Why Deep Learning Represents The Future Of AI

The concept of Artificial Intelligence (AI) has taken the world of science and technology by storm. John McCarthy, one of the first AI researchers, defined it simply as “the science and engineering of making intelligent machines”. AI then, represents the attempt to use computer science to create a synthetic, improved version of the human trait of intelligence; a quality of the species that, of course, originates in the human brain. “AI” has expanded to become an all-encompassing term that includes any form of machine-based intelligence, so even a device as simple as a handheld calculator can be considered a primitive form of Artificial Intelligence. The forms of AI being implemented by modern practitioners and businesses however, have far greater capabilities and are able to mimic and even improve on several cognitive activities once considered solely the purview of human beings, such as Image Recognition, Voice-to-Text Conversion, Language Translation, Art Creation and Time-Series Predictions. These breakthroughs in AI research have primarily been achieved due to the rise and success enjoyed by the AI discipline known as Deep Learning.

Artificial Intelligence represents humanity’s attempt to use computer science principles to create a synthetic version of the human capacity for intelligence.

Deep Learning is formally defined as the approach of representing information as a nested hierarchy of concepts, with each concept defined in relation to simpler concepts, and more abstract representations computed on the basis of less abstract ones. In layman’s terms, Deep Learning essentially uses a network of nodes (called a Neural Network) to enable a computer to learn from patterns in data to generate the right output. The name “Neural” Network was given because this architecture of interconnected nodes in hidden layers loosely resembles the way the human brain learns from its stimuli as well. Deep Learning enables a computer to learn categories in an incremental manner; for example in an English sentence, the hidden layer architecture first learns to define the lowest-level categories of alphabets. In another layer, it then learns to define the combinations of low-level alphabets that form higher-level words. Finally a third layer identifies the combinations of words that form sentences, the highest-level textual concept in this example.

Deep Learning uses a network of nodes, called a Neural Network, that enables a computer to understand abstractions in the information inside datasets, to learn its patterns and generate the desired output.

Deep learning has gained immense traction over the last decade for the following three main reasons:

  • Increase in the amount of training data available.
  • Increase in computational power available at affordable rates.
  • Improvement in algorithm designs/ideas.

The increase in the amount of data available to train computers has primarily come about because of the explosive growth of the World Wide Web. The rise of social networks such as Facebook, Twitter and Instagram have further amplified this modern data deluge, while data is also being collected spontaneously by devices connected to the Internet of Things (IoT). This massive amount of data being generated is presently stored using Cloud Computing, which has provided great scalability to big data storage through its distributed nature.

The rise of Big Data has directly contributed to the success of Deep Learning in modern-day AI Research.

Computational power, as predicted by Moore’s Law, has continued to increase while becoming more affordable due to Economies of Scale. This has taken the form of advanced CPUs (Central Processing Units) and even specialized high-performance chips called GPUs (Graphical Processing Units), that are capable of the kinds of large-scale parallel processing required to run Neural Networks. Lastly, while the algorithms behind the design of these Neural Networks have existed since before the recent advent of Deep Learning, the latency and design of these algorithms have improved in recent years, and have also been adapted to take advantage of the distributed storage and increased processing power available to today’s machines.

The success of Deep Learning has also been made possible due to the increased computational power and improved algorithms available to computers today.

The end-result of this is that with the massive datasets that are accessible and the processing power available to these systems, the Deep Learning algorithms of today have enabled computers to achieve fantastic results in many difficult, human-level cognitive problems. Image Recognition solutions, using techniques from the fields of Computer Vision and Deep Learning, can accurately identify nearly every single object in images and group large numbers of similar images together. They are now even capable of generating one-sentence textual descriptions of these images with no prior training whatsoever. In addition, this technology is also being used in Medical Imaging Systems to read X-rays, MRIs and CT scans and identify abnormalities with the precision of experienced radiologists. Image Recognition is even central to the development of smart vision mobility solutions such as autonomous drones and self-driving cars, showing the wide ranging impact this kind of neural network will have on human society.

Image Recognition is one of the most successful applications of Deep Learning to real-world scenarios.

Speech Recognition is another complex problem that has seen great success with the use of Neural Networks from Deep Learning. The personal assistant solutions offered by the major tech companies, such as Siri from Apple, Alexa from Amazon and Cortana from Microsoft, can respond well to a large set of speech commands, which they can understand and execute due to the Deep Learning algorithms running in the software. This understanding of natural language is even more comprehensive for text, due to the larger amount of data available and smaller processing requirements demanded. Google Translate, for instance, provides textual translation for over a hundred different languages. The achievements of Neural Networks in the area of Natural Language Processing have helped give rise to intelligent chatbots that can understand customer or user requests in a chat interface and respond with the right message/function.

The success of Deep Learning in both text-based and voice-based Natural Language Processing, is seen as a critical step on the path to Artificial Intelligence.

The achievements of Neural Networks in all these areas provide credence to the quality of the contributions Deep Learning has made towards research in Artificial Intelligence. More recently, a new class of Neural Networks called GANs (Generative Adversarial Networks) are now being used to generate artificial content such as text, images and audio, that closely resemble properties of real versions of this content. With the proliferation of these Deep Learning-based tools into solutions for general use, Neural Networks will soon be highly ingrained into human society. Their high performance over the vast datasets available today give hope to the idea that eventually, these algorithms can learn efficiently from smaller amounts of data, and some combination of these Neural Networks from disparate areas of research can give rise to a solution approaching the general, awareness-like intelligence possessed by human beings. In other words, Deep Learning represents the furthest point humanity has reached on the path to full Artificial Intelligence, and it already possesses numerous potential applications that are transforming several modern industries in the world today.

Skcript https://www.artelliq.com/blog/why-deep-learning-represents-the-future-of-artificial-intelligence/ https://www.artelliq.com/media/heroes/Artificial-Intelligence-Machine-Learning-Deep-Learning3.png