Deep learning - What is deep learning? This is the method of obtaining knowledge of Artificial Intelligence

Deep learning - What is deep learning? This is the method of obtaining knowledge of Artificial Intelligence
Deep learning is a set of algorithms for machine learning (machine learning) that attempts to model high - level abstractions data using computational architectures that support multiple non - linear transformations and iterative data expressed in matrix form or tensor.

Deep learning is part of a broader set of machine learning methods based on assimilating data representations. An observation (for example, an image) can be represented in many forms (for example, a vector of pixels), but some representations make it easier to learn tasks of interest (for example, "is this image a human face?") on the basis of examples, and research in this area tries to define which representations are better and how to create models to recognize these representations.

Several deep learning architectures, such as deep neural networks, deep convolutional neural networks, and deep belief networks, have been applied to fields such as computer vision, automatic speech recognition, and recognition of audio and music signals, and have been shown to produce cutting-edge results in various tasks.

The   deep learning is a topic that is becoming increasingly important in the field of artificial intelligence (AI). Being a subcategory of machine learning, deep learning deals with the use of neural networks to improve things such as voice recognition, computer vision and natural language processing.

Is quickly becoming one of the most popular fields inform to tica. In recent years, deep learning has helped to achieve progress in areas as diverse as object perception, machine translation and voice recognition (all of them are especially complex areas for AI researchers).

DEEP LEARNING DEFINITION

There is no single definition of deep learning. In general, it is a class of algorithms designed for machine learning. From this common point, different publications focus on different characteristics, for example:

• Use a cascade of layers with non-linear processing units to extract and transform variables. Each layer uses the output of the previous layer as input. Algorithms can use supervised learning or unsupervised learning, and applications include data modeling and pattern recognition.
• Be based on the learning of multiple levels of characteristics or representations of data. The higher-level characteristics are derived from the lower level characteristics to form a hierarchical representation.
• Learn multiple levels of representation that correspond to different levels of abstraction. These levels form a hierarchy of concepts.

All these ways of defining deep learning have in common: multiple layers of nonlinear processing; and supervised or unsupervised learning of feature representations in each layer. The layers form a hierarchy of characteristics from a lower level of abstraction to a higher one.

The deep learning algorithms contrast with the shallow learning algorithms by the number of transformations applied to the signal as it propagates from the input layer to the output layer. Each of these transformations includes parameters that can be trained as weights and thresholds (p6). There is a de facto standard for the number of transformations (or layers) that turns a deep algorithm, but most researchers in the field believes that deep learning involves more than two intermediate transformations (p7) .

Deep learning, known as deep neural networks, is an aspect of AI that emulates the learning that human beings use to obtain certain knowledge.

The Artificial Intelligence is becoming time that is more human. We are facing a technology that already revolutionizes the world, but that still has a lot to evolve and change in our society. We have already talked in Transkerja.com about how the AI and robots are going to change our planet, and how they will replace humans in millions of jobs.

However, we should not hold negative thinking or fear that machines are intelligent and can learn. We have the key to control the complex technology that, today, is already implemented in many more tasks than we imagine.

Although we understand the capabilities of an Artificial Intelligence, it is difficult for us to understand how they can have these incredible qualities. One of the most important processes is the 'machine learning’, that is, how an AI can learn and obtain knowledge while working.

The basis of this technology is to ensure that a robot can enjoy the same cognitive qualities as a human (or better) and that can help us. A clear example are virtual assistants such as Aura or Alexa.

DEEP LEARNING VS MACHINE LEARNING

To understand what deep learning is, the first thing to do is distinguish it from other disciplines that belong to the field of AI.

One consequence of IA was the learning m to machine, where the computer extracts knowledge through supervised experience. This used to involve a human operator who helped the machine learn by providing hundreds or thousands of training examples and manually correcting their mistakes.

Although machine learning has become a dominant element in the field of AI, it presents certain problems. On the one hand, it consumes a lot of time. On the other hand, it is not yet a true measure of machine intelligence since it relies on human ingenuity to propose the abstractions that allow the computer to learn.

Unlike what occurs with machine learning, deep learning is to less subject to supervision or n. It implies, for example, the creation of large-scale neural networks that allow the computer to learn and "think" on its own without the need for direct human intervention.

Deep learning “does not really look like a computer program," says Gary Marcus, a psychologist and AI expert at New York University. In this sense, he comments that the ordinary computer code is written following very strict logical steps. "But what can be found in deep learning is something different. There is not a series of instructions stating that: "if a thing is true, do this other thing," he says. Instead of being based on linear logic, deep learning theory is based on í as about c or how the human brain works. The program consists of nested layers of interconnected nodes. After each new experience, learn by rearranging the connections between the nodes.

