Earlier magnetic resonance machines were narrow and sounded huge. They were considered the most important machine, yet those scans took a minimum of an hour based on the part of the body that is being scanned.
At recent times, researchers have come up with new algorithms for neural networks that can used with machine learning techniques to create useable images with comparatively less data. These algorithms look like, “puzzles that begs solution” to Paul Hand, Northeastern University assistant professor.
“We don’t have good justifications for why these neural networks tend to work,” says Hand
Earlier computer techniques researchers would destine the important aspect of an image through algorithms. In case, if the researcher wanted to define a characteristic of an animal, he programs an algorithm accordingly.
Researchers indeed train neural networks using tons of data that involved practice. It is said that every single time the network comes up with an right answer no matter it is an accurate reconstruction of MRI image or detecting a cat, the machine is said to learn the stuff that it has to reflect in the future as the right answer which humans expect.
But it was understood that machines where better in recognizing images than humans. The neural networks in machine were designed to educate themselves about the important characteristic of the particular animal.
Hence Paul hand have decided to work on algorithms that can be used to reconstruct images from microscopes, astronomical data or medical technology based on how they are trained.
Hand says, “The thrill of the profession is that you get harder and harder puzzles.”
Further Hand and his colleagues were able to come up with the first mathematical proof that neural networks can use to recover the images from the least data. He also believes in his future exploration can come up with a better technology which combines both computer science and applied mathematics.
Justin Aaron