Back in 2016, Google made headlines when the company announced that it had successfully used machine learning (ML) to improve its popular Google Translate service. Thanks to deep learning, said Barak Turovsky, Google Translate’s product lead at the time, the system “has improved more in one single leap than in 10 years combined.”
Up until that point, the term “neural network” was just as likely to be heard in a medical context as in a technological one. But those in the know recognized the potential that neural networks had to completely transform the world as we know it.
In this present moment, neural networks are responsible for the development of self-driving cars, the voice-activated assistant that lives in your phone or on your countertop, as well as the complete reinvention of industries such as financial services and digital advertising.
What exactly is a neural network?
In the fields of computer science and machine learning, a neural network is typically an algorithm designed to analyze data and find the patterns hidden within.
Essentially, they can act as a sorting and labeling system for data, although their accuracy is dependent on the quality and quantity of the data they are trained on.
What makes neural networks exciting is that they can discover patterns in data that no other technique or human can.
This is why computers are suddenly recognizing faces, translating text, and processing speech much better than they ever have before.
The origin story for the discovery of neural networks as a machine learning technique is fascinating. The human brain contains billions of tiny cells called neurons, and these neurons connect together into networks.
They were first proposed by neuroscientists in the early 1940s as a framework for understanding how the brain works. In the 1950s, once computers were more advanced, several researchers began attempting to simulate neural networks in computers with various degrees of success.
In 1959, two scientists from Stanford developed the first neural network suitable for commercial use, creating a network known as MADALINE that would help eliminate echoes on telephone lines.
Despite these initial advances, neural networks never really gained wide-spread popularity in machine learning or industry – in part because they required much more computing power than was available at the time, but also because the techniques that were being used to build them were still limited.
Present day networks are (unsurprisingly) a great deal more advanced than their early predecessors. This is thanks to both the discovery of better neural network designs and methods to train them, and to the exponential increases in computational power driven by the semiconductor industry.
Modern deep learning utilizes networks with multiple layers (the more layers, the ‘deeper’ the network), which provide more accurate results, but also require vast amounts of computational power and memory.
Companies such as Intel and NVIDIA now offer processors, GPUs, and a whole host of hardware designed specifically to facilitate neural network training and implementation. Without this physical infrastructure on which to carry out deep learning, the rapid advances in neural networks that have taken place over the past decade would have been impossible to attain.
All this begs the question, why should companies invest in neural networks in the first place? What benefits do they bring, and what makes them more effective than existing processes? Moreover, what are the best applications?
To answer those questions, it must first be acknowledged that the true benefit of neural networks (and deep learning in general) is their ability to generate results to a degree of accuracy that no human could hope to achieve – specifically, their capacity to parse gigantic data sets in a relatively short amount of time and detect the patterns hidden within.
The most well-known applications for neural networks involve facial detection, image recognition, and speech-to-text or text-to-speech detection and transcription, but there are numerous other opportunities ripe for exploration.
Because of neural networks’ analytical abilities, it is possible to develop an algorithm designed to answer a specific question, whether scientific or business-related.
To give a real world example, some hospitals are using large-scale data analysis to improve diagnosis, identify high-risk patients, and determine areas of improvement.
Other businesses have used neural networks to improve their advertising efficacy by better identifying potential prospects and determining the ideal situations to approach them.
Every company, no matter how big or small, has the ability to incorporate neural networks into their organization in some way. Whether it is by offering a chatbot that customers can interact with or utilizing analytical tools to identify previously unforeseen opportunities, neural networks are fast becoming an integral part of today’s landscape.
Jeremy Fain is the CEO and co-founder of Cognitiv, the first marketing AI company to offer plug-and-play deep learning products that enable marketers to improve results through custom algorithms. Cognitiv’s award-winning technology creates and executes self-learning, fully automated deep neural networks for multi-touch, full-funnel marketing campaigns.Reblogged 6 months ago from www.clickz.com