number they correspond to. By looking at more than one example, i.e. more than one handwritten piece of paper, the system or machine in the supervised system will learn to recognize the groups of pixels and shapes associated with each number, and eventually be able to recognize handwritten numbers, so it will be able to reliably distinguish between the numbers 9 and 4 or 6 and 8. The goal of entering data based on certain classifications is to find a relationship between the input Or does he have fins? Each True or False answer will delete a large number of options using the exclusion system until a certain point is reached that is the most accurate and correct. Supervised learning depends primarily, as is clear in the previous example, on the feature that distinguishes each entry, so it can be applied to many things such as length. Or speed, or something else, and this is considered only one of the machine learning algorithms through supervised learning. However, training these systems typically requires massive amounts of labeled data, with some systems having to be exposed to millions of examples to master a task. It has other names, such as Predictive Learning, one of its most famous and important types: - Classification This type of supervised machine learning is the most commonly used type, where the input is classified into two or more types. The purpose of this learning is to classify any model input into one of the previously known categories of the model, i.e. the types that were previously trained on, such as the process of recognizing faces, as well as classifying types of machine learning. Fruits or letter classification. - Regression 3 Al-Tariba College of Pure Sciences computer section  Artificial intelligence/third stage Dr.. Hanaa Fathi Mahmoud It is similar to the first type (classification), except that it gives results with continuous values instead of classifying them into separate categories, and is often used in forecasting, such as forecasting stock prices and forecasting temperatures based on climate information and aerial images. Unsupervised Learning:It is learning that results from the presence of data without Its correct types are that the input data (inputs) are available, but the outputs of the system are not known in advance. It is also called descriptive learning. In this type the algorithm learns from clear examples without any associated response, leaving the algorithm to identify data patterns on its own. This type of algorithm tends to restructure data into something else, such as new features that may represent a new class or series of unrelated values. They are very useful in providing humans with insights into the meaning of data and new useful inputs for supervised machine learning algorithms. One of the most famous types of unsupervised learning is cluster analysisFor Clustering It is similar to the methods humans use to know that certain objects or events are from the same class, such as noting the degree of similarity between objects. Some of the recommendation systems you find on the web in the form of marketing automation are based on this type of learning. This type of machine learning algorithms sorts (groups) the inputs to the system into previously unknown groups. This type of learning is used in e-commerce applications and websites, where users are grouped into groups based on the purchases they have previously made by studying their online browsing behavior. Later, emails are sent or advertisements appear directed to these groups that have been grouped based on The type of materials purchased or sites previously visited. Other examples include learning the movements of a person standing in front of a camera that records his movements and studying the reactions to each movement, so that the system can later recognize these movements and link them to appropriate reactions. 4 Al-Tariba College of Pure Sciences computer section  Artificial intelligence/third stage Dr.. Hanaa Fathi Mahmoud Semi-supervised learning:It is when there is data with its types Correct, but limited or incomplete. As the name suggests, this method is a combination of approaches between supervised and unsupervised learning. This technique relies on using a small amount of labeled and labeled data and a large amount of unlabeled data to train the systems. The labeled data is used to partially train a machine model, and this partially trained model is then used to label the unlabeled data, a process called pseudo-labelling. The resulting mixture of labeled and unlabeled data is then trained. The importance of large sets of pre-labeled data to feed learning systems may decrease over time, as semi-supervised learning becomes more important. Reinforcement learning:In this type you learn how to act At a specific event by giving signals symbolizing reward or punishment based on the current behavior. In which the machine interacts with the environment and builds its experiences based on this interaction, reinforcement learning is considered a promising type of education that may have a major role in solving complex issues in the future. Reinforcement learning should be used when the environment is unknown, otherwise it will require a lot of computational resources without any real benefit to the learning process. It teaches how to act in the event of a specific event by giving signals that symbolize reward or punishment based on the current behavior. 5 Al-Tariba College of Pure Sciences computer section  Artificial intelligence/third stage Dr.. Hanaa Fathi Mahmoud Augmented learning works under the principle of feedback learning or reaction and response. Perhaps you did not understand the meaning... So let us imagine that you provided the machine’s system with a picture of a dog and asked it to determine what that thing is. He identified it on the basis that it is a cat, perhaps because there are similarities between them, as if they had 4 Feet, or because the size is approximately the same, or for any reason, the machine produced an incorrect result with a completely negative accuracy. After that, I told the system that the image was of a dog. It is assumed that if it follows the reinforcement learning method, the system will learn that, and therefore any other image of a dog, the machine will be able to classify it correctly. True, you can apply that example to any system that operates on the principle of response and feedback. To understand reinforcement learning, you can imagine how someone might learn to play a very old computer game for the first time, when they are not familiar with the rules or how to control the game. While this person may be a beginner and it's their first time playing, ultimately, by looking at the relationship between the buttons they press, what happens on the screen and the outcome within the game, their performance will get better and better as time goes on. An example of reinforcement learning is Google DeepMind's, which has outperformed humans in a wide range of ancient video games. The system is fed specifications of every inch of each game and determines various information about the state of the game, such as the distance between objects on the screen. It then looks at how the game state and the steps you take in the game relate to the score you achieve. By playing for a significant period of time on the same game, the system eventually builds a model of actions and steps that will increase the outcome of successfully finishing the game better than a human.. Machine learning applications Now, most, if not all, applications use machine learning in one way or another. Examples of this Medical and diagnostic fields Fields of security and protection Commercial areas Natural language processing – natural language processing Syntactic pattern recognition Search engines – search engines 6 Al-Tariba College of Pure Sciences computer section  Artificial intelligence/third stage Dr.. Hanaa Fathi Mahmoud Medical diagnostics, bioinformatics, cheminformatics, classification of DNA sequences Speech recognition Handwriting recognition Object recognition Computer vision Robot locomotion 7 Al-Tariba College of Pure Sciences computer section