USPTO on Patent Eligibility -- Examples 38 & 39

By Michael Borella --

On January 7, 2019, the U.S. Patent and Trademark Office published updated examination guidance, instructing the examining corps and the PTAB how they should apply 35 U.S.C. § 101.  On the same day, the USPTO also published the latest in its series of examples of how this application of the § 101 inquiry should be carried out.  This latest set, encompassing Examples 37-42, apply the updated guidance (Examples 1-36 were previously published over the last four years).  Our analysis of Example 37 was recently published.

The USPTO emphasizes that these examples are "hypothetical and only intended to be illustrative of the claim analysis" under the updated guidance.  Furthermore, the examples "should be interpreted based on the fact patterns set forth below as other fact patterns may have different eligibility outcomes."  In other words, even if an applicant's claim recites similar language and functionality as that of one of the examples, that does not mean the applicant's claim is patent-eligible.  Context matters.

The updated guidance modified only part of the § 101 analysis (step 2A in the USPTO's parlance).  As set forth in Alice Corp. v. CLS Bank Int'l, this step involves determining whether a claim is directed to a judicial exception, such as an abstract idea.  If not, then no § 101 rejection can be made.

The updated guidance breaks step 2A into a pair of sub-steps:

• In sub-step 2A(i), one is to determine whether the claim recites a judicial exception, such as an abstract idea. Abstract ideas are now limited to three categories: mathematical concepts, certain methods of organizing human activity, and mental processes.  If there is no exception recited, the claim is eligible.

• If the claim recites such an exception, then in sub-step 2A(ii) one is to determine further "whether the recited judicial exception is integrated into a practical application of that exception." If so, the claim is eligible.

If the claim fails to establish its eligibility in step 2A, the second part of the § 101 analysis (step 2B) is applied to determine whether any element or combination of elements in the claim is sufficient to ensure that the claim amounts to significantly more than the judicial exception.  If this is the case, the claim is patent-eligible under § 101.  If not, it can be rejected.

Example 38

Example 38 relates to simulating an analog audio mixer.  The background provided by the USPTO is as follows (abbreviated to focus on key aspects):

[Many audiophiles prefer] listening to music in its analog form, as digital audio files are considered to "lose" much of the sound quality in the conversion from analog to digital.  Prior inventions attempted to create digital simulations of analog audio mixers to simulate the sounds from analog circuits.  However, the prior art audio mixer simulations do not produce the same sound quality as the actual analog circuits.

Applicant's invention . . . begins with a model of an analog circuit representing an audio mixing console.  The model includes a location of all the circuit elements within the circuit, an initial value for each of the circuit elements, and a manufacturing tolerance range for each of the circuit elements.  A randomized working value of each element is then determined using a normally distributed pseudo random number generator (PRNG) based on the initial value of the circuit element and the manufacturing tolerance range.  The model is then simulated using a bilinear transformation to create a digital representation of the analog circuit.

The claim of Example 38 recites:

A method for providing a digital computer simulation of an analog audio mixer comprising:
    initializing a model of an analog circuit in the digital computer, said model including a location, initial value, and a manufacturing tolerance range for each of the circuit elements within the analog circuit;
    generating a normally distributed first random value for each circuit element, using a pseudo random number generator, based on a respective initial value and manufacturing tolerance range; and
    simulating a first digital representation of the analog circuit based on the first random value and the location of each circuit element within the analog circuit.

Applying the first sub-step of 2A, the USPTO states that this claim does not recite any of the three types of abstract ideas.  Notably, the USPTO concludes that despite explicitly reciting "generating a normally distributed first random value," the claim does not recite a mathematical calculation.  The USPTO's reasoning to support this rather surprising outcome is that "[w]hile some of the limitations may be based on mathematical concepts, the mathematical concepts are not recited in the claims."  The USPTO quickly concludes that the claim also does not recite a mental process ("because the steps are not practically performed in the human mind") or a method of organizing human activity.  Therefore, the claim is eligible and the analysis ends without moving on to the second sub-step.

