[00:00:25] Speaker ?:
I'm sorry.

[00:01:03] Speaker 02:
Next case is SAP America versus Invest Pick 2017-2008.

[00:01:09] Speaker 02:
Mr. Key.

[00:01:13] Speaker 02:
Good morning, Your Honors.

[00:01:27] Speaker 01:
May it please the Court, I'm Jay Kason for the Appellant Invest Pick.

[00:01:32] Speaker 01:
I will begin by providing a brief overview of the invention of the 2-9 patent, and then discuss why the 2-9-1 patent claims constitute eligible subject matter.

[00:01:44] Speaker 02:
Well, why don't we look at the claim, because that's what counts.

[00:01:46] Speaker 02:
The method for calculating, analyzing, and displaying data.

[00:01:51] Speaker 02:
Isn't that quintessentially abstract?

[00:01:56] Speaker 01:
The preamble, yes, Your Honor.

[00:01:58] Speaker 01:
because it's simply setting out what it does.

[00:02:01] Speaker 01:
But the claims go on to recite very specific claim limitations, such as a bias parameter.

[00:02:11] Speaker 00:
So there are a bunch of details about how the analysis takes place.

[00:02:16] Speaker 00:
That is correct.

[00:02:17] Speaker 00:
Can you end with, okay, we have a display, namely a graph, a plot of

[00:02:26] Speaker 00:
the result of the analysis, full stop, nice information.

[00:02:32] Speaker 00:
How is that not squarely in the abstract category?

[00:02:38] Speaker 00:
It doesn't even say what you do with the information, even to move it into some market choice, which I'm not sure would help, but it doesn't even do that, does it?

[00:02:48] Speaker 01:
Your Honor, this is a method for predicting how a portfolio would perform, and the method

[00:02:56] Speaker 01:
is very precise.

[00:02:58] Speaker 00:
For example, it talks about getting a... But there are lots of very precise mathematical calculations.

[00:03:07] Speaker 01:
These are not just calculations, Your Honor.

[00:03:09] Speaker 01:
This is a user-defined parameters and user-defined queries are posed to the system.

[00:03:16] Speaker 01:
So it begins with basically the user deciding the conditions under which the modeling is going to take place.

[00:03:22] Speaker 01:
And then you have the creation of a sample space

[00:03:25] Speaker 01:
And as this court has previously held in 2016, then one of the key inputs from the user is the bias parameter.

[00:03:32] Speaker 01:
And that bias parameter doesn't have anything to do with creating the sample space, which the user sets.

[00:03:37] Speaker 01:
But it has to do with modeling the conditions under which the user wants the investments to be modeled.

[00:03:44] Speaker 04:
But isn't this much like there's a garbage in, garbage out problem, and that what you all have said is that when the assumptions being used are bad,

[00:03:55] Speaker 04:
the data that gets produced are bad, or at least not good enough.

[00:04:00] Speaker 04:
And what you're doing is coming up with a better set of assumptions.

[00:04:05] Speaker 04:
And so you put those assumptions into a computer, and you get a better result at the end.

[00:04:11] Speaker 04:
But how is that an actual technological advancement, as opposed to just a brilliant idea?

[00:04:18] Speaker 04:
I mean, it sounds like a brilliant idea.

[00:04:20] Speaker 04:
Don't get me wrong.

[00:04:21] Speaker 01:
So it doesn't stop at the brilliant idea, Your Honor.

[00:04:24] Speaker 01:
It operationalizes the idea because it allows you to set the bias parameter and it is set in sample selection.

[00:04:33] Speaker 01:
It's not in creating the sample space, but it actually tells you, pick more of whatever kinds of market conditions you're interested in, the number of updays, number of downdays, and then you pick the samples

[00:04:47] Speaker 01:
based on that.

[00:04:48] Speaker 01:
So in other words, it operationalizes that idea.

[00:04:51] Speaker 01:
So it's not just simply about improving it, but improving it in a very specific way.

[00:04:56] Speaker 04:
And then... So what's novel about allowing a different set of biases to be set?

