Vision and Language
Hyundai Big Data and artificial intelligence technology are two of the most blooming fields of vision and language. In-depth neural network technology has revolutionized the visual perception and language understanding of computers.
The field of research on how machines process human language is called ‘natural language processing’. With the development of machine learning, various processing technologies have been developed in the field of natural language processing. One of the notable technologies is ‘word twovec’.
Wordtubeck is an algorithm developed by research team led by Google’s Tomas Mikolov. It works by translating words into vectors as they appear in the name. That is, it converts human horses into vectors that machines can understand and manipulate. A vector is an element with direction and magnitude in vector space, which can be expressed as a vector on a three-dimensional space, for example, how strongly a force acts in some direction.
Word-to-word converts words into these vectors, which have a vector space of up to hundreds of dimensions that we can not imagine. One of the good things about expressing a horse in this vector is that it can understand the context in which the machine is spoken (or works).
How useful and peculiar it is to be able to understand people. But the way a machine understands man ‘s words is different from ours because it is his own way. These differences often produce interesting results.
An iOS developer, who writes an online name called GraceAvery, was featured on his blog last month with an interesting example of a human word that Wordtubeck machines understand. The word Wu-Tube engine has a fresh interpretation of human speech so that it can be called ‘machine humor.’
Before looking at the content, it is necessary to mention a simple example and a formula first. This is a representative example of how Wu-Tuck-Pek understands the context of human speech. Let’s say you have the question ‘France: Paris = South Korea: ??’ We can easily say “Seoul” but it is not easy to expect answers that understand this context to a machine that interprets the given formula like 1 + 1 = 2. By the way, I am able to deduce the answer to the question even if there is no formula in the Ward-to-Peke machine which learns language with vast amount of documents. It is Wardtubeck’s ability to deduce ‘Korea: Seoul’ in ‘France: Paris’. It is expressed in other ways as ‘France + Paris – Korea = Seoul’.
Grace Avery’s Words of the Week textbook is 3 million words translated into a 300-dimensional vector published by Google News. So people who do not speak English as their first language are only interested. We are ready now. Let’s take a look at the way human language is understood by a machine, which is a bit odd and somewhat philosophical.