Deep learning and AI
Deep learning, sometimes referred to as hierarchical or deep structured learning, is a machine-learning method inspired by the human brain and based on brain-like neural networks. Capable of studying and analysing data representations of images, sounds, video and other data, deep learning aims to replicate the activity that occurs in our grey matter, recognizing patterns and making sophisticated deductions. In the past 10 years, deep learning has advanced considerably, bringing the field closer to one of its original goals: producing artificial intelligence capable of making smart decisions where human capacity is lacking.
Deep learning, applied
Practical examples of deep learning in action include the ability of self-driving cars to recognize pedestrians crossing the street and to differentiate between them and other cars or buildings. Lately there has been much controversy about a handful of highly publicized accidents involving autonomous cars – while the technology isn’t perfect yet, unmanned vehicles have the potential to reduce accidents on a large scale and are far less accident-prone than their human-operated counterparts.
In agriculture, machine-learning applications are being successfully used to predict seasonal crop yield, to monitor the water levels and to detect plant diseases early on, so farmers can respond timely and adequately.
Large insurance companies are starting to employ deep learning algorithms in determining damage to cars involved in traffic accidents. Employing machine learning, it’s possible to train an algorithm to identify damaged car parts, assess the damage, make predictions about the needed repairs and estimate the costs.
In medicine, with the help of deep learning tools, imaging equipment can detect what’s important on medical scans, surveillance images and other diagnostic tests.
In 2012, Richard Rashid, Chief Research Officer at Microsoft, showcased a revolutionary deep learning-based invention at a conference in Tianjin, China. While he spoke in English, his speech was getting translated into Mandarin real-time, first appearing as subtitles on video screens and then as computer-generated voice, which had the same intonation as Rashid’s real voice.
Advanced computer vision
In 2011, there was a competition between different algorithms for recognizing traffic signs. Traditional computer vision methods were able to detect and classify traffic signs but they required considerable amount of time-consuming manual work to distinguish important features in images. The deep learning algorithm won with ease – it was twice as fast in recognizing road signs than humans, and much more reliable. Perhaps not surprisingly, many human drivers don’t know how to read the signs properly.
Computers can write
Data scientist Jeremy Howard developed a deep learning algorithm that can describe patterns without the algorithm having seen them beforehand. For the algorithm to succeed, it is necessary for it to analyse separate parts of the pattern in advance. Still, this is a revolutionary and very promising development as the algorithm is able to fully identify and describe patterns on its own.
Deep learning algorithms can identify structures independently, which is why it comes in handy in health care applications. Notably, a team of radiology specialists has successfully used the algorithm to analyse millions of tumour images, which revealed dozens of new relevant properties about tumours. As a result, radiologists can now discover and treat tumours at an earlier stage, with a subsequent higher treatment success rate.
Deep learning algorithms are becoming increasingly popular as they continue to get better and faster, and due to the ubiquity of big data. You don’t need to be a data scientist to understand how they work in a nutshell and to be able to take advantage of what they can offer. Whether you see future deep learning advancements as an opportunity for your business or a threat is up to you, of course.
Copywriter: Ina Danova