[00:05:02] Speaker 01:
Actually, Your Honor, such a concept didn't exist.

[00:05:05] Speaker 01:
It was completely an artificial construct that was developed by Dr. Varma.

[00:05:09] Speaker 01:
And this court has, in 2016, found that it was a novel and non-obvious concept.

[00:05:15] Speaker 01:
and that was then specifically operationalized because this court found that the bias parameter is not about algebraic calculations of statistical measures or algebraic combinations.

[00:05:29] Speaker 01:
In fact, it is about the degree of randomness in sample selection that the user sets.

[00:05:35] Speaker 01:
So once the user sets that,

[00:05:37] Speaker 01:
then the user can see how these particular investments will perform.

[00:05:42] Speaker 01:
And even more importantly, the process requires that you maintain the temporal correlation between investments so that you can more accurately predict what is going to happen.

[00:05:55] Speaker 01:
So the claims recite these limitations very precisely.

[00:06:01] Speaker 01:
On top of that, the claims talk about a single statistical request

[00:06:08] Speaker 01:
that then corresponds to two or more investments.

[00:06:12] Speaker 01:
So all of these elements are designed to claim a very specific method with very specific limitations, which were completely ignored by the district court.

[00:06:29] Speaker 01:
And the district court simply oversimplified and overgeneralized the claims.

[00:06:34] Speaker 04:
So what does the bias parameter do other than allow for

[00:06:38] Speaker 04:
a manipulation of data?

[00:06:41] Speaker 01:
So the bias parameter is basically a construct, and it's designed to set the parameters for the analysis.

[00:06:53] Speaker 01:
It's very similar to, for example, the static price index, which was found to be an inventive concept in trading technologies.

[00:07:02] Speaker 01:
It is very similar to the coefficients relating to the temperature in exogens.

[00:07:07] Speaker 01:
It's not just a number.

[00:07:10] Speaker 01:
It is, in fact, a key way of capturing what the user wants.

[00:07:17] Speaker 00:
But both in Exogen and in trading technologies, there were assertively new physical things going on.

[00:07:26] Speaker 00:
Exogen, the new physical thing of a certain kind of swiping that allowed essentially

[00:07:36] Speaker 00:
a thermometer to get a good reading without even knowing where the artery was.

[00:07:42] Speaker 00:
And in trading technologies, it was the creation of a particular kind of visual display that arrayed two different kinds of information next to each other.

[00:07:57] Speaker 00:
But here, there is nothing physical that happens at the end of this process.

[00:08:01] Speaker 01:
Actually, Your Honor, resampling itself is a physical process.

[00:08:07] Speaker 01:
It actually requires you to basically fetch and load data and input them in registers.

[00:08:13] Speaker 01:
And then you basically create additional data based on the data that exists.

[00:08:21] Speaker 01:
So by itself, it's a physical process.

[00:08:24] Speaker 01:
In addition, just like trading technologies, we do have an output as well.

[00:08:28] Speaker 01:
We have an explicit output of distributions.

[00:08:32] Speaker 00:
So they may be... You have a display.

[00:08:35] Speaker 00:
Right.

[00:08:35] Speaker 00:
Training Technologies was in part about a specific way of displaying on a screen certain information.

[00:08:44] Speaker 00:
You have a plot the information.

[00:08:48] Speaker 01:
We understand that.

[00:08:49] Speaker 01:
And that was about an improved graphical user interface.

[00:08:53] Speaker 01:
And I agree with you, Your Honor.

[00:08:55] Speaker 01:
My only point here is that we also have physical processes going on here with resampling.

[00:09:01] Speaker 01:
And we also have a meaningful output for, say, maximum drawdown, gross rate of return, and so on.

[00:09:11] Speaker 01:
So how does the resampling occur?

[00:09:13] Speaker 01:
I'm sorry?

[00:09:14] Speaker 01:
How does the resampling occur?

