Patent Application 17868267 - SYSTEM AND METHOD FOR TEST-TIME ADAPTATION VIA

Title: SYSTEM AND METHOD FOR TEST-TIME ADAPTATION VIA CONJUGATE PSEUDOLABELS
Application Information

Invention Title: SYSTEM AND METHOD FOR TEST-TIME ADAPTATION VIA CONJUGATE PSEUDOLABELS
Application Number: 17868267
Submission Date: 2025-05-15T00:00:00.000Z
Effective Filing Date: 2022-07-19T00:00:00.000Z
Filing Date: 2022-07-19T00:00:00.000Z
National Class: 706
National Sub-Class: 012000
Examiner Employee Number: 99243
Art Unit: 2126
Tech Center: 2100
Rejection Summary

102 Rejections: 1
103 Rejections: 5
Cited Patents

No patents were cited in this rejection.
Office Action Text


    DETAILED ACTION
Notice of Pre-AIA  or AIA  Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
In the event the determination of the status of the application as subject to AIA  35 U.S.C. 102 and 103 (or as subject to pre-AIA  35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA  to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.  
Response to Amendment
Applicant’s submission filed on April 28, 2025, has been entered. Claims 1-20 are pending.
Response to Amendment
Applicant’s arguments filed on 04/28/2025 have been fully considered and not found persuasive.
In response to Applicant’s amendment, Examiner withdraws the previously raised objection to the Specification.
Response to 101 Rejection
	Starting on page 1 and ending on page 1, Applicant remarks:
“While the Examiner characterictizes several limitations (e.g., generating prediction data, pseudo-reference data and loss data) as mental processes or mathematical concepts, this characterization oversimplifies the technical nature of invention. The claims are not directed to abstract ideas performed in the human mind, by rather to a specific method of updating parameter data of a machine learning system via gradient-based pseudo-reference data derived from sensor data in the second (or target) domain… [and controlling an actuator…]—a process that cannot practically be performed as a mental process in the human mind.”
Examiner’s Response:
	Applicant’s argument is unclear. For the purposes of compact prosecution, Examiner shall assume that Applicant is arguing that the controlling of an actuator cannot be practically be performed as a mental process and therefore the entire claims do not recite an abstract idea.
	Examiner did not state that the controlling of an actuator is a mental process. Instead, for example, Examiner identified limitations the following limitations as a mental process: 
“generating, via the machine learning system, prediction data based on the sensor data;” – this limitation is recited at a high level of generality and is directed to the abstract idea of a mental process, i.e., generating prediction data based on sensor data (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.] “via the machine learning system” will be interpreted under Prong Two.
“generating pseudo-reference data based on a gradient of a predetermined function evaluated with the prediction data;”  – this limitation is recited at a high level of generality, and when interpreted in view of the specification, is directed to the abstract idea of a mental process, i.e., as pseudo-reference data refers to creating labels for unlabeled data. (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.]). Additionally, this limitation recites the abstract idea of a mathematical concept (see MPEP 2106.04(a)(2)), i.e., as a gradient, refers to calculating the mathematical operation.
“generating loss data based on the pseudo-reference data and the prediction data;” – this limitation is recited at a high level of generality is directed to the abstract idea of a mathematical concept (see MPEP 2106.04(a)(2)), i.e., generating loss data, when read in light of the specification, refers to calculating the loss function.
The use of an actuator is not recited until after the limitations identified as abstract ideas. Notably, Examiner did not solely identify limitations that were directed to abstract idea of a mental process but also limitations directed to the abstract idea of mathematical concepts. Therefore, the independent claims positively recite abstract ideas.
Starting on page 2 and ending on page 2, Applicant remarks:
“Even if a judicial exception were implicated, the claims are integrated into a practical application. For example, the claims include:- obtaining sensor data from the second domain,- generating, via the machine learning system, prediction data based on that sensor data,- generating loss data based on a gradient of a predetermined function evaluated with the prediction data,- updating parameter data of the machine learning system based on the loss data, and- performing, via the machine learning system, a task in the second domain after the parameter data has been updated....
This is not a generic recitation of applying an abstract idea to a field of use.”
Examiner’s Response:
	As a threshold matter, Examiner maintains that the claims are recited in a generic matter. The claims are recited using broad language that lack specifics. For example, the claims recite a “predetermined function” but do not recite examples of possible predetermined functions. The claims recite “updating parameter data” but not include examples of parameter data. The claims recite “via the machine learning system” but do not recite examples of machine learning systems.  Additionally, when the claims recite the act of generating data, they are recited at a broad level such as “generating loss data based on the pseudo-reference data and the prediction data.” Similarly, the use of an actuator, is recited at a high level of generality, and as explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible “simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use.” MPEP 2106.05(h), like claiming the mental processes/mathematical concepts are being used with an actuator for an (unspecified) particular technological use. Therefore, for these reasons and the reasons under the 35 U.S.C. 101 rejection header, Examiner maintains the 35 U.S.C. 101 rejection.
Applicant Remarks:
