Image creation in computers resembles human dreams where we sometimes imagine things that we have not actually experienced. This is the conclusion of a research collaboration between DTU and MIT (Massachusetts Institute of Technology) in the US.
.
The researchers in the project are working to improve artificial intelligence based on so-called deep learning. They work to make computers imagine new images on the basis of information from images uploaded to computers.
Professor Lars Kai Hansen and Postdoc Søren Hauberg from DTU Compute are behind the Danish part of the new project finding which has just been presented at the ‘Artificial Intelligence and Statistics’ conference in Cadiz in Spain.
“The aim of the research is to make image recognition stronger when only a few images are available. This will—in particular—have a large impact on the analysis of medical scan images of which usually only a very limited number is available,” says Lars Kai Hansen.
Deep learning is a technique employing artificial intelligence to train simulated brains, so-called 'neural networks'.
Even though similar algorithms have existed for a long time, it was not until recently that it became possible to train networks with many layers of simulated neurons, the so-called deep networks. Faster computers and big data are two factors which have made deep learning possible.