Applies to deep learning in math

Deep learning and its applications today

ORIGINAL ITEM

CHAGAS, Edgar Thiago De Oliveira [1]

CHAGAS, Edgar Thiago De Oliveira. Deep learning and its applications today. Revista Científica Multidisciplinar Núcleo do Conhecimento. 04 year, Ed. 05, Vol. 04, pp. 05-26 May 2019. ISSN: 2448-0959

Contents

SUMMARY

Artificial intelligence is no longer a plot for feature films. Research in this area is increasing every day, giving new insights into machine learning. Deep learning methods, also known as deep learning, are currently being used on many fronts, such as facial recognition in social networks, automated cars, and even some medical diagnostics. Deep learning enables computer models, which are composed of innumerable processing layers, to "learn" data with different levels of abstraction. These methods improved speech recognition, visual objects, object recognition, among the possibilities. However, this technology is still poorly known and the purpose of this study is to clarify how deep learning works and demonstrate its current uses. Of course, with the spread of this knowledge, deep learning may present other applications in the near future that are even more important to all of humanity.

Keywords: deep learning, machine learning, IA, machine learning, intelligence.

INTRODUCTION

Deep learning is understood as a branch of machine learning based on a group of algorithms that attempt to shape high level abstractions of data using a deep diagram with multiple layers of processing, consisting of multiple linear and nonlinear changes.

Deep learning works the computer system to perform tasks such as speech recognition, image identification and projection. Rather than organizing the information to act by given equations, this learning determines the basic patterns of that information and teaches computers to evolve by identifying patterns in processing layers.

This type of learning is a broad branch of machine learning methods based on learning representations of information. In this sense, deep learning is a set of learning machine algorithms that try to integrate multiple levels that are accepted statistical models that correspond to different levels of definition. Lower levels help define many notions of higher levels.

There is a lot of current research going on in this area of ​​artificial intelligence. The improvement in deep learning techniques has made it possible for computers to improve their ability to understand what is required. Research in this area aims to promote better representations and sophisticated models to identify those representations from information that is not labeled on a large scale, some as a basis in neuroscience insights and in the interpretation of computing and communication patterns in the Nervous system. Since 2006, this type of learning has evolved into a new branch of machine learning research[2].

More recently, new deep learning techniques have emerged that have impacted several studies of signal processing and pattern identification. Note a number of new problematic commands that can be resolved by these techniques, including machine learning and artificial intelligence key points.

According to Yang et al. will attract media attention to those in the[3]The progress made in this area has made great strides. Large technology organizations have made many investments in deep learning research and its new applications.

Deep learning encompasses learning on different levels of representation and incomprehensibility that help to understand information, images, sounds and texts.

Among the exhibits available through Deep Learning, it is possible to identify two salient points. The first shows that we are dealing with models that are formed by innumerable layers or steps of non-linear data processing and that they are also supervised learning practices or not, of the representation of attributions in later and immaterial layers.

It is understood that deep learning resides in the seams between the branches of neural network research, AI, graphical modeling, pattern identification and optimization, and signal processing. The attention paid to deep learning has been due to the improvement in chip processing ability, the significant magnification of the information used for education, and recent advances in machine learning and processing studies. Signals.

This advancement enabled in-depth learning practices to effectively use complex and non-linear applications, identify representations of distributed and hierarchical resources, and make effective use of labeled and unlabeled information.[4]

Deep learning refers to a broad class of machine learning methods and projects that bring together the property of using many layers of non-linearly processed data of a hierarchical nature. Using these methods and projects, much of the studies in this area can be broken down into three main clauses, Pang said [5]et al, who are the deep networks for unsupervised learning; Supervised and hybrid.

Deep unsupervised learning networks are available to capture the high sequence correlation of the analyzed or identifiable information to review or associate standards when data is not available on the stereotypes of the classes. Available in the database. Learning mapping or unattended representation relates to deep networks. Also, you can look for the mapping of grouped statistical distributions of the visible data and associated classes when they are available and can be covered as part of the visible data.

Deep neural networks for supervised learning should provide discrimination in the classification of patterns, usually individualization of the subsequent distribution of classes with the visible information that is always available for this supervised learning, also known as deep discriminatory networks.

The deep hybrid networks are highlighted by the discrimination identified with the results of generative or unsupervised deep networks that can be achieved through the improvement and regularization of the deeply monitored networks. Its attributions can also be achieved when the discriminatory supervised learning guidelines are used to assess standards in a generative or unsupervised deep network[6].

The deep and recurring networks are models that demonstrate high performance in identifying questionable patterns in Ivain and language[7]n. Despite its power of representation, the great difficulty of forming deep neural networks with generic use still exists today. Studies initiated in relation to the recurring neural networks [8]by Hinton et al. the formulas in layers.

The present study aims to clarify the progress of deep learning and its applications according to the latest research results. For this purpose, a qualitative descriptive research is carried out in which books, theses, articles and websites are conceptually carried out on the advances in the field of artificial intelligence and especially in deep learning.

Interest in machine learning has been growing over the last decade as there is an ever greater interaction between applications, whether mobile or computing devices, with individuals, through programs for spam detection, recognition in photos on social networks, smartphones with facial recognition, among others. According to Gartner, I will[9]n all business programs to be connected to machine learning by 2020. These elements attempt to justify the preparation of this study.

HISTORICAL DEVELOPMENT OF DEEP LEARNING

Artificial intelligence is not a new discovery. It dates back to the 1950s, but despite the evolution of its structure, some aspects of credibility were lacking. One such aspect is the volume of data that has arisen in a great variety and speed, allowing the creation of standards with high accuracy. However, a relevant point was how large machine learning models with large amounts of information were processed because computers could not perform such actions.

At that moment the second aspect related to parallel programming in GPUs was identified. The graphical processing units, which allow the realization of mathematical operations in parallel, in particular those with matrices and vectors that are present in models of artificial networks, enabled the current development, i.e. the big data summation (large volume of data); Parallel processing and machine learning models are presented as the result of artificial intelligence.

The basic unit of an artificial neural network is a mathematical neuron, also called a node, based on the biological neuron. The connections between these mathematical neurons are linked to those of the biological brain, and most importantly, in the way these associations develop over time, called "training".

Between the second half of the decade of 80 and the beginning of the decade of 90, there were several relevant advances in the structure of artificial networks. However, the amount of time and information it took to get good results delayed the assumption of what drove interest in artificial intelligence.

In the early 2000s, computing power expanded and the market experienced a "boom" in computing that was not possible before. It was when deep learning emerged from the great computational growth of that time as an essential mechanism for the development of artificial intelligence systems and won several machine learning competitions. Interest in deep learning continues to grow today and several commercial solutions are emerging.

Over time, several researches have been done to simulate brain function, particularly during the learning process, to create intelligent systems that could emulate tasks like classification and pattern recognition, under Other Activities. The conclusions of these studies gave the model of the artificial neuron that was later placed in a networked network called a neural network.

In 1943, Warren McCulloch, neurophysiologist, and Walter Pitts, mathematician, created a simple neural network with electrical circuits and developed a computer model of neural networks based on mathematical concepts and algorithms called thresholds. Logic or threshold logic that made research on the neural network possible, divided into two strands: the focus on the biological process of the brain and another one that focused on the application of these neural networks aimed at artificial intelligence.[10]

Donald Hebb[11] wrote a work in 1949 in which he reported that the more neural circuits are used, the more they are used as the essence of learning. With the advancement of computers in 1950, the idea of ​​a neural network gained strength and strength[12]Thanial Rochester from IBM study labs tried to form one but failed.

Practice the Dartmouth Summer Research Project[13]r Artificial Intelligence in 1956 promoted the neural networks as well as artificial intelligence and promoted the research in this field related to neural processing. In the following years, John Von Neumann imitated simple functions of neurons with vacuum tubes or telegraphs, while Frank Rosenblatt initiated the Perceptron project and analyzed the functioning of the fly's eye. The result of this research was hardware that is the oldest neural network in use to this day. However, the perceptron is very limited, which Marvin and Papert have proven.[14]

Figure 1: Neural network structure

A few years later, in 1959, Bernard Widrow and Marcian Hoff developed two models called "Adaline" and "Madaline". The nomenclature is derived from the use of several elements: ADAptive LINear. Adaline was created to identify binary patterns to make predictions about the next bit, while "Madaline" was the first neural network applied to a real problem, with an adaptive filter. The system is still in use, but only commercially.[15]

The advances previously made led to the belief that the potential of neural networks was limited to electronics. It was asked about the impact that "intelligent machines" would have on people and society as a whole.