The USPTO’s 2019 PEG Example 39 (“Robust Training of Neural Networks”) supports eligibility for claims directed to improving training techniques that enhance model robustness. In Example 39, the claims recite training a neural network using an expanded training set of facial images to train the neural network. This expanded training set is developed by applying mathematical transformation functions on an acquired set of facial images. The training method was found to be patent eligible because it improves the robustness of a neural network in a specific, practical context — image recognition. The claimed method passes Step 2A because it does not recite any of the judicial exceptions, and in particular, the claim does not recite a mental process because the steps are not practically performed in the human mind. Similarly, Applicant’s claimed concepts outlined above cannot be practically performed in the human mind.
Examiner’s Response:
Example 39 is facially distinct from the presented claims. Claim 1, involves steps such as “generating pseudo-reference data based on a gradient of a predetermined function evaluated with prediction data” and “generating loss data based on the pseudo-reference data and the prediction data” and can be practically performed in the human  mind. In contrast, Example 39 recites “applying one or more transformations to each digital face image including mirroring, rotating, smoothing, or contrast reduction to create a modified set of digital facial images” and “training the neural network in a first stage using the first training set,” are considered not practically performable in the human mind. The present claims do not recite a limitation or step that is analogous to applying transformations on a digital images.
As described above, unlike how Example 39 provides examples of a transformation, claim 1 recites terms like “a predetermined function,” “prediction data,” “pseudo-reference data” and “loss data,” without further specificity, could encompass a wide range of mathematical or algorithmic processes within the broadest reasonable interpretation. Example 39 provides more concrete examples such as the specific types of data, digital facial images and non-facial images and transformations. Additionally, while a human could theoretically transform an image, the act of applying “one or more transformations” to “each digital facial image’ positively recites a scale of manipulation that exceeds a mental process. Lastly, as discussed above, unlike Example 39, the independent claims still recite the abstract idea of mathematical processes in addition to the abstract idea of mental processes. Therefore, the independent claims of the present application are unlike Example 39, recite mental processes and mathematical concepts and for these reasons, and the reasons under the 35 U.S.C. 101 rejection header, Examiner is maintaining the 35 U.S.C. 101 rejection.
Applicant Remarks:
	The Examiner asserts that the use of machine learning is well-understood and routine. However, the claimed recitation of generating, for example, pseudo-reference data based at least on a gradient of a predetermined function is not a conventional application of machine learning. These claimed features enable test-time adaptation, for example, without requiring human-provided labels of sensor data in the target domain. The Examiner has provided no evidence that these claimed features is well-understood or routine. In actuality, quite the opposite is true.
	Accordingly, the claims are not directed to an abstract idea, or in the alternative, are significantly more than any judicial exception, and therefore satisfy 35 U.S.C. § 101.
Examiner’s Response:
	Examiner did not state that the claim, as a whole, is well-understood and routine. However, specific limitations in the claims are well-understood and routine (WURC). For example, “obtaining the sensor data from the second domain;” was identified a well-understood, routine and conventional. Evidence was provided in the form of citation, such as “Further, the insignificant extra-solution activity of data gathering is also WURC, see MPEP 2106.05(d)(II) “The courts have recognized the following computer functions as well-understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. i. Receiving or transmitting data over a network”. Additionally, examiner cited to MPEP 2106.05(d) II (ii) of performing repetitive calculations. ) “The courts have recognized the following computer functions as well-understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. iii. Performing repetitive calculations, (citing to Flook).” The additional elements, as identified in the non-final rejection, are mere instructions to apply or extra-solution activities that are WURC.
	Therefore, the claims recite abstract ideas including mental process and mathematical operations, the abstract ideas are not integrated in to a practical application nor do significantly more. For these reasons, and the reasons under the 35 U.S.C. 101 rejection header, Examiner is maintaining the rejection.
Response to 102/103 Rejection
On page 5, Applicant remarks:
Applicant respectfully traverses this rejection because Xu fails to disclose at least “generating, via the processor, pseudo-reference data based on a gradient of a predetermined function evaluated with the prediction data,” as recited in independent claim 15.
In FIG. 2 and FIG. 3, Xu discloses generating pseudo-labels from (i) an input image 202/302 and (ii) weak annotations 238/316 via a weak supervision technique. See Xu at paragraphs [0069]-[0070] and [0077]. However, Xu does not disclose generating pseudo- reference data based on a gradient of a predetermined function evaluated with the prediction data, as recited in independent claim 15
Also, with respect to paragraph [0072], which is relied upon in the rejection, Xu merely refers to gradient descent in the context of backpropagation to locally update a current model 218. However, Xu does not generate pseudo-reference data by applying a gradient to the loss function, as set forth in the claims. This is a fundamental difference. The claimed invention generates pseudo-reference data via a gradient of a predetermined function (e.g., a loss function) applied to the prediction data. Xu does not disclose at least this feature, as recited. Claim 15 is therefore patentable for at least these reasons.