[00:09:16] Speaker 01:
So the resampling occurs by once you have samples of the investments, and you've decided to sample them across multiple days, and you've determined

[00:09:30] Speaker 01:
how many of up downs and down days to sample based on the bias parameter, then you basically physically make multiple copies of that data, maintaining that temporal correlation between the stocks so that that way you know how the stocks relate to each other and also how they relate within each other.

[00:09:47] Speaker 01:
And that's your cross correlation and auto correlation.

[00:09:49] Speaker 01:
This is to make sure that you actually have

[00:09:52] Speaker 01:
captured how they relate to each other, because you might think you've got diversified investment, but you actually don't.

[00:09:57] Speaker 01:
And so you keep the temporal correlation, and then you make multiple- But resampling is not new.

[00:10:03] Speaker 04:
And in fact, resampling across a series of stock transactions is not new.

[00:10:08] Speaker 04:
What's new here is what your measure for resampling is, right?

[00:10:12] Speaker 01:
So what is new is two things.

[00:10:15] Speaker 01:
One is, you're absolutely correct, Your Honor.

[00:10:17] Speaker 01:
Resampling is not new, and Bradley Efron

[00:10:21] Speaker 01:
came up with resampling and bootstrapping and so on.

[00:10:24] Speaker 01:
His work is part of the record.

[00:10:26] Speaker 01:
But what we have done is not just resampling any data.

[00:10:31] Speaker 01:
We decide how to resample the biased samples.

[00:10:38] Speaker 01:
So we let the user pick what samples that you're going to do resampling on.

[00:10:44] Speaker 01:
And you're not just doing resampling generically.

[00:10:47] Speaker 01:
You're doing resampling in a particular way.

[00:10:51] Speaker 01:
keeping the temporal correlation, for example.

[00:10:54] Speaker 01:
And the prior art methods on the record have other ways of doing resampling.

[00:10:58] Speaker 01:
For example, the Sortino case, the Sortino reference, which was mentioned in the 2016 case before this court, that had a different method of doing resampling.

[00:11:09] Speaker 01:
So what we're talking about here is actually a new and improved method of analysis that uses features such as bias parameters.

[00:11:19] Speaker 01:
comes up with new and specific limitations on the resampling itself.

[00:11:25] Speaker 01:
And in addition, we try to trigger this with user input and provide the output in a form so that the user can perhaps redo the analysis again and change the conditions.

[00:11:41] Speaker 01:
that they want so that there is an accurate prediction of how that portfolio will perform under the kinds of conditions that the user is interested in modeling.

[00:11:53] Speaker 01:
So here, our key point is that there are critical claim limitations that should have been taken into account by the district court.

[00:12:04] Speaker 01:
And if they had taken it into account, they would realize that the claims were not directed to an abstract idea.

[00:12:09] Speaker 04:
You're not disputing that claim one is representative, are you?

[00:12:14] Speaker 01:
Well, it's a representative claim of the method claim.

[00:12:19] Speaker 01:
This claim 11 as well is a method claim and claim 22 is a system claim.

[00:12:26] Speaker 01:
So in addition, what the 291 patent does is it comes up with a computerized system where this method, if we have a lot of data, how we can efficiently

[00:12:40] Speaker 01:
Execute that.

[00:12:41] Speaker 04:
We had that issue in ALICE, where there was a system claim or an apparatus claim.

[00:12:47] Speaker 04:
And the Supreme Court said it didn't make any difference if all of it was doing was practicing the method.

[00:12:53] Speaker 01:
Yes, Your Honor.

[00:12:54] Speaker 01:
The difference is this.

[00:12:56] Speaker 01:
Here, we have very specific kinds of hardware that are explained and recited.

[00:13:03] Speaker 01:
We talk about not just any generalized resampling computerized system,

[00:13:09] Speaker 01:
which would be just like the Alice model.

[00:13:11] Speaker 01:
But here we're talking about having parallel processors where the data is vectorized to specifically handle that kind of data and to do it efficiently.

[00:13:22] Speaker 01:
For example, figure 12 in the packing talks about the specific vectorization.

[00:13:26] Speaker 01:
It talks about specific kinds of databases.