Example 39

Example 39 relates to training a neural network for facial detection.  The background provided by the USPTO is as follows (again abbreviated to focus on key aspects):

In facial detection, a neural network classifies images as either containing a human face or not, based upon the model being previously trained on a set of facial and non-facial images.  However, these prior methods suffer from the inability to robustly detect human faces in images where there are shifts, distortions, and variations in scale and rotation of the face pattern in the image.

Applicant's invention addresses this issue by using a combination of features to more robustly detect human faces.  The first feature is the use of an expanded training set of facial images . . . developed by applying mathematical transformation functions on an acquired set of facial images.  Unfortunately, the introduction of an expanded training set increases false positives when classifying non-facial images.  Accordingly, the second feature of applicant's invention is the minimization of these false positives by performing an iterative training algorithm, in which the system is retrained with an updated training set containing the false positives produced after face detection has been performed on a set of non-facial images.

The claim of Example 39 recites:

A computer-implemented method of training a neural network for facial detection comprising:
    collecting a set of digital facial images from a database;
    applying one or more transformations to each digital facial image including mirroring, rotating, smoothing, or contrast reduction to create a modified set of digital facial images;
    creating a first training set comprising the collected set of digital facial images, the modified set of digital facial images, and a set of digital non-facial images;
    training the neural network in a first stage using the first training set;
    creating a second training set for a second stage of training comprising the first training set and digital non-facial images that are incorrectly detected as facial images after the first stage of training; and
    training the neural network in a second stage using the second training set.

Applying the first sub-step of 2A, the USPTO states that this claim does not recite any of the three types of abstract ideas.  It is reasonably well known that neural networks can be considered mathematical structures (albeit very complex ones), and are often modelled using linear algebra and calculus.  But, not unlike the analysis of Example 38's claim, the USPTO finds this claim to not recite any mathematical concepts.  Also like that of Example 38's claim, the USPTO further concludes that the claim does not recite a mental process or a method of organizing human activity.  Even the admittedly mathematical image transformations escape unscathed.  Therefore, the claim is eligible and the analysis ends at the first sub-step.

Analysis

If the goal of Example 38 is to draw a clearer line around the nebulous notion of abstract ideas, it has failed spectacularly.  Randomly generating a variable according to proscribed distribution is almost always a mathematical calculation.  And if the USPTO is giving patentable weight to the notion of pseudorandom generation being the differentiator, that would be a mistake.  Under a broadest reasonable interpretation, pseudorandom number generators include linear congruential generators, which are simple recursive mathematical functions.

Of course, even if one were to conclude that the claim of Example 38 does recite the abstract idea of a mathematical calculation, it could be argued that the additional elements of the claim integrate this abstract idea into a practical application thereof -- notably, a circuit simulation that can be used to general audible analog impurities.  Still, it remains difficult to rectify the USPTO's reasoning with the scientific and technical reality underlying its example.

Example 39 on the other hand, provides some relief to patentees who are attempting to protect advances in machine learning.  This example does not describe the structure of the neural network being trained, and essentially treats it like a black box.  But the claim is eligible because of the two-phase training technique in combination with the transforming of the facial images.  In this example, the USPTO is signaling that it will not treat a reasonably-claimed machine learning invention as "just a bunch of math."

Comments

[MathStudent]

So, of course, there are, in more or less clear order and not exhaustively: the Constitution, statutes, the CFR, SCOTUS (in their role interpreting statutes and regulation), the CAFC, PTO Guidance and then... PTO examples. While the take-home of the examples seems clear enough, i.e. that new step 2A,1st identification of an abstract idea in the form of a mathematical calculation should turn on recitation of explicit and possibly basic mathematical operations, what is not clear is what case law supports a take-home bifurcating at this particularly engineered line in claim interpretation. If put into practice, this would have the ironic effect of rendering a BROADER claim patent eligible, as in in these examples, but causing a hypothetically NARROWER dependent claim more explicitly reciting the math to be directed to an abstract idea, at least in step 2A,1st. And what if the broader claim, looking to the specification, only is supported for embodiments such as in the hypothetical dependent claim, then would the broader claim still be interpreted according to the take-home here, based only on its plain language? The PTO Guidance and, presumably, the examples are subject to a period of review, so there's time for revision. Meanwhile, conversations between examiners and applicant representatives citing these examples will be interesting, especially given the above hierarchy.

