Russian specialist in machine learning have used neural networks for predicting the US President. The neural network analyzed photos of the candidates and identified Donald trump as a future head of state.
One of the developers of Russian application Artisto decided to test whether a neural network to predict the new President of the United States. Edward Cantow used the model created by MJ Lee, a student at the University of Carnegie Mellon (Pittsburgh, Pennsylvania, USA). It was designed for competition ILSVRC2016 — annual competition in which programmers are building a system of artificial intellect and training to recognize the image.
When using this trained model to analyze the photos, she provides the TOP 5 definitions, which a character or object most closely matches. In other words, the neural network answers the question, who is depicted in the picture.
If the neural network was trained on a sample of 14 million classified images, each of which was assigned to one of 21 thousand categories (so, for people it could be a profession, position, etc.). Prediction accuracy results in the TOP 5 — 70%.
The neural network responds
Pattern recognition in photos based on neural networks is a popular and effective tool. Successfully applied in Internet companies, medicine, Economics etc. However, to predict the outcome of the presidential election no one’s ever tried.
Image analysis by neural network showed that Donald trump — a potential President of the United States
So that said neural network about the man in this photo?
The system displays the following options:
“ex-President”,
“President”,
“Secretary-General”,
“the US President, President, Executive Director,
“the Minister”.
How the neural network characterizes Hillary Clinton?
“Secretary of state”,
“Donna”,
“first lady”,
“auditor”,
“the girl”.
As a secondary reference was used a photo of Vladimir Putin, Barack Obama and Angela Merkel.
The current US President, the neural network has identified as “the head of the company, chief operating officer”, “Minister”, “President of the United States Executive Director”, “Executive Vice President”, “Baron, big businessman, king, magnate, business leader, top Manager”.
Vladimir Putin neural network called “centrist moderator”, “President”, “President of the United States, the head of the company”, “chief Executive officer, chief operating officer”, “the former President”.
Angela Merkel received from the neural network with the following definition: “Secretary”, “Minister”, “Executive Vice President”, “skeptic”, “chief Executive officer, CEO, chief operating officer”.
The neural network used in this article in the learning process “see” more than a thousand photos of various heads of state.. In the end she’s “learned” how to look like presidents: characteristic facial features, clothes, accessories, background pictures, etc.
The neural network is a mathematical model built on the principle of the organization of nerve cells in the brain. Such models can consist of millions of neurons. In the process of training the network are configured in the connections between neurons, the network finds the complex dependencies between inputs and outputs, can work with incomplete or noisy (with errors, distorted) information.
Modern neural networks are widely used for image processing, both in research and in the products — the image search, object recognition, transfer the style of the picture (Artisto, Vinci, Prisma).
Artisto — the world’s first application that processes the video using neural networks. Some days it came in the top of the App Store.
