PCGS LAUNCHES AI-POWERED COIN GRADING TECHNOLOGY

1630Boston

@ricko et al, if this was posted here before, I apologize, I just stumbled upon this.
Coin Grading Giant Partners with Positronic to Develop Machine Learning Tool

PCGS Gold Shield highlights questionable “keypoints” harvested from their vast proprietary imaging database.
.

Earlier this year, Professional Coin Grading Service rolled out PCGS Gold Shield™ Service utilizing new coin grading technology developed in partnership with Positronic. Coin grading is the process of determining the grade or condition of a coin in order to determine its value. PCGS is one of the top coin grading services in the United States, providing the standard for coin authenticity.

PCGS graders have long been considered the best in the business - setting the bar for finding counterfeits and consistently grading coins with accuracy. Since 1986, they have graded more than 37 million coins. As they looked to serve their customers with a higher degree of accuracy and a quicker turnaround on coin grades, their team knew that upgraded machine learning technology and artificial intelligence would be key.

In partnering with Positronic, PCGS has been able to decrease the amount of time it takes to grade coins while increasing accuracy of their grading. The new system Their new system benefits from artificial intelligence and machine learning, allowing coin graders to quickly determine coin authenticity by checking coins against their vast proprietary imaging database.

“We have been testing and refining the system for over two years and will continue to do so for the foreseeable future. It will become a crucial component of our fight against counterfeiting, " shared Don Willis, President of PCGS

John Nelson, PCGS’ Chief Technology Officer, shared: “With coins going through PCGS Gold Shield, our system is even more intelligent, and faster than before. It speeds the grading and authentication process.”

Positronic specializes in making businesses perform better and faster through strategic implementation of technology. Is there an area of your business that you need to up your game to stay competitive? Contact us to explore solutions.

Jimnight

Wow ... it's hard for me to get my head around just how far grading has come.

PerryHall

How old is that press release?

BryceM

AI recognition of certain patterns will be really helpful in the fight against counterfeits too.

MrEureka

@PerryHall said:
How old is that press release?

2+ years old.

https://www.positronic.ai/blog/2018/5/10/pcgs-partners-with-positronic-to-develop-coin-grading-technology

CoinJunkie

I found the article through a Google search. It was dated Sept. 13, 2018.

chesterb

They have been doing this in certain radiology studies for a long time. It makes sense to adapt it for the coin industry. It would make finding counterfeits easier too.

TurtleCat

Without more insight into how it is actually used, it doesn’t mean a whole lot.

Kudbegud

If they are using it in a test mode I'd like to know how the machine compares to the human. What is the percentage of matching grades and who was the leader?

ricko

I knew it was being investigated....and it is the future of coin grading/authentication. I am not surprised it is two years down the road already and still a long way to go. The programming etc. is a major task. Cheers, RickO

PacificWR

Glad to see this. I use gold shield a lot.

1630Boston

@MrEureka said:

@PerryHall said:
How old is that press release?

2+ years old.

https://www.positronic.ai/blog/2018/5/10/pcgs-partners-with-positronic-to-develop-coin-grading-technology

I just found it and thought others might like to read it
If you knew about it I apologize.

Lakesammman

Don Willis - name sounds familiar .......... as does the technology.

Justacommeman

@Lakesammman said:
Don Willis - name sounds familiar .......... as does the technology.

Coin bingo

m

BryceM

Looks like a paint-by-numbers guide for a new US mint product.

mannie gray

Who's Al?

roadrunner

@mannie gray said:
Who's Al?

The guy who invented the internet.

1630Boston

Artificial Intelligence / Machine Learning

Jimnight

@Lakesammman said:
Don Willis - name sounds familiar .......... as does the technology.

What the heck does all that stuff mean?

PerryHall

@mannie gray said:
Who's Al?

Assuming that you may be serious, AI is Artificial Intelligence.

PerryHall

Since eye appeal is a grading factor, how does the computer evaluate the coins eye appeal?
I remember back a couple decades ago PCGS invested a lot of money and effort in developing a system of computer grading that went no where and they eventually cut their loses and gave up on it.

1630Boston

@PerryHall said:

how does the computer evaluate the coins eye appeal?
.
Someday, the "definition" of eye appeal described in the below video will be codified.
Think how many million/billion lines of code get processed every time you start your car.
.
https://youtu.be/1UcNbPbu9u8

neildrobertson

@1630Boston said:

@PerryHall said:

how does the computer evaluate the coins eye appeal?
.

I'm of the opinion that a grading system that is 100% consistent and reliable that imperfectly captures eye appeal is preferential to a system that is less consistent that captures eye appeal better. I would even be open to dropping the term eye appeal all-together and just referring to things like brightness, reflectivity, breaks in luster and things like that.

keets

Who's Al?

I am he!! here is my uninformed thought on this.

I wouldn't expect that this system, if it's in place at PCGS, is used to grade coins in lieu of a grading team. more than likely it's used to find counterfiets and to provide a baseline analysis of a coin's grade to speed up the grading room. I would expect that all the information the system provides plus the actual grade assigned by PCGS is compared.

neildrobertson

@keets said:
I wouldn't expect that this system, if it's in place at PCGS, is used to grade coins in lieu of a grading team. more than likely it's used to find counterfiets and to provide a baseline analysis of a coin's grade to speed up the grading room. I would expect that all the information the system provides plus the actual grade assigned by PCGS is compared.

There are grade limiting factors relating to size of marks and location of marks. You could hypothetically create something that counts that type of stuff and measures the relative size of marks to set some grade boundaries.

Catbert

Some machine learning methods:

Machine learning algorithms are often categorized as supervised or unsupervised.