[00:13:29] Speaker 01:
And in addition, the system is designed to accomplish the resampling, keeping temporal correlation.

[00:13:37] Speaker 01:
So the

[00:13:39] Speaker 01:
system claims are much narrower.

[00:13:41] Speaker 01:
It is not just some generic system.

[00:13:45] Speaker 02:
Counsel, you're into, well into your rebuttal time.

[00:13:48] Speaker 02:
Would you like to save it or continue?

[00:13:50] Speaker 01:
I would like to save it, Your Honor.

[00:13:52] Speaker 02:
Thank you.

[00:13:57] Speaker 02:
Ms.

[00:13:58] Speaker 02:
Vidal?

[00:13:59] Speaker 04:
What is it, Your Honor?

[00:14:00] Speaker 04:
Would you mind starting where he left off, which is what is it about the system claims?

[00:14:05] Speaker 04:
I mean, do you

[00:14:05] Speaker 04:
Do you dispute his contention that the system claims are farmer-specific?

[00:14:11] Speaker 03:
The system claims do add generic computer components.

[00:14:14] Speaker 03:
If you look at all the components that are added, like parallel processors and some of the other components, they're all generic.

[00:14:20] Speaker 03:
There's nothing in the specification that says otherwise, and appellants have not otherwise said that these are specific components.

[00:14:29] Speaker 03:
There's something unique about them or something inventive about them.

[00:14:36] Speaker 03:
This case is not distinguishable from Fluke.

[00:14:39] Speaker 03:
The district court recognized that on page 16 of their opinion.

[00:14:42] Speaker 03:
In both cases, in Fluke and in this case, you have a mathematical algorithm, and you have a result.

[00:14:50] Speaker 03:
In Fluke, you had a mathematical algorithm related to setting an alarm limit, and the result was a varying alarm limit that was set.

[00:14:58] Speaker 03:
Here you have a mathematical algorithm, and the result is a distribution plot.

[00:15:04] Speaker 03:
So for that reason,

[00:15:05] Speaker 03:
This claim, these claims, should be held not patent eligible for the same reasons.

[00:15:11] Speaker 02:
They've been held to be non-obvious, right?

[00:15:16] Speaker 02:
They've been held to be non-obvious.

[00:15:18] Speaker 03:
That's correct.

[00:15:19] Speaker 03:
That's correct.

[00:15:19] Speaker 03:
The abstract idea may be non-obvious, but that does not affect the analysis when it comes to determining whether the abstract idea is indeed abstract.

[00:15:28] Speaker 03:
And because the abstract idea is math, according to Gottschalk, it is abstract.

[00:15:34] Speaker 04:
So in terms of what was held in our prior case as to why it was not obvious, are you saying that the only thing they said was not obvious was the choice of data input?

[00:15:48] Speaker 03:
It's a combination.

[00:15:49] Speaker 03:
They said what was not obvious was the choice of data input, so the bias parameter combined with the resampling method.

[00:15:56] Speaker 03:
That in and of itself is math.

[00:15:58] Speaker 03:
And if you look at the joint appendix pages 51 through 53, you'll see all the mathematical formulas.

[00:16:04] Speaker 03:
that control that math.

[00:16:06] Speaker 03:
That is the only thing that was alleged to be non-obvious, and that is not a proper determination when determining whether the math is an abstract idea.

[00:16:14] Speaker 04:
So your response to your friend on the other side is that his argument that his resampling method was unique and different doesn't matter?

[00:16:24] Speaker 03:
It doesn't matter, and the district court assumed that it would be.

[00:16:27] Speaker 03:
The district court said even if it is new and unique, that does not control.

[00:16:32] Speaker 03:
When determining whether

[00:16:33] Speaker 03:
Um, whether an idea is an abstract idea, you need to look beyond that.

[00:16:37] Speaker 03:
You need to look at whether it's mathematical as in, as in Gottschall that that determines that this is abstract.

[00:16:44] Speaker 02:
But if it's abstract, you can then look, go to step two to see if it's inventive.

