MACHINE LEARNING
In few words, machine learning is the computer area that is responsible for computers to learn.
But it is not only about learning, it is about creating the rules to perform actions that humans could be taught. So for example I could teach a child to differentiate between a dog and a cat. Also, you could teach a computer to do the same without having to execute the action or schedule that action in advance. Thus and returning to the previous example, to teach a child that he is a dog and that he is a cat, he could start with descriptions of each species, for example a dog barks and a cat meows, so for the child to recognize these differences, he would have I have to show you many examples so that you can 'train' and get it right when you select the correct species.
So from this learning through experience, the important question is: when is it considered that a child is “learning”? When the ability or skill that is acquired was not present in their birth traits.
Thus, transferring our example to computer learning, what is sought is to teach machine tools that allow them to execute actions if these actions are explicitly scytho by researchers; that is, without it being necessary to remotely control the machine, but rather it is this in an 'autonomous' way that learns to carry out these actions to achieve the objective that is sought to teach, for example, to choose the right dog instead of the cat or the reverse.
Another very important aspect to highlight is that machine learning is closely related to pattern recognition (a science that deals with engineering, computing and mathematical processes related to physical or abstract objects, with the purpose of extracting information that allows establishing properties from among sets of said objects. This area of knowledge is also called pattern reading, figure identification and shape recognition, since it consists of the recognition of signal patterns).
In this way, machine learning can be seen as a way to automate some parts of the scientific method using mathematical methods. Automatic learning has a wide spectrum of applications, including search engines, for example. Thus, when you type a word on the web and the result shows you different links to internet pages with information related to the word entered, the effectiveness of this search and its efficiency lie in the training received by the model in charge of executing the search. and return related results.
Another concrete example is speech and written language recognition, so when you play a video on any platform on the web and activate the subtitles, a mathematical model is executing its own rules to get it right when transcribing the audio or translating the subtitles which has the audio file incorporated by default. Repeat after repetition, the model will improve its performance through experience, gradually acquiring a skill that it did not always have.
Machine learning is also widely used in the field of robotics to automate processes performed by machines and again, to make machines learn from experience.
Now, there are three models within machine learning: i) supervised learning: in this case the technique consists of deducing a function from training data, that is, input and output data or expected result. The objective in this case is to create a function capable of predicting the value corresponding to any data that has been entered into the model based on the data that has already been provided previously; that is, the training data. The method in this case is to generalize from the training data and obtain the results for situations not seen previously, on second place, we can find the ii) unsupervised learning: in second place we find unsupervised learning, which is a method of machine learning that is based on observations. The main difference with supervised learning is that in the former there is no prior knowledge. Their main characteristics are: they do not need an external teacher, they show a certain degree of self-organization, the network discovers the input data and autonomously, they usually require less training times and have a simple architecture. Finally, we find iii) reinforcement learning, in which the model learns by observing the world around it. Your information or input is the feedback you get from the outside world in response to your actions; that is, the system in this case learns based on the error-test. This type of learning is the most general within the three categories, in this model there is no instructor to tell the entity what to do. The intelligent entity must learn how the environment behaves through rewards (reinforcements) or punishments, which are a consequence of success or failure respectively. Finally, we find iv) multitasking learning, which consists of models that use the knowledge previously learned by the system in order to face problems similar to those already seen.
Within machine learning it is important to highlight that it is possible to obtain three types of knowledge:
a) Growth: which is acquired from what surrounds us and stores the information in memory as if it were traces. b) Restructuring: this consists of the interpretation of knowledge by the individual that generates reasoning and in turn new knowledge and c) adjustment: which is obtained by generalizing several concepts.
These three types of knowledge are made during a process of automatic learning but their importance lies in the characteristics of what you are trying to learn.
In this short text it is also important to highlight some differences between supervised and unsupervised learning. In the first place, supervised learning is characterized by having information that establishes in advance which group or set of data are satisfactory for the objective of the learning model. A clear example in this specific case is the one we used at the beginning of this text. In this particular case, it could be a piece of software that identifies whether or not an image corresponds to the image of a cat or a dog. For this learning model, we would have to provide the program with different images, specifying in the process if it is a cat or a dog, as we would teach the process to a child.
On the other hand, in unsupervised learning, the program would not have data that defines what information satisfies the expected conditions or not. The main objective of these programs is usually to find patterns that allow the data to be separated and classified into different groups according to their attributes. Returning to the previous example, an unsupervised program would not be able to tell us if an image corresponds to a cat or not, but it could, for example, classify the images and separate those that contain cats, buts or another different animal. Given this limitation, it is important to highlight that the information obtained by an unsupervised learning model must be subsequently interpreted by a human to give you usefulness according to what is expected to be obtained with the program.
Now, what are the fields in which machine learning is used the most today?
* DNA sequence classification
* Economic predictions and fluctuations in the stock market
* Mapping and 3D modeling
* Fraud detection
* Medical diagnostics
* Internet search engines
* Voice recognition systems
* Optimization and implementation of digital advertising campaigns
Some important dates:
<1950s Statistical methods are discovered and refined.
1950s Pioneering machine learning research is conducted using simple algorithms.
1960s Bayesian methods are introduced for probabilistic inference in machine learning.
1970s 'AI Winter' caused by pessimism about machine learning effectiveness.
1980s Rediscovery of backpropagation causes a resurgence in machine learning research.
1990s Work on Machine learning shifts from a knowledge-driven approach to a data-driven approach. Scientists begin creating programs for computers to analyze large amounts of data and draw conclusions – or "learn" – from the results. Support vector machines (SVMs) and recurrent neural networks (RNNs) become popular. The fields of computational complexity via neural networks and super-Turing computation started.
2000s Support Vector Clustering [5] and other Kernel methods [6] and unsupervised machine learning methods become widespread.[7]
2010s Deep learning becomes feasible, which leads to machine learning becoming integral to many widely used software services and applications.
An finally, some quotes:
“A baby learns to crawl, walk and then run. We are in the crawling stage when it comes to applying machine learning.” ~Dave Waters
“Interesting in predictive analytics? Then research artificial intelligence, machine learning, and deep learning.” ~SupplyChainToday.com
“Machine learning will automate jobs that most people thought could only be done by people.” ~Dave Waters
“Don’t let the digital supply chain scare you. Big data, IoT, cloud, AI, drones and deep learning are just ways to improve the supply chain.” ~SupplyChainToday.com
“Computers are able to see, hear and learn. Welcome to the future.” ~Dave Waters
“Machine Learning: A computer is able to learn from experience without being specifically programmed.”
“Although still in its infancy, machine learning will be a game changer in supply chain. ~SupplyChainToday.com
“Machine learning will increase productivity throughout the supply chain.” ~Dave Waters
Sources:
Wikipedia.