Deep Learning
Abstract
The most recent technologies to be released include the deep learning technology, a form of machine learning that empowers computers to learn based on experience and understanding the world in terms of the hierarchy of concepts rather than having a human being satisfy all the knowledge the computer needs. (Mohri, Rostamizadeh, Sutton, Barto, & Charniak, 2018). Deep Learning is slowly gaining popularity in today’s world and the number of applications of the technology continues to rise on a daily basis. In a world of such an advancement in technology, Deep Learning is useful in helping us determine how best to use technology to our own benefits. Therefore, this paper will discuss Who, When and How the Development of Deep Thinking came out: Advantages and Disadvantages of Deep Learning, How Deep Learning Works, and The Future of the Technology.
Introduction
The field of Information Technology is the most radiant, with new technological advancements every now and then. Most of these technologies are developed with an aim of addressing a specific challenge that humans face on their day to day activities such as keeping financial records and agribusiness. The technology is always evolving, with each successive technology being an improvement of the prior one. The role of these technological advancements in the current world cannot be overlooked since humans are now more dependent on technology than ever. Among the most recent technologies to be released include the deep learning technology that is slowly gaining popularity in today’s world and the number of applications of the technology continues to rise on a daily basis. Other recent technological advancements include the user behavior analytics that learns the behavior of one user of a technology. The technology is effective in determining in case another individual gains access to technology by observing the changes in user behavior. It is therefore important to learn these technologies as such information is key to understanding how to use the technology to our benefits and also how to harness the technology for our own use.
Who, When and How the Development Came about
Deep learning was developed and first used back in 2015 by deep instinct, a technological company that specifies in offering cyber security, in cyber security. Deep learning or the deep neural networks is based on Artificial Intelligence that was developed in back in the 1950’s. There have been three waves of development of deep learning: the first wave was known as cybernetics in the 1940-1960s, the second wave was connectionism in the 1980 and 1990s while the current one began in 2006 (Goodfellow, Bengio ; Courville, 2017). In regard to the subset of AI Deep learning is the most advanced. It is an improvement of machine learning but. It goes beyond machine learning and borrows heavily from the way that the human brain functions. Deep learning has algorithms that mimic the biological structure of the human brain. The technology is the only method that is capable of training on raw data as compared to machine learning that only depends on processed data (Kaftzan, 2018). It also relies on a vast base of information that makes it more reliable. The development of deep learning was fueled by the improvement in algorithms and the use of Graphics Processing Units (GPU). Before the improvement in algorithms, only shallow neural network could be trained but with the improvement, training of deep neural networks is now possible. The deeper the neural network, the more complex and refined the process of data processing becomes. Deep learning is commercially used in a number of fields such as the development of autonomous cars, agriculture and image recognition.
Advantages and Disadvantages of Deep Learning.
The main advantage of deep learning is that it allows for advanced analysis. Advanced analysis is as a result of improved data processing models that leads to the generation of highly reliable results. Additionally, deep learning allows for unsupervised learning techniques that make the system to become better on its own, as compared to machine learning that only works with labeled data. By the system becoming better on its own, the ability to determine the crucial features enables the system to come up with reliable and precise results.
Moreover, deep learning is able to generate new features from the limited number of features that are provided in the training dataset. As a result, the algorithm is able to come up with new tasks that are more effective in solving the current tasks. The ability to come up with new tasks without human intervention allows scientists to save time that they would otherwise use in creating new tasks.