[00:16:50] Speaker 02:
And isn't that almost a, that's the European word for non-obvious.

[00:16:55] Speaker 03:
So you cannot look at whether the abstract idea is inventive.

[00:16:58] Speaker 03:
You need to look at what was added to the abstract idea.

[00:17:01] Speaker 03:
Alice dictates that.

[00:17:03] Speaker 03:
In this case, that argument was never made.

[00:17:06] Speaker 03:
That said, the district court did go that step further.

[00:17:08] Speaker 03:
The district court applied Alice step one, determined that this was all mathematical manipulation and formulas and that that was abstract.

[00:17:16] Speaker 03:
And then the district court stepped through each and every dependent claim and each and every additional limitation and found that all of those additional limitations were either generic computer components, like processors, or they were insignificant pre and post activity.

[00:17:32] Speaker 04:
But why isn't the ordered combination of bias parameter plus resampling system, why isn't that enough to make it inventive under step 2?

[00:17:43] Speaker 03:
If you look at Fluke, Fluke talks about the circumference of a circle and says that the formula for circumference of a circle is abstract.

[00:17:52] Speaker 03:
It's 2 pi r. And basically, what you would be saying is because r is an input to that,

[00:17:58] Speaker 03:
If you choose r one way and then you put r into the formula, now somehow because you've combined a way of choosing r and a formula, that that is somehow inventive.

[00:18:08] Speaker 03:
And it's not.

[00:18:08] Speaker 03:
The bias parameter is just an input into the distribution, just as r, the radius of a circle, is an input into the equation for the circumference of a circle.

[00:18:24] Speaker 03:
I have nothing further to add.

[00:18:28] Speaker 00:
Would it be fair to say that the consequence of adopting the position that you're urging is that some fairly wide range of mathematical innovations used for finance would be categorically ineligible under 101?

[00:18:55] Speaker 03:
It would be fair to say

[00:18:57] Speaker 03:
I would say yes, in the sense that if you're using probability distributions to mimic behavior in the market or any other behavior, that is not patent eligible.

[00:19:09] Speaker 03:
I mean, if you look at the patent, it talks about the bell curve, which is used all over the place.

[00:19:15] Speaker 03:
It's used to estimate intelligence.

[00:19:16] Speaker 03:
It's used to estimate height.

[00:19:18] Speaker 03:
You can't patent that formula just the same way you can't patent the formulas in this patent.

[00:19:23] Speaker 03:
And yes, a lot of those can be used to model

[00:19:26] Speaker 03:
financial markets.

[00:19:27] Speaker 03:
You can use the Gaussian normal distribution to model financial markets.

[00:19:32] Speaker 03:
That doesn't make it patentable just because the financial market, the data happens to fall in a certain curve.

[00:19:39] Speaker 04:
What about the time length limitation of the claims?

[00:19:42] Speaker 03:
It's to get a formula.

[00:19:43] Speaker 03:
If you look at pages 51 to 53 of the patent, there's a specific formula for that.

[00:19:47] Speaker 03:
So all of those additional parameters just say how you shift and how you adjust that curve.

[00:19:53] Speaker 03:
It's all still a distribution.

[00:19:55] Speaker 03:
Is it a curve?

[00:19:56] Speaker 04:
every invention ultimately mathematical in the long run?

[00:19:59] Speaker 04:
I mean, is the overlay of math the end of the inquiry?

[00:20:05] Speaker 03:
No, the overlay of math would be one thing.

[00:20:07] Speaker 03:
This is only math.

[00:20:08] Speaker 03:
If there was math, and then there was something more, as in trading tech.

[00:20:11] Speaker 03:
In trading tech, there was math.

[00:20:13] Speaker 03:
But then there was a GUI.

[00:20:14] Speaker 03:
There was an interface that was new and inventive under Alice Step 2.

[00:20:18] Speaker 03:
Here, we only have the math.

[00:20:20] Speaker 03:
We only have the formulas on pages 51 and 53 of the joint appendix, on those pages of the patent.