The debate over how Artificial Intelligence would affect humans criticized research in neural networks that led to a cut in funding and, consequently, studies in the region that stayed until 1981.

In the following year, several events aroused interest in this area again. Caltech's John Hopfield presented an approach to creating useful equipment and demonstrating what it does[16]n. 1985 the American Institute of Physics began an annual meeting called Neural Networks for Computation. In 1986, the media began reporting on the multiple layers of neural networks, and three researchers came up with similar ideas, called backpropagation networks, because they identify patterns of distributing failures across the network.

The hybrid networks only had two layers, while the backpropagation network[17]Works many present, so this network retains the information more slowly because it takes thousands of iterations to learn but also present more results. Exactly. The first international conference on neural networks of the Institute for Electrical and Electronic Engineers (IEEE) took place as early as 1987.

In 1989, scientists created algorithms that made use of deep neural networks, but the "learning" time was very long, which prevented their application to reality. In 1992, Juyang Weng Diulga created The Cresceptron Method to perform the detection of 3D objects from turbulent scenes.

In the mid-2000s, the term "deep learning" or "deep learning" according to an article by Geoffrey Hinton and Ruslan Salakhutdinov begins to become widespread [18]who showed how a multi-layered neural network could be trained beforehand, one layer at a time.

In 2009, the Neural Network Systems Processing Workshop on Deep Learning for Speech Recognition will be held, and it will be confirmed that neural networks do not require prior training when dealing with a large group of data and failure rates decrease. to tell[19]d.

In 2012, the research provided identification algorithms of artificial patterns with human performance in some tasks. And the Google algorithm identifies cats.

In 2015, Facebook used deep learning to automatically tag and recognize users in photos. The algorithms perform face recognition tasks with deep networks. In 2017 there was a large-scale adoption of deep learning in various business applications and mobile devices, as well as advances in research[20]G.

Deep learning has set itself the task of showing that a fairly extensive range of data, fast processors, and a fairly sophisticated algorithm enable computers to perform tasks such as recognizing images and speech, among other possibilities.

The research on neural networks has gained in importance with promising attributions presented by the neural network models, due to the recent technological innovations of implementation that make it possible to develop bold neural structures parallel to the hardware, the satisfactory performance of these systems, with superior performance than traditional systems, including. The development of neural networks is deep learning.

DEEP LEARNING

The first thing to do is to differentiate between artificial intelligence, machine learning and deep learning.

Figure 2: Artificial Intelligence, Machine Learning, and Deep Learning

The research area of ​​artificial intelligence is the research and design of intelligent sources, i.e. a system that can make decisions based on a characteristic that is considered intelligent. In Artificial Intelligence there are several methods that model this property and among them is the sphere of machine learning, where decisions (intelligence) are made based on examples and not specific programming.

Machine learning algorithms require information in order to functionions and learning functions that can be used for future decisions. Deep learning is a subset of machine learning techniques that typically use deep neural networks and require a large amount of information for training[21]

According to Santana, there is e[22]Some differences between machine learning techniques and deep learning methods, and the most important ones are the necessity and impact of data volume, computing power, and flexibility in problem modeling.

Machine learning needs data to identify patterns, but there are two problems with the data related to dimensionality and stagnation of performance by introducing more data beyond the restrained line. It is proven that there is a significant decrease in performance when this happens. With regard to dimensionality, the same occurs as there is a lot of information that, through the classical techniques, can identify the dimension of the problem.

Figure 3: Comparison of deep learning with other algorithms on the amount of data.

The classical techniques also represent a saturation point in terms of the amount of data, that is, have a maximum limit to extract the information that does not occur with deep learning, created to work with a large volume of data.

Regarding the computing power for deep learning, its structures are complex and require a large amount of data for its training, which demonstrates its reliance on a large computing power to implement these practices. While other classic practices than CPU require a lot of computing power, deep learning techniques are superior.

