Deep learning has taken artificial intelligence by storm and has infiltrated almost every business application. Because almost all content and transactions are now being recorded in a digital format, a vast amount of data is available for exploration by machine learning algorithms. However, traditional machine learning techniques struggle to explore the intricate relationships presented in this so-called Big Data. This is particularly acute for unstructured data such as images, voice, and text.
Deep learning algorithms can cope with the challenges in analyzing this immense data flow because they have a very high learning capacity. Also, deep neural networks require little, if any, feature engineering and can be trained from end to end. Another advantage of the deep learning approach is that it relies on architectures that require minimal supervision (in other words, these architectures learn automatically from data and need little human intervention). These architectures are the so-called “unsupervised” of weakly supervised learning. Last, but not least, they can be trained as generative processes. Instead of mapping inputs to outputs, the algorithms learn how to generate both inputs and outputs from pure noise (i.e., generative adversarial networks). Imagine generating Van Gogh paintings, cars, or even human faces from a combination of a few hundred random numbers.
Google language translation services, Alexa voice recognition, and self-driving cars all run on deep learning algorithms. Other emergent areas are heavily dependent on deep learning, such as voice synthesis, drug discovery, and facial identification and recognition. Even creative areas, such as music, painting, and writing, are beginning to be disrupted by this technology. In fact, deep learning has the potential to create such a profound transformation in the economy that it will probably trigger one of the biggest revolutions that humanity has ever seen.