[00:20:26] Speaker 03:
and nothing more.

[00:20:27] Speaker 03:
And yes, math may underlie a lot of what goes on in the world, but what's patentable has to be something more than just math.

[00:20:33] Speaker 03:
It's going to transcend financial institutions and other areas.

[00:20:41] Speaker 02:
Thank you, Council.

[00:20:42] Speaker 03:
Thank you.

[00:20:44] Speaker 02:
Mr. Key has a little rebuttal time, two and a half minutes.

[00:20:49] Speaker 01:
Thank you, Your Honors.

[00:20:56] Speaker 01:
Your honor, the Invespic patent claims are not like Fluke.

[00:21:01] Speaker 01:
In Fluke, we had a mathematical formula that was recited in the claims.

[00:21:07] Speaker 01:
There is no mathematical formula recited in Invespic.

[00:21:11] Speaker 01:
What the Supreme Court said was, if you assume that that mathematical equation is in the prior art, there's nothing else left in Fluke.

[00:21:20] Speaker 01:
That's what the Supreme Court said.

[00:21:22] Speaker 01:
That is not what is going on here.

[00:21:25] Speaker 01:
no mathematical expressions in the claims.

[00:21:30] Speaker 01:
What we have here is a particular method of trying to get around some prior limitations.

[00:21:40] Speaker 01:
And counsel is correct.

[00:21:41] Speaker 01:
The patent does talk about the Gaussian distribution.

[00:21:45] Speaker 01:
And what we have is a method to get past making that assumption.

[00:21:51] Speaker 01:
Don't make that assumption.

[00:21:52] Speaker 01:
Instead, look at how the data is and how the investments performed so that you take the distribution as they existed in the past.

[00:22:05] Speaker 01:
And that is basically the so-called frequentist tradition.

[00:22:09] Speaker 01:
And then you apply that, you apply the Bayesian probabilities from the bias parameter, and you combine both those approaches in a very specific method.

[00:22:17] Speaker 01:
So we're not talking about high-level combining of

[00:22:23] Speaker 01:
Frequentus and Bayesian, we're talking about doing so in a very specific way, taking the bias parameter into account, which is the Bayesian approach, taking the resampling, and that's the Frequentus approach, and we put them together in a very specific way.

[00:22:38] Speaker 01:
The equations correspond to how you do the autocorrelation.

[00:22:43] Speaker 01:
They correspond to how you determine the maximum drawdown.

[00:22:47] Speaker 01:
That's not in the claims.

[00:22:49] Speaker 01:
Those are just very specific

[00:22:52] Speaker 01:
enabling details.

[00:22:54] Speaker 01:
There is no math and there's no equations in the claim.

[00:22:57] Speaker 01:
What we have here is a very specific way of doing predictive analytics, and this is what data science is about.

[00:23:04] Speaker 01:
So if somebody else wants to model this in some other way, they want to model it using the bias parameter, that's fine.

[00:23:10] Speaker 01:
If they don't want to do it, they want to do it some other way, they're absolutely free to do that.

[00:23:15] Speaker 01:
They're, of course, free to use resampling.

[00:23:17] Speaker 01:
They're free to simply not even have any user input.

[00:23:21] Speaker 01:
and simply show the user a whole bunch of different conditions.

[00:23:24] Speaker 01:
What we're saying is there is a very specific way of doing this.

[00:23:29] Speaker 01:
Your Honor, the district court did not even consider, cite, or refer to this court's decision in 2016.

[00:23:36] Speaker 01:
So it simply did not engage with the claim limitations at any level.

[00:23:46] Speaker 01:
It simply said this is an abstract method of

[00:23:50] Speaker 01:
manipulating data.

[00:23:52] Speaker 01:
And, Your Honor, we did, in fact, argue in the district court.

[00:23:58] Speaker 02:
Counsel, as you can see, your red light is on.

[00:24:00] Speaker 02:
Your time has expired.

[00:24:02] Speaker 02:
So we will take the case under advisement.

[00:24:05] Speaker 01:
Thank you, Your Honors.