The search for parallel computing and the use of GPUs with CUDA-Compute Unified Device Architecture or Unified Computing Device Architecture initiated deep learning because it was not possible with the use of a simple CPU.

In comparison with the formation of a deep neural network or deep learning with the use of a CPU, it turns out that even with prolonged training it would be impossible to achieve satisfactory results.

Deep learning, also known as deep learning, is part of machine learning and it applies algorithms to process data and reproduce the processing of the human brain.

Deep learning uses layers of math neurons to process data, identify language, and recognize objects. The data is carried through each level, with the output from the previous level providing input to the next level. The first level in a network is called the entry layer and the last is called the entry layer. The intermediate layers are called hidden layers, and each layer of the network is formed by a simple and uniform algorithm that includes a kind of activation function.

Figure 4: Simple neural network and deep neural network or deep learning

The outermost layers in yellow are the input or output layers, and the intermediate or hidden layers are red. Deep learning is responsible for recent advances in computers, speech recognition, language processing, and auditory identification, based on the definition of artificial neural networks or computer systems that affect the way the human brain works.

Another aspect of deep learning is the acquisition of resources that use an algorithm to automatically create relevant parameters of information for training, learning and understanding, a task of the artificial intelligence engineer.

Deep learning is a development of neural networks. Interest in deep learning has gradually grown in the media and several researches have spread in the area and its application has reached automobiles, including in the diagnosis of cancer and autism[23]

The first deep learning algorithms with several layers of nonlinear mappings present their origins in Alexey Grigoryevich Ivakhnenko, who developed the method of data manipulation, and Valentin Grigor 'Evich Lapa, author of the work Cybernetics and Predictive Technology in 1965.[24]

Both used thin and deep models with polynomial activation functions that were examined with statistical methods. Through these methods, they selected the best resources at each tier and carried them to the next level without backpropagation "training" the entire network, but using minimal squares at each tier from which the previous ones were installed. Independently in the later shifts, manually.

Figure 5: Structure of the first deep network known as Alexej Grigorewitsch Ivakhnenko

At the end of the decade of 1970 came the winter of artificial intelligence, a drastic cut in funding for research on the subject. The impact has limited advances in deep neural networks and artificial intelligence.

The first corporate law networks were used in 1979 by Kunihiko Fukushima with several layers of groupings and convolutions. He created an artificial neural network called a neocognitron, with a hierarchical and layered layout that enabled the computer to recognize visual patterns. The networks were similar to modern versions, with "training" focusing on the strategy of strengthening regular activation in myriad layers. In addition, the design of Fukushima allowed manual adjustment of key resources by increasing the importance of certain connections[25]e.

Many Neocognitron guidelines are still in use as top-down connections and new learning practices have encouraged the realization of various neural networks. When multiple patterns are presented at the same time, the Selective Care Model can separate them and identify each pattern, paying attention to each value. A more recent Neocognitron can identify patterns with a lack of data and complete the picture by inserting the missing information, known as the inference.

Backpropagation, which was used for deep learning error training, went ahead from 1970 when Seppo Linnainmaa wrote a dissertation and inserted a FORTRAN code for backpropagation without it succeeding until 1985. Rumelhart, Williams and Hinton then demonstrated backpropagation in a neural network with representations of the distribution.

This discovery allowed the AI ​​debate to reach cognitive psychology, which attracted questions about human understanding and its relationship to symbolic logic, as well as connections. In 1989, Yann LeCun conducted a hands-on demonstration of backpropagation, using the combination of convolutionary neural networks to identify the written digits.

During this period there was again a shortage of funds for research in this area, known as the second winter of the IA, which took place between 1985 and 1990, also impacting research in neural networks and deep learning. Some researchers' expectations did not reach the expected level, which deeply irritated investors.

In 1995, Dana Cortes and Vladimir Vapnik created the supporting vector machine[26]ine or the supporting vector machine, which was a system for mapping and identifying similar information. The Long Short Memory LSTM for regular neural networks was developed in 1997 by Sepp Hochreiter and Jürgen Schmidhuber[27]t.