Examiner’s Response:

The claimed language states “generating pseudo-reference data based on a gradient of a predetermined function evaluated with the prediction data.” None of these terms are explicitly defined by the specification. Therefore, the claims are interpreted under their broadest reasonable interpretation using plain meaning in light of the specification as it would be understood by one of ordinary skill in the art. Xu at 0072 recites:

    PNG
    media_image1.png
    239
    471
    media_image1.png
    Greyscale
) 
	Both Xu and the present application generates pseudo-labels. The pseudo-labels are updated, i.e., pseudo-reference data, based on a gradient of a context loss, i.e., a gradient of the predetermined function. The context loss, i.e., predetermined function, is evaluated with the prediction maps 220, i.e., prediction data, and initial pseudo labels 210. Xu uses these locally updated models to determine updated feature maps that undergo meta label fusion to generate fused labels that update the main model. Therefore, the fused labels represented the updated pseudo-labels. Therefore, Xu at 0072 recites the claimed limitation within the broadest reasonable interpretation. Any additional features or factors identified by the Applicant are not claimed.
Applicant remarks:

Regarding claim 18, the Examiner points to Xu’s use of scaling during gradient-based updates (e.g., using weight y=1), but this scaling occurs during standard backpropagation, not during generation of pseudo-reference data. The claim requires that pseudo-reference data be generated from a gradient, then used to compute a Joss, and that a scaled gradient of that loss is used to update the model. Xu does not disclose this sequence, nor does it treat gradients as pseudo-reference data sources. Claim 18 is therefore patentable for at least these additional reasons. Accordingly, Applicant respectfully requests withdrawal of the 35 U.S.C. § 102 rejection of each of the claims.

Examiner’s response:

Claim 18 recites “wherein the parameter data is updated using a scaled gradient of the loss data.” Claim 18 is dependent on claim 15 where claim 15 recites “updating parameter data of the machine learning system based on the loss data;”1 The terms “parameter data” and “loss data” are not defined insofar as, explained in claim 15, “loss data” is “based on the pseudo-reference data and the prediction data” and that “updating parameter data” is “based on the loss data.” 
Xu explicitly recites “(0072) For each pseudolabel 210 YjAi, j∈[1,n] for each weak annotation 238 type Ai, in an embodiment, a training framework computes a local loss function as described above, and performs a gradient descent to backpropagate 240 said local loss to locally update a current model 218 M into individual locally updated models Mj.” The term “gradient descent” is a well-established optimization algorithm universally used in training machine learning models and particularly neural networks. Gradient descent algorithms use a learning rate to control the size of the gradient descent step, i.e., how much the parameters are adjusted during each iteration. See Deniz, Gamez, (Jan. 9 2022) “Understanding Gradient Descent and Learning Rate,” Medium (“We can set the size of the gradient descent step by multiplying the gradient with a constant, namely a learning rate: 
    PNG
    media_image2.png
    222
    326
    media_image2.png
    Greyscale
”) And the learning rate is a scalar on the gradient that controls the level of magnitude of change in each iteration. See Kwiatkowski, Robert (May 22, 2021) “Gradient Descent Algorithm – a deep dive,” Medium (“Gradient Descent Algorithm iteratively calculates the next point using gradient at the current position, scales it (by a learning rate) and subtracts obtained value from the current position (makes a step).” (emphasis added in italics). Therefore, gradient descent algorithms, which Xu recites using, inherently use a learning rate to scale the gradient descent and Xu recites, within the broadest reasonable interpretation, “the parameter data is updated using a scaled gradient of the loss data.”
Similarly, 0072 additionally recites “During training by a training framework, in an embodiment, a context loss is combined with common segmentation losses to train 240 model 218 M with weight as £ = £Dice + y£context where £Dice is Dice loss and y=1.” y equaling one is an exemplar, but regardless, y is a constant that scales the £Dice, i.e., directly scale the Dice loss that contributes to the total gradient calculated during gradient descent. Therefore, y is a scalar that modifies the Dice loss, i.e., a scaled gradient of the loss data. This gradient is used to update the parameters of the local model which are used to update the parameters of the convolutional neural network (0073-0075). Therefore, Xu recites, within the broadest reasonable interpretation, “the parameter data is updated using a scaled gradient of the loss data.”
For these reasons, and the reasons under the 35 U.S.C. 102/103 rejection header, Examiner is maintaining the 102/103 rejection.
Applicant Remarks:

