Slashdot Mirror
Astronomers Teach a Machine To Analyze Space Images
New submitter Jim Geach writes: Our team of astronomers and computer scientists has developed a novel unsupervised machine learning algorithm — a combination of Growing Neural Gas and Hierarchical Clustering — to automatically analyze astronomical images. In effect, the algorithm performs the same task as a human 'eyeballing' an image, automatically identifying and labeling the points of interest. We're aiming to deploy the algorithm on the next generation of astronomical surveys such as LSST and Euclid where no human, or even group of humans, could closely inspect every piece of data. The algorithm could also find application in other fields, such as medical imaging and early disease diagnosis. The results are being presented at the UK National Astronomy Meeting in Wales, and the details of the algorithm are described in this paper.
The universe is Waldo.
http://bravestwarriors.wikia.c...
https://www.youtube.com/watch?...
In this regard, I think so.
For a computer to look at an image as a whole and not bit-by-bit, the software behind image recognition must be amazing.
I'm impressed with every one of these stories.
> Growing Neural Gas
My doctor diagnosed me with this too.
Do you want Skynet? Because this is how you get Skynet.
Get free satoshi (Bitcoin) and Dogecoins
So this is built on the same principles of the Google image tagging algorithm that decided Blacks were Gorillas. This should prove entertaining even if it fails to be useful.
The algorithm could also find application in other fields, such as medical imaging and early disease diagnosis.
Radiologists already use software that assists in scanning images for potentially interesting features. They aren't a replacement but they apparently do a fairly good job at helping to ensure as little as possible gets overlooked. I did some consulting work in a radiology clinic some years ago and they used this technology there to good effect.
I wouldn't be surprised to see anatomic pathologists using technology like this somewhere in the future. The logistics of it are much more complicated than for radiology but I think somewhere down the line it will probably happen.
So basically a pre-scan filter that leaves all questionable findings to the experts (human) for further review.
Yes but it also serves an error checking function. Sometimes humans overlook things quite by accident and it provides as way to help ensure that an unblinking set of eyes looks things over. Sometimes these systems flag things that the doctor's miss. (and vice-versa) Both human and machine are pretty good individually but together the results are even better.
I believe these folks have been doing this for years. They even have been a participant in Google Summer of Code. They gave a presentation on how they could identify objects from cell phone pictures.
The problem is that GZ relies on a user base that wants to look at the interesting objects - this will look at all the boring stuff too. Plus, data rates of LSST will be 10s TB/night - this will have to be parsed extremely quickly to find transients. Can't envision crowd sourcing doing that.
Test it on this:
http://apod.nasa.gov/apod/ap03...
Table-ized A.I.
Gorillas can't fly. And there are no black people on ISS right now.
"So long and thanks for all the fish."