The next step in the evolution of deep learning came in 1999 when computing and graphics processing (GPUs) became faster. The use of GPUs and their fast processing meant an increase in the speed of computers. Neural networks competed with support vector machines. The neural network was slower than a support vector machine, but they got better results and evolved as more training information was added.

In 2000, a problem called the Vanishing Gradient was identified. The tasks learned in the lower layers were not transferred to the upper layers, but only occurred with those with gradient-based learning methods. The cause of the problem lay in some activation functions that reduced their input, which influenced the output area, and created large input fields that were displayed in a very small area, which led to a sinking color gradient. The solutions that were implemented to solve the problem were the pre-workout shift by shift and the development of long-term and short-term memories[28]s.

In 2009, Fei-Fei Li published ImageN[29]et, with a free database of more than 14 million images, focused on "building" neural networks and pointing out how big data would affect the operation of machine learning.

The speed of GPUs, by 2011, continued to increase, allowing the composition of convolutionary neural networks without the need for pre-workout layer by layer. So it became notorious that deep learning was beneficial in terms of efficiency and speed.

In this day and age, big data processing and the advancement of artificial intelligence depend on deep learning, which can develop intelligent systems and promote the creation of a fully autonomous artificial intelligence that has an impact on the whole of society.

FLEXIVILITY OF NEURAL NETWORKS AND THEIR APPLICATIONS

Despite the existence of several classical techniques, the DEEP learning structure and its basic unit, the neuron is generic and very flexible. If we make a comparison with the human neuron that supplies synapses, we can see some connections between them.

Figure 6: Correlation between a human neuron and an artificial neural network

It should be noted that the neuron is formed by the dendrites, which are the entry points, a nucleus which in the artificial neural networks is the processing nucleus and the exit point which is represented by the axon. In both systems, the information is entered, processed and found out.

If you think of it as a math equation, the neuron reflects the sum of the inputs multiplied by weights, and that value goes through an activation function. This sum was listed by McCulloch and Pitts in 1943[30]

Regarding the infamous interest in deep learning these days, Santana is[31] Complex believes that there are two factors which are the amount of information available and the limitation of the older techniques besides the current computing power to train networks. The flexibility to connect several neurons in a more complex network is the differential of DEEP learning structures. A convolutional neural network is widely used in face recognition, image recognition, and assignment extraction.

A conventional neural network consists of several layers, the so-called layers. Depending on the problem to be solved, the number of layers can vary, being able to have up to hundreds of layers, with factors in the amount affecting the complexity of the problem, time and computing power

There are several different structures with innumerable purposes, and how they work depends on the structure too, and all of them are based on neural networks.

Figure 7: Examples of neural networks

This architectural flexibility makes it possible to solve various problems. Deep learning is a general objective technique, but the most advanced areas have been: computer vision, speech recognition, natural language processing, recommendation systems.

The arithmetic image includes object recognition, semantic segmentation, in particular autonomous cars. It can be confirmed that computer vision is part of artificial intelligence and is defined as a set of knowledge that seeks artificial modeling of human vision with the aim of mimicking its functions through the development of software and hardware. advanced[32]n.

Applications of the computing vision include military use, the marketing market, security, public services and the production process. Autonomous vehicles represent the future of safer transportation, but it is still in the testing phase as it encompasses multiple technologies applied to one function. The computational vision in these vehicles as it allows the recognition of the path and the obstacles, the improvement of the routes.

In the context of security, facial recognition systems have become increasingly prominent, given the level of security in public and private places that are also implemented in mobile devices. Likewise, they can serve as keys to access financial transactions, while on social networks, it detects the presence of the user or his friends in photos.

Regarding the marketing market, a study developed by Image Intelligence indicated that 3 billion images are shared daily from social networks and 80% contain references related to specific companies but with no textual references. Specialized marketing companies offer real-time monitoring and management services. With computer vision technology, the image identification accuracy reaches 99%.

In the public service, it covers the security of the site by monitoring cameras, vehicle traffic through stereoscopic images that make the visual system efficient.

In the production process, companies from various industries use the computational perspective as a quality instrument. In every industry, the most advanced software, coupled with the ever increasing processing capacity of hardware, increases the use of computational visions.