• Supervised machine learning algorithms can apply what has been learned in the past to new data using labeled examples to predict future events. Starting from the analysis of a known training dataset, the learning algorithm produces an inferred function to make predictions about the output values. The system is able to provide targets for any new input after sufficient training. The learning algorithm can also compare its output with the correct, intended output and find errors in order to modify the model accordingly.

• In contrast, unsupervised machine learning algorithms are used when the information used to train is neither classified nor labeled. Unsupervised learning studies how systems can infer a function to describe a hidden structure from unlabeled data. The system doesn’t figure out the right output, but it explores the data and can draw inferences from datasets to describe hidden structures from unlabeled data.

• Semi-supervised machine learning algorithms fall somewhere in between supervised and unsupervised learning, since they use both labeled and unlabeled data for training – typically a small amount of labeled data and a large amount of unlabeled data. The systems that use this method are able to considerably improve learning accuracy. Usually, semi-supervised learning is chosen when the acquired labeled data requires skilled and relevant resources in order to train it / learn from it. Otherwise, acquiring unlabeled data generally doesn’t require additional resources.

• Reinforcement machine learning algorithms is a learning method that interacts with its environment by producing actions and discovers errors or rewards. Trial and error search and delayed reward are the most relevant characteristics of reinforcement learning. This method allows machines and software agents to automatically determine the ideal behavior within a specific context in order to maximize its performance. Simple reward feedback is required for the agent to learn which action is best; this is known as the reinforcement signal.

Machine learning enables analysis of massive quantities of data. While it generally delivers faster, more accurate results in order to identify profitable opportunities or dangerous risks, it may also require additional time and resources to train it properly. Combining machine learning with AI and cognitive technologies can make it even more effective in processing large volumes of information.

https://expertsystem.com/machine-learning-definition/#:~:text=A%20definition.%20Machine%20learning%20is%20an%20application%20of,access%20data%20and%20use%20it%20learn%20for%20themselves.

keets

just so there's no mis-understanding, I don't think anything but humans is actually grading our coins.

1630Boston

It is not AL **it is **Ai

logger7

I can see it for screening coins but not final grading; some AI system is or should be in use for huge numbers of moderns for obvious reasons with a finalizer making sure results were accurate.

Jimnight

Nice link.

Thank you ... now I know ( AI )

TwoSides2aCoin

Cool. AI, meet AT and AU

BLUEJAYWAY

@Lakesammman said:
Don Willis - name sounds familiar .......... as does the technology.

This image reminds me of that medical mock up example of a human head that defines all the areas of the human brain. That denotes what each part of the brains functions is.

dpoole

@PerryHall said:
Since eye appeal is a grading factor, how does the computer evaluate the coins eye appeal?
I remember back a couple decades ago PCGS invested a lot of money and effort in developing a system of computer grading that went no where and they eventually cut their loses and gave up on it.

I have long believed there was merit in bringing some objectivity to grading. AI has the potential to do this, presumably more than when PCGS contemplated it years ago (what do I know? I'm a Liberal Arts major). Garbage in/garbage out, and obviously there is the issue of input and training for AI, but stability is the desired outcome, along with the banishment of grade inflation.

Yes, there's no way to convincingly substitute for something as subjective as eye appeal. And if people still need a human eye to tag on some endorsement of eye appeal to an individual coin, than so be it.

But as far as objective standards for what grade corresponds to wear, marks, cuts, cameo contrast etc. etc., IMHO bring it on.

Kudbegud

I'd like to see a grade off like a boxing match. Different coin each round with a human PCGS finalizer grading pro as the referee. Id'd watch that.

BuffaloIronTail

@Jimnight said:
Wow ... it's hard for me to get my head around just how far grading has come.

Open the pod bay doors, Hal.

Pete

mothra454

@dpoole said:

@PerryHall said:
Since eye appeal is a grading factor, how does the computer evaluate the coins eye appeal?
I remember back a couple decades ago PCGS invested a lot of money and effort in developing a system of computer grading that went no where and they eventually cut their loses and gave up on it.

I have long believed there was merit in bringing some objectivity to grading. AI has the potential to do this, presumably more than when PCGS contemplated it years ago (what do I know? I'm a Liberal Arts major). Garbage in/garbage out, and obviously there is the issue of input and training for AI, but stability is the desired outcome, along with the banishment of grade inflation.

Yes, there's no way to convincingly substitute for something as subjective as eye appeal. And if people still need a human eye to tag on some endorsement of eye appeal to an individual coin, than so be it.

But as far as objective standards for what grade corresponds to wear, marks, cuts, cameo contrast etc. etc., IMHO bring it on.

Great points. There may be a role for the AI bot on the grading side as well, since technology has advanced so much in the last 20 years. To make an analogy to another realm, there was recently a team that created a poker playing AI bot that essentially now plays "perfect" limit Texas hold 'em vs. one other player. They did this by using a "regret minimization" algorithm and letting it play billions of hands. At first it lost tons of (fake) money, but kept adapting to minimize its regret for each play until it basically played perfectly.

Not saying a bot could ever take over the grading duties entirely but once it learns enough maybe it could do initial grading that's reviewed by humans to ensure accuracy?

1630Boston

LanceNewmanOCC

so many topics on this forum !!!

Catbert

Alert - 2 year old thread.

Looks like the technology was used to streamline image processing as per the positronic website:

https://positronic.ai/case-studies/coin-process-automation/

I see no reference that the technology is being used currently for grading purposes. FYI.

Catbert

Yet also for counterfeit detection:

https://positronic.ai/case-studies/counterfeit-detection/://

neildrobertson

Those compute times are longer than I would have expected. However, 6 minutes of processor time is still way cheaper than 6 minutes of people time and you can buy more processers for fairly cheap.