What is Deep Learning?
Deep learning (Deep Learning) is a type of automatic learning (learning) and Artificial intelligence (IA) that mimics the way humans get certain types of knowledge. It is an important element of data science, which includes statistics and predictive models, by ende, is extremely beneficial for data scientists that have the task of collecting, analyzing and interpreting large amounts of information; providing that this process is faster and easier.
In your book Deep Learning (Deep Learning) (2021), Gonzalo Pajares Martinsanz and cols. develop the functioning of this, saying that Deep learning algorithms apply to layer-shaped artificial neural networks: input (input) hidden layer (hidden) and otput layer (salida).
The data enters by the first layer, in which there are several artificial neurons that are activated or not depending on the data. To achieve an acceptable level of accuracy, Deep learning programs require access to immense amounts of data training and processing power, none of which were easily available to programmers until the age of big date and cloud computing.
Since deep learning programming can create complex statistical models directly from its own iterative output, it can also generate precise predictive models from large quantities of unstructured data and without labeling.
Rudolph Russell (2018), in his book Deep Learning. Foundations of Learning Deep for Beginners, states that the above mentioned process is important, since Internet of things (IoT) It continues to become more omnipresent because most of the data that creates humans and machines are not structured or labeled.
It should be noted that this is a question of whether it is a question of whether it is a matter of security. They can use various methods to create solid models of deep learning. These techniques include lowering the learning rate, transfer learning, zero training and desertion. Jordi Houses Rome et al, in his work Deep Learning. Principles and foundations (2020) describes them, stating their characteristics:
Decay of the learning rate: The learning rate is a hyper parameter (i.e. a factor that defines the system or establishes the conditions for its functioning before the learning process) that controls the model change in response to the estimated error whenever the weights are changed. The learning rates that are very high can result in unstable training processes or in learning a suboptimal weight set. The learning rates that are very small can produce a prolonged training process that has the potential of ataskosis.
The method of reducing the learning rate (also called annealed learning rate or adaptive learning rates) is the process of adapting the learning rate to increase performance and reduce training time.
Learning Transfer: This procedure implies improving a previously trained model and requires an interface to the interior of a pre-existing network. First, users feed this network with new data that contain unknown ratings. After the adjustments needed in the network, new tasks with more specific categorization capabilities can be developed. This method has the advantage of requiring less data than others, thus reducing the calculation time to minutes or hours.
Zero training: This process implies that a developer recapsulates a large set of labeled data and configure a network architecture that can learn the features and the model. This technique is especially useful for new applications, as well as for applications with a large number of output categories. However, in general, it is a less common approach, as it requires excessive amounts of data, which generates that training takes days or weeks.
Deertion: This method tries to solve the problem of overadjustment in networks with large amounts of parameters by randomly dropping units and their neuronal network connections during training. It is verified that the defection method can optimize the performance of neural networks in supervised learning tasks, in areas such as voice recognition, document classification and computational biology.
Practical examples/ Scientific and technological advances
Currently, are multiple applications that have the algorithms Deep Learning, Among them we can mention the following exemplifications:
Artificial vision
La Artificial vision acquires the ability to recognize characters, images, objects and even faces, and their impact on Industry 4.0 It's important. Betterview is a company that applies the artificial vision to property insurance, using geospatial data, which can automatically evaluate what material is made a building, in what condition is the ceiling, what is the surface of this, how many garden rubble possesses the property, which so close is a structure of vegetation, and hundreds of other factors that collectively determine the risk profile of the property and the optimal price of the insurance policy.
Predictive analysis
El Salvador prediction analysis It can generate more accurate forecasts of business outcomes, market developments or energy needs. Companies that use this type of method are: Shell, which employs predictive analysis to anticipate failures in oil drilling and inventory management and IBM, that performs it to solve commercial and research problems.
Virtual Assistants
Alexa, Cortana or Sir are assistants who understand and execute the user's voice commands in natural language and are able to learn with time.
Chatbots
The chatbots are used in customer service systems to solve user problems through a chat and also learn progressively. An example of chatbot, is Gala, which belongs to Bank Galicia.
Robotics
Deep learning makes it easy for robots to perform human-like tasks even by making real-time decisions. They can also meet their own maintenance. Sophia and Threat That's two examples.
Health
Through the analysis of medical images, deep learning facilitates the detection of diseases and computer-assisted diagnosis, even without the intervention of personnel. MONA (Medical Open Network for AI) uses techniques Deep Learning and Artificial Intelligence To detect lung lesions in the segmentation of 3D images of computed tomography of patients produced by COVID.
Entertainment
Content companies in streaming, as Netflix, HBO or YouTube, provide automatic recommendations and subtitles to your users.
As you can appreciate, the use of Deep Learning gains more and more ground in various areas, which implies a strong investment. In this sense, it is pertinent to mention that, worldwide, a gradual increase in the investment of this technology in the world market.
Conclusion
It is currently essential that companies and institutions employ Deep Learning in their processes, as this will imply high accuracy in data processing and effectiveness in implementing effective strategies in certain activities. The advancement of Artificial Intelligence allows to improve the quality of life of human beings in various areas (health, education, entertainment, work, among others), which means to continue establishing an ethical regulatory framework so that the benefits it offers to humanity are not distorted.
Although there are still some challenges to consider (especially ethically and morally), several organisms such as UNESCO and the UN of the hand of several experts and industry referents work daily to provide solutions that promote the Sustainable Development Goals in Agenda 2030. In this context, great human knowledge is taken as the cornerstone to solve the emerging problems linked to Artificial Intelligence, as stated by Anais Nin: “The possession of knowledge does not kill the sense of wonder and mystery. There is always more mystery”.
Comments