Click any word in a definition or example to find the entry for that word
a form of machine learning (= a process in which computers formulate problem-solving strategies by learning from large amounts of data) which incorporates the use of neural networks (= computer systems designed to work in the same way as the human brain)
'Over the last half-dozen years, deep learning, a branch of artificial intelligence inspired by the structure of the human brain, has made enormous strides in giving machines the ability to intuit the physical world.'Harvard Business Review 30th January 2017
The human brain is a strange and wonderful thing, an intricate biological network we're born with which gives us the cognitive fuel to effortlessly distinguish between horses and cows, cars and trains, happy and sad faces. Unfortunately, the network of electronic connections commonly known as a computer isn't anywhere near as good at this kind of stuff, and experts across the globe have consequently spent decades trying to work out if it is in fact possible to emulate human perception within an electronic box. Whilst we've seen amazing advances in the capacity to store and process information, artificial intelligence has always lagged behind, but in recent years there's been a quiet buzz around a concept which looks like it could finally be a game-changer – a branch of AI known as deep learning.
deep learning techniques have been used in the experimental development of systems which can diagnose skin cancer
An explanation of what is meant by deep learning crucially depends on an understanding of the term machine learning. This expression is often misinterpreted, and in fact does not refer to using machines 'for' learning (compare mobile learning, which is the use of handheld computers or mobile phones for learning), but rather describes a concept in which the machine is, itself, the learner. In machine learning, computers are fed vast amounts of data which by clever use of algorithms they analyse and 'learn' from. In other words, the computer is effectively 'trained' by the data, enabling it to learn how to perform a particular task without being specifically programmed to do so. Deep learning goes one step further by combining machine learning with neural networks, computational models of the brain which enable data to be classified and interpreted more effectively through mathematical techniques like weighting and probability.
Such concepts sound rather complex and sophisticated outside of the expert domain, but their application in the real world is proving pretty exciting. At Stanford University in the USA for instance, deep learning techniques have been used in the experimental development of systems which can diagnose skin cancer. Making a database of 130,000 skin disease images, researchers were able to train an algorithm to correctly identify whether lesions were benign or malignant, and even succeed in matching the performance of real dermatologists. Another intriguing area where deep learning is making in impact is in the computational recognition of emotions. Though a computer can never of course have an intrinsic sense of how we're feeling, through the input of vast amounts of image data it can be trained to recognise the intricacies of visual cues like facial expressions, and thereby identify smiles, frowns, etc, from which it can draw general conclusions about our likely mood. If you'd like to know more, check out this fascinating TED talk by computer scientist Rana el Kaliouby.
Research into the principles underlying machine learning and deep learning dates right back to the 1950s, but the terms themselves only really emerged in the 2000s alongside revolutionary advances in computing and the setting aside of earlier scepticism about data-driven, statistical approaches to AI. Deep learning and machine learning techniques have also been effectively explored in the world of ELT, as in the brilliantly innovative Write & Improve from Cambridge English, which has used learner corpus data to inform a system which can automatically identify errors in students' writing and give feedback.
Read last month's BuzzWord. moonbow.
This article was first published 11th April 2017.
A must for anyone with an interest in the changing face of language. The Macmillan Dictionary blog explores English as it is spoken around the world today.global English and language change from our blog