The claim language from claim 15 discussed above is also present in the other independent claims. None of these secondary references appear to cure the deficiencies of Xu. Therefore, the rejection of these claims should also be withdrawn for similar reasons as discussed above with respect to at least independent claim 15.

Examiner’s Response:

As discussed above, Xu recites the claim language from claim 15 that is present in the other independent claims. Therefore, Examiner is maintaining 102/103 rejection.
Rejection 35 U.S.C. § 101
U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. 
Step 1 of the subject matter eligibility test (see MPEP 2106.03).
Claims 1-7 are directed to a “method” which describes one of the four statutory categories of patentable subject matter, i.e., a machine.
Claims 8-14 are directed to a “method” which describes one of the four statutory categories of patentable subject matter, i.e., a machine.
Claims 15-20 are directed to a “system” which describes one of the four statutory categories of patentable subject matter, i.e., a machine.
Regarding Claim 1:
Step 2A of the subject matter eligibility test (see MPEP 2106.04).  
Prong One: 
Claim 1 recites (“sets forth” or “describes”) the abstract ideas as follows: 
“generating, via the machine learning system, prediction data based on the sensor data;” – this limitation is recited at a high level of generality and is directed to the abstract idea of a mental process, i.e., generating prediction data based on sensor data (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.] “via the machine learning system” will be interpreted under Prong Two.
“generating pseudo-reference data based on a gradient of a predetermined function evaluated with the prediction data;”  – this limitation is recited at a high level of generality, and when interpreted in view of the specification, is directed to the abstract idea of a mental process, i.e., as pseudo-reference data refers to creating labels for unlabeled data. (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.]). Additionally, this limitation recites the abstract idea of a mathematical concept (see MPEP 2106.04(a)(2)), i.e., as a gradient, refers to calculating the mathematical operation.
“generating loss data based on the pseudo-reference data and the prediction data;” – this limitation is recited at a high level of generality is directed to the abstract idea of a mathematical concept (see MPEP 2106.04(a)(2)), i.e., generating loss data, when read in light of the specification, refers to calculating the loss function.
“performing, via the machine learning system, a task in in  the second domain after the parameter data has been updated; and” – this limitation is recited at a high level of generality and is directed to the abstract idea of a mental process, i.e., performing a task in the second domain (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.] “via the machine learning system” will be interpreted under Prong Two.
Prong Two: Claim 1 does not include additional elements that integrate the mental process into a practical application. 
“obtaining the sensor data from the second domain;” – is recited at a high level of generality and is merely a recitation of an insignificant extra-solution activity of data gathering (see 2106.05(g)).
“via the machine learning system” – is recited at a high level of generality and is merely an instruction to apply an abstract idea (see MPEP 2106.05(f)), i.e., generating prediction data, in a generic computer environment using generic computer functions.
“updating parameter data of the machine learning system based on the loss data;” – is recited at a high level of generality and is merely a recitation of an insignificant extra solution activity of data manipulation (see 2106.05(g)).
“controlling an actuator based on the task performed in the second domain.” – is recited at a high level of generality and is viewed as field of use (see MPEP 2106.05(h)) as it is merely limiting the application to an actuator such as, when viewed in light of the specification, a partially autonomous vehicle or partially autonomous robot.2
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application.
Step 2B of the subject matter eligibility test (see MPEP 2106.05).
When considered individually or in combination, the additional limitations and elements of claim 1 does not amount to significantly more than the judicial exception for the reasons as discussed above as to why the additional limitations to not integrate the abstract idea into a practical application.
“obtaining the sensor data from the second domain;” – is merely a recitation of an insignificant extra-solution activity of data gathering (see 2106.05(g)), which does not integrate a judicial exception into practical application. Further, the insignificant extra-solution activity of data gathering is also WURC, see MPEP 2106.05(d)(II) “The courts have recognized the following computer functions as well-understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. i. Receiving or transmitting data over a network”.
“via the machine learning system” – is recited at a high level of generality and is merely an instruction to apply an abstract idea, i.e., generating prediction data, and therefore fails to amount to significantly more than the judicial exception. See MPEP 2106.05(f).
“updating parameter data of the machine learning system based on the loss data;” – is merely a recitation of an insignificant extra solution activity of data manipulation (see MPEP 2106.05(g)) “Selecting a particular data source or type of data to be manipulated”; which does not integrate a judicial exception in to a practical application. Further, as updating parameter data based on loss requires repetitive calculations for minimizing the loss, this limitation further recites a well-understood, routine and conventional activity under MPEP 2106.05(d) II (ii) of performing repetitive calculations.
“controlling an actuator based on the task performed in the second domain.” – is a recitation of field of use, see MPEP 2106 (citing Intellectual Ventures I LLC v. Capital One Bank (USA), N.A., 792 F.3d 1363, 1366, 115 USPQ2d 1636, 1639 (Fed. Cir. 2015) (“An abstract idea does not become nonabstract by limiting the invention to a particular field of use or technological environment, such as the Internet [or] a computer”) and therefore fails to amount to significantly more than the judicial exception.