Monitoring systems enable the detection of ready-made standards and indicate failures that would not be recognizable when looking at an employee in the production line. In the same context, the replacement automation project is used for inventory control. Real-time inventory and sales control enables technology to control the operations of a company and thus increase its bottom line. There are other uses in medicine, education, and e-commerce.

Conclusions

The present study sought to elucidate what deep learning is and how it can be used in today's world. Deep learning techniques progress mainly through the use of several layers. However, there are still limitations to using deep neural networks as they are just one way to learn several changes to be implemented in the input vector. Changes due to a number of parameters that are updated during the training period.

It is undeniable that artificial intelligence is a closer reality, but there is still a long way to go. The acceptance of deep learning in various fields of knowledge enables society as a whole to benefit from the wonders of modern technology.

With regard to artificial intelligence, it is verified that this technology is adaptive, although very important, has a linear and non-malleable nature than humans, which is a great differential and essential to some areas of knowledge that can be deeply learned not yet implemented.

In some way, the use of deep learning methods will enable the machines to assist society in various activities as shown, by expanding human cognitive capacities and further developing these areas of knowledge.

BIBLIOGRAPHICAL NOTES

ALIGER. Saiba o que é visão computacional e como ela pode ser usada. 2018. Disponível em: https://www.aliger.com.br/blog/saiba-o-que-e-visao-computacional/

BITTENCOURT, Guilherme. Breve história da Inteligência Artificial. 2001

CARVALHO, André Ponce de L. Redes Neurais Artificiais. Disponível em: http://conteudo.icmc.usp.br/pessoas/andre/research/neural/

CHUNG, J; GULCEHE, C; CHO, K; BENGIO, Y. Gated feedback recurrent neural networks. In Proceedings of the 32nd International Conference on International Conference on Machine Learning, volume 37 of ICML, pages 2067-2075. 2015

CULTURA ANALITICA. Entenda o que é deep learning e como funciona. Disponível em: https://culturaanalitica.com.br/deep-learning-oquee-como-funciona/cultura-analitica-redes-neurais-simples-profundas/

DENG, Jia et al. Imagenet: Um banco de dados de imagens hierárquicas em grande escala. In: 2009 Conferência IEEE sobre visão computacional e reconhecimento de padrões. Ieee ,. p. 248-255. 2009

DETTMERS, Tim. Aprendizagem profunda em poucas palavras: Historia e Treinamento. Disponível em: https://devblogs.nvidia.com/deep-learning-nutshell-history-training/

FUKUSHIMA, Kunihiko. Neocognitron: Um modelo de rede neural auto-organizada para um mecanismo de reconhecimento de padrões not afetado pela mudança de posição.Cibernética biológica, v. 36, n. 4, p. 193-202, 1980.

GARDENER. Gartner diz que a AI Technologies estará em quase todos os novos produtos de software até 2020. Disponível em: https://www.gartner.com/en/newsroom/press-releases/2017-07-18-gartner-says-ai-technologies-will-be-in-almost-every-new-software-product -by-2020

GOODFELLOW, I; BENGIOo, Y; COURVILLE, A. Deep learning. MIT Press. Disponível em: http://www.deeplearningbook.org

GRAVES, A. Sequence Transduction with recurrent neural networks. arXiv preprint arXiv: 1211.3711, 2012.

HOCHREITER, Sepp.O problema do gradiente de fuga durante a aprendizagem de redes neurais recorrentes e soluções de problemas. Revista Internacional de Incerteza, Fuzziness and Knowledge-Based Systems, v. 6, n.02, p. 107-116, 1998.

HEBB, Donald O. The organization of behavior; a neuropsycholocigal theory. A Wiley Book in Clinical Psychology., P. 62-78, 1949.

HINTON, Geoffrey E; SALAKHUTDINOV, Ruslan R. Reduce a dimensionalidade dos dados com redes neuraisciencia , v. 313, n.5786, p. 504-507, 2006.

HOCHREITER, Sepp; SCHMIDHUBER, Jürgen. Longa memória de curto prazo. Computação Neural, v. 9, n.8, p. 1735-1780, 1997.

