Deep learning took a major step forward with the announcement that Amazon Web Services and Microsoft have teamed up to support Gluon, an open-source framework for deep learning development. Gluon’s library is designed to make it easy for developers of all skill levels to design and deploy deep learning apps for the cloud, the Internet of Things and mobile devices. Gluon is part of a broader open-source artificial intelligence (AI) initiative supported by Facebook as well as by Amazon and Microsoft. Strikingly, Apple and Google have not yet joined this effort, prompting speculations about what designs these tech giants may have on the booming deep learning market.
The dawn of deep learning is opening up new technological possibilities that will transform the future. Here’s a look at how the deep learning approach to AI works, how far this technology has already progressed and where it’s likely to head in the near future.
The Deep Learning Approach to AI
Deep learning is a specialized application of machine learning, which is a specialized application of artificial intelligence. General artificial intelligence represents the ideal of an artificial brain that would be able to simulate any cognitive task human beings can perform, including sensing, thinking and decision-making. Most AI applications are more specialized, limited to perform specific tasks such as solving mathematical or scientific problems, recognizing images from photographs, processing human speech or winning chess games.
Machine learning introduces flexibility to AI. Traditional AI applications are programmed to follow set rules. Instead of following pre-programmed rules, machine learning applies probability equations to identify mathematical trends in data. This allows for a range of models to be generated from the same set of data instead of restricting the program to a single, predetermined outcome.
Deep learning takes machine learning a step further by using artificial neural networks to guide AI learning. Just as the human nervous system can input and process sensory data, neural networks can input digital data, interpret it in mathematical or logical terms, evaluate how closely it corresponds to a desired outcome and make adjustments to generate an even closer approximation to the envisioned goal. This procedure can be used to spot current trends, predict future outcomes or suggest decisions based on desired outcomes. For instance, natural speech recognition deep learning applications can compare a speaker’s voice to a database of English vocabulary and dialects in order to identify what a speaker is saying.
The Development of Deep Learning
The concept of machine learning has been around since the 1950s, when IBM programmer Arthur Samuel introduced the term to describe an AI program he had designed to play checkers. But machine learning programs required an enormous amount of computing resources, making applications impractical until recently. It was only in the 2010s that machine learning and deep learning applications of AI became practical, thanks to advances in storage capacity, processing power and multi-processor parallelization.
The rise of the cloud helped make deep learning possible by providing access to remote resources far faster than those previously available. A major deep learning breakthrough came in 2012 when Google employee Andrew Ng used the company’s cloud server to teach a computer to recognize cat images from online videos.
Qualcomm achieved another deep learning breakthrough by introducing its Artificial Intelligence platform, which allows machine and deep learning AI applications to be run directly on mobile devices instead of relying on a cloud connection. This avoids download lag time, which empowers mobile devices to run deep learning apps such as biometric facial recognition, smart adjustment of photos and optimization of battery life.
The Future of Deep Learning
The deep learning market, already worth over a quarter of a billion dollars in 2016, is accelerating at an explosive CAGR of 52.1 percent, on track to reach over $10 billion by 2025, Grand View Research projects. Image and speech recognition, computer vision and natural language processing promise to be some of the most important applications of deep learning.
The healthcare industry will become one of the biggest adopters of deep learning, using it for purposes such as analyzing genetic data to personalize healthcare for individual patients, performing computer-assisted diagnosis based on medical images, and providing remote healthcare via data obtained from IoT sensors. Deep learning is also increasingly important in the defense industry, where it is used for purposes such as assessing battlefield data. Other important emerging deep learning applications include automotive manufacturing, financial services and data mining to optimize business performance and increase sales.
Once a remote possibility, deep learning has become a dynamic reality that is already seeing important applications in image recognition, speech recognition and other areas. As deep learning adoption becomes more widespread, its use will increasingly permeate every area of society and business, yielding deeper insights into data, making more accurate predictions and generating higher performance.