Claim 2 is clarifying the machine learning system as performing the task of a classifier, therefore the claim is directed towards the abstract idea of a mental process, i.e., classifying (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.]) and therefore fails to amount to significantly more than the judicial exception.
	Claim 3 is clarifying the machine learning system, i.e., the machine learning system uses the same predetermined function (which is recited at a high level of generality), and is therefore directed towards clarifying the instructions to apply the abstract idea and therefore fails to amount to significantly more than the judicial exception.
	Claim 4 is clarifying the abstract idea, i.e., the predetermined function as a loss function, used to perform the mental evaluations and is therefore directed towards the abstract idea of the mathematical operations, as a loss function is a mathematical operation, and therefore fails to amount to significantly more than the judicial exception.
	Claim 5 is clarifying the abstract idea, i.e., the predetermined function comprises one of several loss functions, is therefore directed towards clarifying abstract idea of the mathematical operations, as specifying the loss function comprises one of several loss functions, and therefore fails to amount to significantly more than the judicial exception.
	Claim 6 is clarifying the abstract idea, i.e., updating the parameter data using a scaled gradient of the loss data, and is therefore directed towards clarifying the abstract idea of mathematical operations, as using a scaled gradient is a mathematical operation, and therefore fails to amount to significantly more than the judicial exception.
	Claim 7 is merely a recitation of an insignificant extra-solution activity of data gathering (see 2106.05(g)), which does not integrate a judicial exception into practical application. Further, the insignificant extra-solution activity of data gathering is also WURC, see MPEP 2106.05(d)(II) “The courts have recognized the following computer functions as well-understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. i. Receiving or transmitting data over a network”. Additionally, this limitation recites “one or more sensor” which is interpreted as generic machinery or computer component being used in its ordinary capacity to measure data, therefore, this limitation invokes a machinery merely as a tool to perform an existing process (see MPEP 2106.05(f)(2)).
Regarding Claim 8:
Step 2A of the subject matter eligibility test (see MPEP 2106.04).  
Prong One: 
Claim 8 recites (“sets forth” or “describes”) the abstract ideas as follows: 
“generating, via the machine learning system, prediction data based on the sensor data;” – this limitation is recited at a high level of generality and is directed to the abstract idea of a mental process, i.e., generating prediction data based on sensor data (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.] “via the machine learning system” will be interpreted under Prong Two.
“generating loss data based on a negative convex conjugate of a predetermined function applied to a gradient of the predetermined function, the predetermined function being evaluated based on the prediction data;” - this limitation is recited at a high level of generality is directed to the abstract idea of a mathematical concepts (see MPEP 2106.04(a)(2)), i.e., negative convex conjugate applied to a gradient of the predetermined function. 
“performing, via the machine learning system, a task in the target domain after the parameter data has been updated; and” – this limitation is recited at a high level of generality and is directed to the abstract idea of a mental process, i.e., performing a task in the second domain (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.] “via the machine learning system” will be interpreted under Prong Two.
Prong Two: Claim 8 does not include additional elements that integrate the mental process into a practical application. 
“obtaining the sensor data from the target domain;” – is merely a recitation of an insignificant extra-solution activity of data gathering (see 2106.05(g)).
“via the machine learning system” – is recited at a high level of generality and is merely an instruction to apply an abstract idea (see MPEP 2106.05(f)), i.e., generating prediction data, in a generic computer environment using generic computer functions.
“updating parameter data of the machine learning system based on the loss data;” – is recited at a high level of generality and is merely a recitation of an insignificant extra solution activity of data manipulation (see 2106.05(g)).
“controlling an actuator based on the task performed in the target domain.” – is recited at a high level of generality and is viewed as field of use (see MPEP 2106.05(h)) as it is merely limiting the application to an actuator such as, when viewed in light of the specification, a partially autonomous vehicle or partially autonomous robot.3
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application.
Step 2B of the subject matter eligibility test (see MPEP 2106.05).
When considered individually or in combination, the additional limitations and elements of claim 8 does not amount to significantly more than the judicial exception for the reasons as discussed above as to why the additional limitations to not integrate the abstract idea into a practical application.
“obtaining the sensor data from the target domain;” – is merely a recitation of an insignificant extra-solution activity of data gathering (see 2106.05(g)), which does not integrate a judicial exception into practical application. Further, the insignificant extra-solution activity of data gathering is also WURC, see MPEP 2106.05(d)(II) “The courts have recognized the following computer functions as well-understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. i. Receiving or transmitting data over a network”.
“via the machine learning system” – is recited at a high level of generality and is merely an instruction to apply an abstract idea, i.e., generating prediction data, and therefore fails to amount to significantly more than the judicial exception. See MPEP 2106.05(f).