In-depth learning has shown that it has potential as the basis for creating software capable of determining emotions or events described in a text even without being explicitly cited, recognizing objects in photographs and making sophisticated predictions about the likely future behavior of people.

To understand what learning is, a great example that you can find in a Transkerja.com article is great. Imagine a child that the first word he learns is "dog." The child learns what it is, and what it is not, a dog, pointing to objects and pronouncing the word "dog" before his father. The father says "Yes, that's Dog”, or "No, that's not a dog”. While the child continues to aim at the objects, he becomes more aware of the characteristics that all the dogs he points to have, and his father says whether he is wrong or not.

Little by little, the child clarifies a complex abstraction by building a hierarchy in which each level of abstraction is created with the knowledge obtained in the previous hierarchical layer. That is, the more dogs you see, the more you know what a dog is.

NEURAL NETWORK

In information technology, a neural network is a system of programs and data structures that approximates the functioning of the human brain.  A neural network usually involves a large number of processors running in parallel, each having its own small sphere of knowledge and access to data in its local memory. Usually, a neural network is "trained" or fed with large amounts of data and rules about relationships (for example, "a grandparent is older than a person's father"). Then, a program can tell the network how to behave in response to an external stimulus (for example, to a data input by a computer user that is interacting with the network) or can initiate the activity by itself (within the limits). Of his access to the external world).
   

FROM MACHINE LEARNING TO DEEP LEARNING

Do you remember how you learned to communicate? You were listening or seeing the same words many times and you formed patterns, to learn to identify words and connect concepts, until you know how to generate an answer.

Computers learn in a similar way. Chatbots like Cortana and virtual assistants like Alexa are able to "understand" our questions and generate relevant answers thanks to the processing of high volumes of data, whether text data or voice recordings.

The virtual assistant is trained by means of algorithms that expose you to millions and millions of examples in the pronunciation of each word. When we talk about machine learning or machine learning, in general it refers to this process of "training" with data.

DEEP LEARNING NETWORKS IN MACHINES

In this case, we talk about deep learning, one of the keys in the process. Also known as deep neural networks, it is an aspect of AI that emulates the learning that we human beings use to obtain certain knowledge. We could consider it as a way to automate predictive analysis.

It is important to know that deep learning is a specific branch of learning an AI. In this case, while the traditional machine learning algorithms are linear, the deep learning algorithms are stacked in a hierarchy of increasing complexity and abstraction. If you have more questions, you can look at our post about the differences between machine learning and deep learning.

THE ROLE OF NEURAL NETWORKS

We as humans have a brain that facilitates this learning process, without us noticing. Instead, machines depend on certain algorithms to guide them in learning processes, the rules they use to review and make sense of all the data that is taught. We can say that the most important gear of deep learning are neural networks, an automated implementation of statistical analysis. The "mystery" behind neural networks is pure statistics and classical probability theory: non-linear regressions, and Bayesian classification, to name a few, science generated since the 17th century. What neural networks do is to iteratively process large volumes of data by changing the weighting of these formulas, with enormous automatic capacity.

AN AI NEEDS MILLIONS OF DATA TO LEARN

This procedure of the child is the same as the computer programs do. Each algorithm in the hierarchy applies a non-linear transformation in its input, and uses what it learns to create a statistical model as output. The iterations continue until a certain level of precision is reached.

What is achieved with deep learning is that the system has less and less margin for error. AI is trained by introducing a gigantic amount of data that, following that repetition process, makes software impressively accurate. The programmers who are in charge of this work are in luck, if there is something that abounds in the digital era, it is information and data.

THE GAME OF DEEP LEARNING

In 2011, Google launched the Google Brain project, which created a neural network trained with deep learning algorithms, highlighting its ability to recognize high-level concepts. Last year, Facebook created the Unit research or n AI using in - depth-learning expertise to help create solutions that best identify faces and objects in the 350 million videos and photos that are transferred daily to Facebook. An example of deep learning in action is the speech recognition system that is included in Google Now or in the Apple system called Siri.

THE FUTURE

Deep learning promises many advances, making the construction of automatic or vile driverless and the creation of robotic butlers constitute real possibilities. They are still limited, but what they can achieve was unthinkable just a few years ago. In addition, the pace at which they advance is unprecedented.

The ability to analyze massive data sets and to use deep learning in computer systems that can adapt to the experience, instead of depending on a human programmer, will lead to significant technological advances.

Ranging from drug discovery or development of new materials to the creation or n of robots with a higher level of consciousness about the world around them. Perhaps this explains why Google has gone shopping lately and robotics companies have been at the top of their list. In a matter of months, they have acquired eight robotics companies.