Discussion in 'Coin Chat' started by JCro57, Feb 25, 2018.
How can you measure accuracy when there's no ground truth?
Log in or Sign up to hide this ad.
Ground truth is assigned grades by a panel of graders, the same as it is now. A portion of an assigned grade is eye appeal. You can never suit everyone but with averages, you can get close, just the same as AI has been used to predict the weather, elections and believe it or not, the next suit of clothes that you will buy.
Sorta what I've been doing with CAD for awhile - except I am not using it for grading - just for the purpose of identifying dies. I unofficially call it ANDIS called "and is" (Automated Numismatic Die Identification System). Yes my system is kinda crude - just don't have the expertise to take it to the next level.
Any work on early large cents?
Not a whole lot - here are some
Very cool; I can recognize the varieties by the images- will take the time later to overlay a couple on genuine images I have saved. Thanks justafarmer!
No they didn't, but that's because there was no reason to adjust them. The mintage numbers are a matter of fact, and unchangeable. What you're trying to do is to compare mintage numbers to census numbers - and they are two completely different things. Mintage are how many were made to begin with while census numbers are how many still exist.
There are very, very accurate records for mintage numbers. There are no records for census numbers. When it comes to census numbers the best anyone can do is guess at them.
Now about this idea of computer grading, there is one very good reason why the TPGs will never employ it. And it doesn't have anything to do with whether or not a computer can do it. Just for the sake of argument let's say with absolute certainty that a computer could do it - let's just stipulate that as a fact. The TPGs would still never use it.
Do you know why ? Because it would put them out of business in a very, very, short time. Because once graded by the computer there is absolutely no reason for any coin to ever be re-graded again.
The system that they have is the only thing there is that keeps them in business. They are not ever going to change that.
Exactly. And that is all the grading services can do. Once the coin leaves their possession, they lose control of what happens to it. So they are very similar.
I partially agree with this. Algorithms and techniques used by computers are not perfect. I spent much of my career developing new algorithms and techniques to process meteorological data. These algorithms and techniques lead to newer and better algorithms and techniques. Software to analyze and grade coins would be no different. Also, humans are the ones who come up with the rules so as improvements are made to imaging coins (quality, resolution), rules may also change accordingly.
That assumption is predicated on the system being right the first time for all coins and grading practices never changing. Fluid grading practices can be implemented in a computerized grading system as easily as in the current system. It can also easily be sold as "bug fixing," "fine tuning" or "keeping current," all of which are legitimate, and all will mean, "you might want to resubmit to make sure your coins are graded using the latest technology." People will also resubmit to probe the system, then word will get out about which series is loose this week.
You're probably right John, one way or another they'd probably figure out how to keep changing the grading standards just like have been doing for years - and getting away with it. And getting people to buy into their new nonsense just like have been getting people to buy into their current nonsense.
My degrees are also in criminology. In addition, I judge our regional science fairs. And, when I worked at Coin World 1999-2000, we also had a working file on "computer grading" just in case it actually became real enough to write about. It had been tried perhaps a decade earlier, but with inconclusive results.
At the last science fair, one of the entrants was on facial recognition, testing a common program against humans for the correct recognition of emotion. The program won.
That said, you may be missing the point, however, if you want to prevent upgrades. We really do not want our coins to be objectively graded. We want them to be optimistically graded in our favor. Perhaps my favorite example is the famous Byron Reed holding of the Parmelee 1804 Dollar. It was held on display in a museum for many years. It toned naturally... and the curator would wipe off the toning... and the coin upgraded ... Just sayin... The purpose of the Red Book is to allow us to show our wives how clever we are when we spend $100 on a coin valued at $138.
You cannot take coin collecting too seriously. It is a hobby. You can make money at it, just as a carnival sideshow is an honest living that sells entertainment. The people walking the fairway with teddy bears proves that.
Computerized coin grading has been worked on for at least 25 years. Facial recognition software is now more accurate than humans looking at other humans. As I said above, I just judged a high school science fair where this was an entrant. The kid did not develop the software, just downloaded it, to test people against it. The future is here.
Speaking for me, I would like to see the recognition software used to screen out the latest deceptive struck counterfeits out there; I would think it embarrassing to actually certify and holder them...
At the 2017 Emergency Management Conference in San Antonio, I met several free market weather forecasting services. They are amazingly accurate. You get what you pay for and most of us rely on free (government) services, which are close enough for a broad picture. I did some research into the development of weather forecasting software and it was one of the very first applications soon after gunnery. 16k bits were enough to get started on the problem.
As for coin grading, there is no challenge to objective recognition, but no one really wants that.
PCGS is doing that with the gold shield service already. They announced late last year their AI program to help with counterfeits and pointing out abnormalities/areas of concern. NGC is a lot more hush hush about what they use but if they don’t have something similar I would assume they’re working on it
8 of my struck fake early coppers are in TPG holders; hopefully this latest technology will make a difference.
The development of meteorological algorithms isn't limited to weather forecasting. My area of expertise was in weather radar research and development. This area has evolved significantly over the past 50 years in both hardware and software. New data processing techniques have been valuable in identifying signals in the data that can be used to detect various phenomena.
How does this relate to numismatics? As technology evolves, one would expect information about a specific coin to increase. How far can this go and what effect would this have on grading a coin? It all depends on whether the new information would have an effect on grade and whether the grading rules change based on this new information.
Separate names with a comma.