“updating parameter data of the machine learning system based on the loss data;” – is merely a recitation of an insignificant extra solution activity of data manipulation (see MPEP 2106.05(g)) “Selecting a particular data source or type of data to be manipulated”; which does not integrate a judicial exception in to a practical application. Further, as updating parameter data based on loss requires repetitive calculations for minimizing the loss, this limitation further recites a well-understood, routine and conventional activity under MPEP 2106.05(d) II (ii) of performing repetitive calculations.
“controlling an actuator based on the task performed in the second domain.” – “controlling an actuator based on the task performed in the second domain.” – is a recitation of field of use, see MPEP 2106 (citing Intellectual Ventures I LLC v. Capital One Bank (USA), N.A., 792 F.3d 1363, 1366, 115 USPQ2d 1636, 1639 (Fed. Cir. 2015) (“An abstract idea does not become nonabstract by limiting the invention to a particular field of use or technological environment, such as the Internet [or] a computer”) and therefore fails to amount to significantly more than the judicial exception.
	Claim 9 is clarifying the machine learning system as performing the task of a classifier, therefore the claim is directed towards the abstract idea of a mental process, i.e., classifying (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.]) and therefore fails to amount to significantly more than the judicial exception.
	Claim 10 is clarifying the machine learning system, i.e., the machine learning system uses the same predetermined function (which is recited at a high level of generality), and is therefore directed towards clarifying the instructions to apply the abstract idea and therefore fails to amount to significantly more than the judicial exception.
	Claim 11 is clarifying the abstract idea, i.e., the predetermined function as a loss function, used to perform the mathematical operations and is therefore directed towards clarifying the abstract idea of the mathematical operations and therefore fails to amount to significantly more than the judicial exception.
	Claim 12 is clarifying the abstract idea, i.e., the predetermined function comprises one of several loss functions, used to perform the mathematical operations and is therefore directed towards clarifying the abstract idea of the mathematical operations and therefore fails to amount to significantly more than the judicial exception.
	Claim 13 is clarifying the abstract idea, i.e., updating the parameter data using a scaled gradient of the loss data, used to perform the mathematical operations and is therefore directed towards clarifying the abstract idea of the mathematical operations, by specifying a scaled gradient of the loss data is used, and therefore fails to amount to significantly more than the judicial exception.
	Claim 14 is merely a recitation of an insignificant extra-solution activity of data gathering (see 2106.05(g)), which does not integrate a judicial exception into practical application. Further, the insignificant extra-solution activity of data gathering is also WURC, see MPEP 2106.05(d)(II) “The courts have recognized the following computer functions as well-understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. i. Receiving or transmitting data over a network”. Additionally, this limitation recites “one or more sensor” which is interpreted as generic machinery or computer component being used in its ordinary capacity to measure data, therefore, this limitation invokes a machinery merely as a tool to perform an existing process (see MPEP 2106.05(f)(2)).
Regarding Claim 15:
Step 2A of the subject matter eligibility test (see MPEP 2106.04).  
Prong One: 
Claim 15 recites (“sets forth” or “describes”) the abstract ideas as follows: 
“generating, via the machine learning system, prediction data based on the sensor data;” – this limitation is recited at a high level of generality and is directed to the abstract idea of a mental process, i.e., generating prediction data based on sensor data (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.] “via the machine learning system” will be interpreted under Prong Two.
“generating pseudo-reference data based on a gradient of a predetermined function evaluated with the prediction data;” – this limitation is recited at a high level of generality, and when interpreted in view of the specification, is directed to the abstract idea of a mental process, i.e., as pseudo-reference data refers to creating labels for unlabeled data. (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.]). Additionally, this limitation recites the abstract idea of a mathematical concept (see MPEP 2106.04(a)(2)), i.e., as a gradient, refers to calculating the mathematical operation.
“generating loss data based on the pseudo-reference data and the prediction data;” – this limitation is recited at a high level of generality is directed to the abstract idea of a mathematical concept (see MPEP 2106.04(a)(2)), i.e., generating loss data, when read in light of the specification, refers to calculating the loss function.
“performing, via the machine learning system, a task in the second domain after the parameter data has been updated; and” – this limitation is recited at a high level of generality and is directed to the abstract idea of a mental process, i.e., performing a task in the second domain (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.] “via the machine learning system” will be interpreted under Prong Two.
Prong Two: Claim 15 does not include additional elements that integrate the mental process into a practical application. 
“a processor;” – Does not particularly nor specifically identify a machine because a processor is a generic computer component within a generic computer environment and does not do more to integrate an abstract idea in to a practical application. Processors are merely using an apparatus/computer with instructions, as a tool to perform the abstract idea (MPEP 2106.05(f)). 