Deep learning also depends on architectures that require minimal supervision as compared to the traditional machine learning. These architectures allow the system to learn automatically from the data provided. Deep learning systems can also be trained as generative processes that can generate both outputs and inputs from a large data set. Deep learning is also used in cybersecurity. The key advantage here is that the deep neural networks are able to go through all the bites in a file without missing or ignoring any detail (Daly, 2018)
One of the disadvantages of deep learning is that most of the systems require supervised learning since they are not capable of learning on their own. Also, there is the disadvantage of adversarial input or feature, in which adjusting the input image in a specific way easily fools the deep learning neural network. The deep learning software is incapable of justifying the results or providing arguments on how and why it has come to a certain conclusion. One cannot follow an algorithm that is used by the software to determine the reason behind the conclusion. The lack of arguments or an algorithm that can be reviewed to determine the reasoning behind the conclusion means that in case of an error, one has to revise the whole algorithm that is both tiring and time-consuming. (Vieira ; Ribeiro 2018).
Deep learning is resource intensive. It requires high-performance graphics processing units, very powerful GPUs and sizeable amounts of storage space, at least 12 GB, (Perez, 2017) to store the vast amounts of data required to train the models. Also, the technology requires more training time as compared to the traditional machine learning. Furthermore, deep learning is dependent on analyzing vast amounts of data, especially during training, but the data is always constantly changing. Therefore, scientists are forced to adopt the deep learning algorithms in a way that the large amounts of continuous data can be easily handled by the neural networks (Vieira, ; Ribeiro, 2018)
How Deep Learning Works
Deep learning is a machine learning method that is used to train an AI to predict the output based on a specific set of inputs. Deep learning heavily borrows from the brain of an AI that has neurons, such as a human brain. The neurons are grouped into three types of layers: an input layer, hidden layer, and the output layer. The input layer is responsible for receiving the input data. The hidden layers perform complex mathematical calculations based on the inputs provided. The deep learning technology is characterized by having more than one hidden layer. Upon computation, the output layer is the one that returns the output data. Each neuron has an activation function that standardizes the output from the neuron.
Training the AI is quite a complicated process as it requires large datasets and also large amounts of computational powers. The first step involves giving the inputs from the data set. The inputs are then compared with the outputs. At first, the outputs will be wrong as the AI is still untrained. After going through the data set, one then creates a cost function that is used to show how wrong the AIs outputs were from the actual outputs (Skansi, 2018). The ideal cost function should be equal to zero. In order to reduce the cost functions, one changes the weight between the neurons by using a technique known as the gradient descent. Gradient descent is used to obtain the minimum of a function by changing the weights in small increments after changing the data set (Skansi, 2018). The gradient obtained from the cost functions is then used to show the direction that leads to the minimum of the cost function that requires large sets of data. The deep learning systems update the weights using gradient descent automatically, thereby simplifying the entire process. After training the AI, then deep learning can now be used to predict certain variables based on the input information.
The Future of the Technology
In the future, some changes will definitely occur in the technology, given that the technology is still in its baby stages. Even though deep learning has had its bright moments, there are still many more improvements that scientists are working on so as to make it more credible and reliable. Among the problems that users of the deep learning technology users experience are the bias problem, that has seen some of the companies using it ditch it. Among the companies that have ditched the technology is Amazon, as it favored the male over female candidates in the industry, especially those seeking for employment (Wallace, 2018). Another example is on an application that used to make people more attractive by making them whiter (Wallace, 2018). Such forms of bias are common and scientists are currently working on how to eliminate the bias, even though the bias is primarily based on people’s behavior.
Additionally, most of the input data obtained is rarely used for deep learning processes. In the future, unsupervised learning will enable the technology to obtain its own data and use it to come up with more accurate conclusions. Also, for deep learning to evolve, scientists are working on unsupervised learning techniques that are less dependent on the input data but will be able to copy the learning behavior of humans.
Conclusion
Learning about the technologies is useful in helping us determine how best to use the technology to our own benefits. The deep learning technology tries to replicate the working of the human brain in solving complex technologies. It is the most advanced AI technology there is so far. Based on how it works, the technology has a wide range of applications from autonomous cars to speech recognition that is widely used by the intelligence services. Even though technology has some limitations such as being biased, further research and development is most likely to correct such limitations that will make it more effective in the future.