[Skeptical]

MathStudent,

As I have noted already (but it bears emphasis), this new guidance is a FAIL for both it's lack of reach and it's over-reach.

The "over-reach" is because while explicitly calling out the Supreme Court-induced Common Law law (re)writing of 35 USC 101 as having created a Gordian Knot of contradictions, the director attempts to incorporate those same contradictions (albeit, he attempts to do in "layers," with what appears to be "escape hatches" at some of the higher layers.

The lack of reach is that THIS is merely Office guidance (which of course is a necessary "lack," given our government structure of three branches and separation of powers). It has already been pointed out that MANY existing cases, following the layered (provided escape hatches) approaches would come out with OPPOSITE conclusions that SOME panels of the CAFC (as well as other courts of the Article III nature) would come to.

So,...

EVEN IF one is "fortunate" to be able to employ successfully the new "escape hatches" in some of the new layers, one STILL has no ability to predict what possible future "luck of the draw" panel one may obtain at a CAFC level, nor with ANY ability to generate any reliance that a patent that passes the executive branch will -- or will not -- stand in view of what the judicial branch "knows when they see it."

Of course, this is a direct result that the Supreme Court's broken score board remains in place.

Let me ALSO note the recent advancement by Justice Kavanaugh in the oral arguments of the California Franchise Tax Board v. Hyatt:

“Kavanaugh suggested, overruling is proper when a precedent is “egregiously wrong,” has “severe practical consequences,” and generates “no reliance.”

There IS a growing realization that what the Court has done with its fingers in the wax nose of patent law eligibility satisfies each and every one of these three factors.

[21f2f3fff]

The problem with ex. 38 is that a model IS a mathematical equation. That's what a model is. F=ma is a model for how an object having a given mass responds to a force. The 'elements' of the model are just variables of the equation. So ex. 38 is essentially claiming an equation having certain variables, randomly generating a value for one of the variables, and solving the equation.That substituting the word "model" for "equation" apparently makes the claim "not recite a mathematical concept" is stupidity of the highest order.

[Martin Snyder]

Once again, my test is easier and way more direct as to eligibility on a policy basis.

The invention is a method. The result of the method is information. The information is consumed by an amplifier circuit, a non-human actor. It should be eligible.

As to abstraction in the claims; PHOSITA could make and use the invention from the description, and the claims are not reciting an aspiration.

I don't know enough about the art to make a 103 determination, but that seems to me where the claims would fail, if at all.

If the method were producing information consumed by persons, it would not be eligible. This method is not infringed when a person hears the amplified material- but when the transformation of the signals occurs.

Also, PS, gift-certificates are accounting entries, and accounting information is consumed by persons to make decisions about goods, services, and financial assets. Not eligible.

[Skeptical]

Mr. Snyder,

Your "PS" is most odd, as you seem to want to draw attention to the fact that in a recent case ( https://www.patentdocs.org/2019/01/coqui-technologies-llc-v-gyft-inc-d-del-2018.html#comments ), you have chosen to contradict your own (rather odd) pet theory.

To wit, the claim there is explicitly "directed to" machine consumption: "...request by a user's communication terminal through a wired network, wireless network, or both, comprising:"

So let me repeat a comment from there (to which you have not deigned to reply) here:

Mr. Snyder apparently has abdicated his stance as to eligibility depending on whether consumption of information is by machine (not possible to be abstract) or by a human.

Here, even machine consumption appears to be "fundamentally abstract" because some later utility is utility by humans.

I wonder how a consistent application of a "any later utility by humans" would play out in ALL of patent law....

Would we even HAVE patent law?

Being entirely objective and consistent with the position advocated by Mr. Snyder, everyone should be: