- fevereiro 15, 2019
- Posted by: Equipe Ustore
- Category: Computação em Nuvem, Internacional
Artigo em inglês
Machine learning, sometimes called computational intelligence, has broken down barriers in recent years and has made significant advances in a number of areas, such as robotics, machine translation, social networking, e-commerce, and even in areas such as medicine and healthcare. Machine Learning is an area of AI with a goal to develop computational techniques on learning as well as the construction of systems capable of acquiring knowledge automatically.
A learning system is a computer program that makes decisions based on accumulated experiences through the successful solution of past problems. Despite the short definition, there are numerous different learning algorithms and the area is one of the hottest in the field of computing, with several new techniques and algorithms being published regularly.
Machine Learning vs. Artificial Intelligence
Many people think machine learning and artificial intelligence mean the same thing, but that’s not quite accurate. There are several defininitions of Artificial intelligence, which is a broad concept that includes machine learning. One widely accepted definition is that artificial intelligence consists of computational mechanisms that rely on human behavior to solve problems. In other words, technology makes the computer “think” like a person to perform tasks.
Humans are able to analyze data, find patterns or trends in them, make more informed analysis from there, and then use conclusions to make decisions. In a sense, artificial intelligence follows this same principle. Usually, the more we perform a task, the more skillful we become. This is a result of our ability to learn. Frequent repetition or execution of related procedures works as a training for us. Something similar occurs in artificial intelligence systems: data available publicly or recorded on proprietary platforms serve as training for artificial intelligence algorithms.
How is this training done? There is no way, there are several algorithms for this purpose. It all depends on the application and the organizations or people that are behind them. Here, what matters most is knowing that it is at this point that machine learning makes sense.
What is Machine Learning?
Machine learning is also a concept with several definitions, but at its core, machine learning is a system that can modify its behavior autonomously based on its own experience. The human interference here is minimal. Such a behavioral modification basically consists of establishing logical rules that aim to improve the performance of a task or, depending on the application, to make the most appropriate decision for the context. These rules are generated based on pattern recognition within the analyzed data.
For example, is a person types the word “brave” into a search engine, the service needs to analyze a series of parameters to decide whether to display results equivalent to infuriated or brave, two possible meanings. Among the numerous parameters available is user search history: if minutes before he has looked for courage, for example, the second meaning is most likely. This is a very simple example, but it illustrates some important aspects of machine learning.
It is important that systems make analyzes based on a significant amount of data, a criterion that searchers have to spare because of the millions of accesses that they receive and that, consequently, serve as training.
Another aspect is the constant data entry that favors the identification of new standards. Suppose the word “brave” becomes a slang term associated with a cultural movement. With machine learning, the search engine will be able to identify patterns that point to the new meaning of the term and, after some time, will be able to consider it in the search results.
There are several approaches to machine learning. A well-known one is called “deep learning,” where large amounts of data are treated from several layers of artificial neural networks – algorithms inspired by the structure of brain neurons that solve very complex problems, such as object recognition in images.
Examples of Machine Learning
The use of machine learning is evolving into diverse applications and many technological resources that we have today are based on AI and machine learning.
- Autonomous Database – With the help of machine learning, autonomous databases handle several tasks previously performed by an administrator (DBA), allowing those professionals to take care of other activities, reducing the risk of application unavailability due to human error.
- Combating Fraud in Payment Systems – Various attempts at credit card fraud and other payment methods are generated every second. Machine learning has allowed anti-fraud systems to identify most of these before they are successful.
- Translation of Texts – A translation can never be made to the lette – context, regional expressions and other parameters must be taken into account. Thanks to machine learning, automatic translators are getting more and more accurate.
- Content Recommendation – Video and audio streaming platforms use machine learning to analyze the history of content viewed or rejected by users to give them recommendations consistent with their tastes.
- Marketing and Sales – Websites that recommend products and services based on previous purchases are using machine learning to analyze purchase histories and promote other items customers may be interested in. This ability to capture data, analyze it and use it to customize the shopping experience or implement a marketing campaign is the future of retail.
- Transportation – Analyzing data to identify patterns and trends is essential for the transportation industry, which depends on developing more efficient routes and predicting potential problems to increase reliability and profitability. The machine learning data modeling and analysis aspects are important tools for carriers, public transportation, and other organizations in the industry.
- Oil and Gas – Machine learning is helping discover new sources of energy, analyze minerals in soil, predict faults in refinery sensors, and accelerate the distribution of oil to make it more efficient and economical. The number of machine learning applications in this industry is vast and continues to grow.
- Healthcare – Machine learning is a growing trend in healthcare, thanks to the emergence of wearable devices and sensors that enable healthcare professionals to access patient data in real time. The technology can also help medical experts analyze data to identify trends or alerts, leading to improved diagnoses and treatments.
Methods Used in Machine Learning
Two of the most widely adopted methods of machine learning are supervised learning and unsupervised learning, but they are not the only ones.
Supervised Learning Algorithms are trained by means of labeled examples, as an input in which the desired output is known. For example, equipment might have data points labeled “F” (failure) or “E” (performs). The learning algorithm receives a set of inputs along with the corresponding correct outputs, and learns by comparing the actual output with the correct outputs to find errors. It then modifies the settlement model. Through methods such as classification, regression and gradient boosting, supervised learning uses standards to predict label values ??in additional non-labeled data. Supervised learning is commonly employed in applications where historical data predicts likely future events. For example, it can anticipate when credit card transactions are likely to be fraudulent or which insured tends to claim their policy.
Non-Supervised Learning is used against data that does not have historical labels. The “right answer” is not reported to the system. The algorithm must find out what is being shown. The goal is to explore the data and find some structure within them. Unsupervised learning works well with transactional data. For example, it can identify customer segments with similar attributes that can then be treated similarly in marketing campaigns; or it can find the key attributes that separate distinct customer segments. Popular techniques include self-organizing maps, proximity mapping, k-means grouping, and decomposition into singular values. These algorithms are also used to segment text topics, recommend items, and identify discrepant points in the data.
Semi-Supervised Learning is used for the same applications as supervised learning, but handles both labeled and unlabelled data for training – usually a small amount of data labeled with a large amount of unlabeled data (because data without labels is cheaper and requires less effort to be acquired). This type of learning can be employed with methods like classification, regression and prediction. Semi-supervised learning is useful when the cost associated with labeling is too high to enable a fully labeled training process. Basic examples include identifying a person’s face on a webcam.
Reinforcement Learning – is commonly used in robotics, gaming and navigation. With it, the algorithm discovers, through trial and error, which actions yield the greater rewards. This type of learning has three main components: the agent (the learner or decision maker), the environment (everything the agent interacts with) and actions (what the agent can do). The goal is for the agent to choose actions that maximize the expected reward over a given period of time. The agent will reach the goal much faster if he follows a good policy. So the focus of reinforcement learning is figuring out the best policy.
What are the Differences between Data Mining, Machine Learning and Deep Learning?
While all of these methods have the same goal, extracting insights, patterns, and relationships that can be used in decision making, they have different approaches and capabilities.
Data Mining can be considered a superset of many different methods for extracting insights from data. It may involve traditional statistical methods and machine learning. Data mining applies methods from several areas to identify previously unknown patterns in the data. This may include statistical algorithms, machine learning, text analysis, time series analysis, and other areas of analytics. Data mining also includes the study and practice of data storage and manipulation.
With Machine Learning, the objective is to understand the structure of the data – to fit theoretical distributions in well understood data. Thus, there is a theory behind statistical models that is mathematically proven, but this requires that the data also meet certain assumptions. Machine learning has been developed from the ability to use computers to examine the structure of data, even if we do not know what this structure looks like. The test for a machine learning model is a validation error in new data and not a theoretical test that proves a null hypothesis. As machine learning generally uses an iterative approach to learning from data, learning can be easily automated. The steps are performed through the data until a robust standard is found.
Deep Learninglearning combines advances in computing power and special types of neural networks to learn complicated patterns in large amounts of data. Deep learning techniques are the most advanced today to identify objects in pictures and words in sounds. Researchers are trying to apply successes in pattern recognition to more complex tasks, such as machine translation, medical diagnostics, and a host of other social and corporate issues.
Although the concepts of artificial intelligence and machine learning emerged long ago, they are starting to become part of mainstraem applications. But, we’re still only at the beginning. If they are useful and impressive today, they will be even more effective when they are better trained and improved.
Source: TechZone 360
Author: Lenildo Morais, Project Manager at Ustore, a leader in the Brazilian cloud storage market, based at the Porto Digital technology campus in Recife. He is also a Researcher at the ASSERT Research Technologies Laboratory – Advanced System and Software Engineering. Lenildo has experience in Software Engineering, working in the following research topics: software quality, methodologies and processes of software development, requirements engineering, software testing and project management software. He earned a Master in Computer Science at the Center of Informatics of the Federal University of Pernambuco.