INTRODUCTION to artificial neural networks,Proceedings Electronic Technology Directions to the Year 2000, Adelaide, SA, Australia, 1995, pp. 36-62. Disponível em: https://ieeexplore.ieee.org/document/403491

KOROLEV, Dmitrii. Set of different neural nets. Neuron network. Deep learning. Cognitive technology concept. Vector illustration Disponível em: https://stock.adobe.com/br/images/set-of-different-neural-nets-neuron-network-deep-learning-cognitive-technology-concept-vector-illustration/145011568

LIU, C .; CAO, Y; LUO Y; CHEN, G; VOCCARANE, V, MA, Y; CHEN, S; HOU, P. A new Deep Learning-based food recognition system for dietary assessment on an edge computing service infrastructure. IEEE Transactions on Services Computing, PP (99): 1-13. 2017

LORENA, Ana Carolina; DE CARVALHO, André CPLF. Uma introdução to support vector machines. Revista de Informática Teórica e Aplicada, v. 14, n. 2, p. 43-67, 2007.

MARVIN, Minsky; SEYMOUR, paper. Perceptrons. 1969. Disponível em: http://134.208.26.59/math/AI/AI.pdf

MCCULLOCH, Warren S; PITTS, Walter. To cálculo lógico das idéias imanentes na atividade nervosa. O boletim de biofísica matemática, v. 5, n. 4, p. 115-133, 1943.

NIELSEN, Michael. How the backpropagation algorithm works. 2018. Disponível em: http://neuralnetworksanddeeplearning.com/chap2.html

NIPS. 2009. Disponível em: http://media.nips.cc/Conferences/2009/NIPS-2009-Workshop-Book.pdf

PANG, Y; SUN, M; JIANG, X; LI, X .. Convolution in convolution for network in network. IEEE Transactions on Neural Networks and Learning Systems, PP (99): 1-11. 2017

SANTANA, Marlesson. Deep Learning: do Conceito às Aplicações. 2018. Disponível em: https://medium.com/data-hackers/deep-learning-do-conceito-%C3%A0s-aplica%C3%A7%C3%B5es-e8e91a7c7eaf

UDACITY. Conheça 8 aplicações de Deep Learning no mercado. Disponível em: https://br.udacity.com/blog/post/aplicacoes-deep-learning-mercado

YANG, S; LUO, P; LOY, C.C; TANG, X. From facial parts responses to face detection: A Deep Learning approacH. In Proceedings of 2015 IEEE International Conference on Computer Vision (ICCV), pages 3676-3684. 2015

