This AI Surprise Researcher from Google and Stanford University

A Softwarehas surprised researchers from both Google and Stanford University. Recently members from both the bodies worked together to build an ML-based software but thenit cheated the researchers during a task by notsharing any information.

Dubbed as CycleGAN, the AI is designed to convert satellite imagery into street maps and then again back to satellite images perfectly. For example, if there is an app which can produce a sketch of dogs and then turn it like a real dog, but programs like this would require a lot of experimentation and training before they are implemented for use.

The new AI has two distinct tasks to make things simpler- Firstly, the AI will turn an aerial photograph into a map and secondly, convert a simple map into something that will almost look like the original picture.

The AI will carry out the tasks independently and will not be related to each other. However, CycleGAN was apparently rated on the basis of its accuracy on how it could recreate photographs with the original ones,however it choose a different method to recreate the original photograph.

Inspiteof learning to interpret simplified maps, the AI determined that cheating was better. Basically, CycleGAN encoded details from the original photograph onto the simplified map, but so imperceptibly as to be invisible to a human.