  1. Liu, C., Cao, Y., Luo, Y., Chen, G., Vokkarane, V., Ma, Y., Chen, S., and Hou, P. (2017). A new Deep Learning-based food recognition system for dietary assessment on an edge computing service infrastructure. IEEE Transactions on Services Computing, PP (99): 1-13.
  2. Yang, S., Luo; P., Loy, C.-C; Tang, X. From facial parts responses to face detection: A Deep Learning approacH. In Proceedings of 2015 IEEE International Conference on Computer Vision (ICCV), p. 3676
  3. Chung, J; Gulcehre, C; Cho, K; Bengio, Y. Gated feedback recurrent neural networks. In Proceedings of the 32nd International Conference on International Conference on Machine Learning, volume 37 of ICML, p. 2069
  4. PANG, Y; Sun, M., Jiang, X., and Li, X. (2017). Convolution in convolution for network in network. IEEE Transactions on Neural Networks and Learning Systems, PP (99): 5. 2017 ↑
  5. Goodfellow, I., Bengio, Y., and Courville, A. Deep Learning. MIT Press. http://www.deeplearningbook.org.
  6. GRAVES, A. Sequence transduction with recurrent neural networks. arXiv preprint arXiv: 1211.3711, 2012.
  7. GARDENER. Gartner diz que a AI Technologies estará em quase todos os novos produtos de software até 2020. Disponível em: https://www.gartner.com/en/newsroom/press-releases/2017-07-18-gartner-says-ai-technologies-will-be-in-almost-every-new-software-product -by-2020
  8. MCCULLOCH, Warren S; PITTS, Walter. To cálculo lógico das idéias imanentes na atividade nervosa. O boletim de biofísica matemática, v. 5, n. 4, p. 115-133, 1943.
  9. HEBB, Donald O. The organization of behavior; a neuropsycholocigal theory. A Wiley Book in Clinical Psychology., P. 62-78, 1949.
  10. BITTENCOURT, Guilherme. Breve história da Inteligência Artificial. 2001.p. 24
  11. Idbiem p. 25th
  12. MARVIN, Minsky; SEYMOUR, paper. Perceptrons. 1969. Disponível em: http://134.208.26.59/math/AI/AI.pdf
  13. Introduction to artificial neural networks, "Proceedings Electronic Technology Directions to the Year 2000, Adelaide, SA, Australia, 1995, pp. 36-62. Disponível em: https://ieeexplore.ieee.org/document/403491
  14. BITTENCOURT, Guilherme op cit. p. 32
  15. NIELSEN, Michael. How the backpropagation algorithm works. 2018. Disponível em: http://neuralnetworksanddeeplearning.com/chap2.html
  16. HINTON, Geoffrey E; SALAKHUTDINOV, Ruslan R. Reduce a dimensionalidade dos dados com redes neuraisciencia , v. 313, n.5786, p. 504-507, 2006.
  17. NIPS. 2009. Disponível em: http://media.nips.cc/Conferences/2009/NIPS-2009-Workshop-Book.pdf
  18. Santana, Marlesson. Deep Learning: do Conceito às Aplicações. 2018. Disponivel em: https://medium.com/data-hackers/deep-learning-do-conceito-%C3%A0s-aplica%C3%A7%C3%B5es-e8e91a7c7eaf
  19. SANTANA, Marlesson. Deep Learning: do Conceito às Aplicações. 2018. Disponível em: https://medium.com/data-hackers/deep-learning-do-conceito-%C3%A0s-aplica%C3%A7%C3%B5es-e8e91a7c7eaf
  20. idem
  21. UDACITY. Conheça 8 aplicações de Deep Learning no mercado. Disponível em: https://br.udacity.com/blog/post/aplicacoes-deep-learning-mercado
  22. Dettmers, Tim. Aprendizagem profunda em poucas palavras: Historia e Treinamento. Disponível em: https://devblogs.nvidia.com/deep-learning-nutshell-history-training/
  23. FUKUSHIMA, Kunihiko. Neocognitron: To modelo de rede neural auto-organizada para um mecanismo de reconhecimento de padrões not afetado pela mudança de posição. Cibernética biológica, v. 36, n. 4, p. 193-202, 1980.
  24. LORENA, Ana Carolina; DE CARVALHO, André CPLF. Uma introdução to support vector machines. Revista de Informática Teórica e Aplicada, v. 14, n. 2, p. 43-67, 2007.
  25. HOCHREITER, Sepp; SCHMIDHUBER, Jürgen.Longa memória de curto prazo. Computação Neural, v. 9, n.8, p. 1735-1780, 1997.
  26. HOCHREITER, Sepp.O problema do gradiente de fuga durante a aprendizagem de redes neurais recorrentes e soluções de problemas. Revista Internacional de Incerteza, Fuzziness and Knowledge-Based Systems, v. 6, n.02, p. 107-116, 1998.
  27. DENG, Jia et al. Imagenet: Um banco de dados de imagens hierárquicas em grande escala. In: 2009 Conferência IEEE sobre visão computacional e reconhecimento de padrões. Ieee, 2009. p. 248-255.
  28. MCCULLOCH, Warren S .; PITTS, Walter. To cálculo lógico das idéias imanentes na atividade nervosa. O boletim de biofísica matemática, v. 5, n. 4, p. 115-133, 1943.
  29. SANTANA, Marlesson. Deep Learning: do Conceito às Aplicações. 2018. Disponivel em: https://medium.com/data-hackers/deep-learning-do-conceito-%C3%A0s-aplica%C3%A7%C3%B5es-e8e91a7c7eaf
  30. ALIGER. Saiba o que é visão computacional e como ela pode ser usada. 2018. Disponível em: https://www.aliger.com.br/blog/saiba-o-que-e-visao-computacional/

[1] Bachelor of Business Administration.

Submitted: May, 2019

Approved: May 2019

Degree in information security, bachelor in business administration, MBA in project management, specialist in cybersecurity, blockchain, artificial intelligence and international certifications.