“a non-transitory computer readable medium in data communication with the processor, the non-transitory computer readable medium having computer readable data including instructions stored thereon that, when executed by the processor, cause the processor to perform” – Does not particularly nor specifically identify a machine because a computer readable medium is a generic computer component within a generic computer environment and does not do more to integrate an abstract idea in to a practical application. A computer readable medium are merely using an apparatus/computer with instructions, as a tool to perform an existing process in its ordinary capacity (e.g., to receive, store or transmit data) (MPEP 2106.05(f)(2)). 
“obtaining the sensor data from the second domain;” – is merely a recitation of an insignificant extra-solution activity of data gathering (see 2106.05(g)).
“via the machine learning system” – is recited at a high level of generality and is merely an instruction to apply an abstract idea (see MPEP 2106.05(f)), i.e., generating prediction data, in a generic computer environment using generic computer functions.
“updating parameter data of the machine learning system based on the loss data;” – is recited at a high level of generality and is merely a recitation of an insignificant extra solution activity of data manipulation (see 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application.
Step 2B of the subject matter eligibility test (see MPEP 2106.05).
When considered individually or in combination, the additional limitations and elements of claim 15 does not amount to significantly more than the judicial exception for the reasons as discussed above as to why the additional limitations to not integrate the abstract idea into a practical application.
“a processor;” – Invokes a computer merely as a tool for performing an existing process (see MEPP 2106.05(f)(2)) and therefore fails to amount to significantly more than the judicial exception.
“a non-transitory computer readable medium in data communication with the processor, the non-transitory computer readable medium having computer readable data including instructions stored thereon that, when executed by the processor, cause the processor to perform” – Invokes a computer merely as a tool for performing an existing process (see MEPP 2106.05(f)(2)) and therefore fails to amount to significantly more than the judicial exception.
“obtaining the sensor data from the second domain;” – is merely a recitation of an insignificant extra-solution activity of data gathering (see 2106.05(g)), which does not integrate a judicial exception into practical application. Further, the insignificant extra-solution activity of data gathering is also WURC, see MPEP 2106.05(d)(II) “The courts have recognized the following computer functions as well-understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. i. Receiving or transmitting data over a network”.
“via the machine learning system” – is recited at a high level of generality and is merely an instruction to apply an abstract idea, i.e., generating prediction data, and therefore fails to amount to significantly more than the judicial exception. See MPEP 2106.05(f).
“updating parameter data of the machine learning system based on the loss data;” – is merely a recitation of an insignificant extra solution activity of data manipulation (see MPEP 2106.05(g)) “Selecting a particular data source or type of data to be manipulated”; which does not integrate a judicial exception in to a practical application. Further, as updating parameter data based on loss requires repetitive calculations for minimizing the loss, this limitation further recites a well-understood, routine and conventional activity under MPEP 2106.05(d) II (ii) of performing repetitive calculations.
	Claim 16 is clarifying the machine learning system as performing the task of a classifier, therefore the claim is directed towards the abstract idea of a mental process, i.e., classifying (concepts performed in the human mind, including observation and evaluation [MPEP 2106.04(a)(2) III. C.]) and therefore fails to amount to significantly more than the judicial exception.
	Claim 17 is clarifying the machine learning system, i.e., the machine learning system uses the same predetermined function (which is recited at a high level of generality), and is therefore directed towards clarifying the instructions to apply the abstract idea and therefore fails to amount to significantly more than the judicial exception.
	Claim 18 is clarifying the abstract idea, i.e., updating the parameter data using a scaled gradient of the loss data, and is therefore directed towards clarifying the abstract idea of mathematical operations, as using a scaled gradient is a mathematical operation, and therefore fails to amount to significantly more than the judicial exception.
Claim 19 recites “an image sensor or a microphone” which is interpreted as generic machinery or computer component(s) being used in their ordinary capacity to measure data, therefore, this limitation invokes a machinery merely as a tool to perform an existing process (see MPEP 2106.05(f)(2)). Additionally, claim 19 is merely a recitation of an insignificant extra-solution activity of data gathering, i.e., sensor data is digital image data from the image sensor or digital audio data from the microphone, (see 2106.05(g)), which does not integrate a judicial exception into practical application. Further, the insignificant extra-solution activity of data gathering is also WURC, see MPEP 2106.05(d)(II) “The courts have recognized the following computer functions as well-understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. i. Receiving or transmitting data over a network”.
Claim 20 recites “an actuator” which is interpreted as generic machinery or computer component(s) being used in their ordinary capacity to measure data, therefore, this limitation invokes a machinery merely as a tool to perform an existing process (see MPEP 2106.05(f)(2)). Lastly, reciting the actuator is used in the second domain is a field of use, see MPEP 2106 (citing Intellectual Ventures I LLC v. Capital One Bank (USA), N.A., 792 F.3d 1363, 1366, 115 USPQ2d 1636, 1639 (Fed. Cir. 2015) (“An abstract idea does not become nonabstract by limiting the invention to a particular field of use or technological environment, such as the Internet [or] a computer”) and therefore fails to amount to significantly more than the judicial exception.
Claim Rejection - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –

(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.


(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.

Claims 15, 16 and 18 are rejected under 35 U.S.C. 102(a)(1) [and/or 102(a)(2)] as being unpatentable over Xu, et al. (US 20220027672 A1) (hereafter referred to as “Xu”).
Regarding claim 15, Xu recites “A system comprising:” (Xu at 0001: For example, at least one embodiment pertains to processors or computing systems used to generate one or more partial or pseudolabels and combine, using one or more neural networks, said partial or pseudolabels with information about features in an input image to generate a label, according to various novel techniques described herein.)
“a processor;” (Xu at 0099: In at least one embodiment, any portion of code and/or data storage 701 may be internal or external to one or more processors or other hardware logic devices or circuits.)
“a non-transitory computer readable medium in data communication with the processor, the non-transitory computer readable medium having computer readable data including instructions stored thereon that, when executed by the processor, cause the processor to perform” (Xu at 0567: In at least one embodiment, code is stored on a computer-readable storage medium, for example, in form of a computer program comprising a plurality of instructions executable by one or more processors. In at least one embodiment, a computer-readable storage medium is a non-transitory computer-readable storage medium that excludes transitory signals…)
“a method for adapting a machine learning system that is trained with training data in a first domain to operate with sensor data in a second domain, the method including:” (Xu at 0084: A training framework performs region grow to specify foreground and background points in a weak supervision technique 304, in an embodiment, as follows, for commonly used weak annotation methods in medical image analysis:… See also 0068) [The weak annotated image set is the source domain or first domain, the unlabeled image set is the target domain or second domain.]
“obtaining the sensor data from the second domain;” (Xu at 0068: In at least one embodiment, a pseudolabel 210 is data comprising classes for unlabeled data, such as an input image 202, as if said classes were true labels. In at least one embodiment a pseudolabel 210 comprises information about an input image 202 that is less specific than a true label in that said information is predicted and not known as true. For example, in an embodiment, a pseudolabel 210 can indicate that a region of an input image 202 comprises a kidney, while a label directly indicates which pixels in an input image 202 correspond to a kidney. In another example, a pseudolabel 210 indicates that a region of an input image 202 comprises something medically interesting, while a label indicates exactly which pixels correspond to that medically interesting thing, and what that medically interesting thing is, in an embodiment.) [The unlabeled data is the second domain, such as an image comprises a kidney, which is obtained from a sensor.]
“generating, via the machine learning system, prediction data based on the sensor data;” (Xu at 0072: In at least one embodiment, a training framework updates 222 each pseudolabel 210 in each pseudolabel group 212, 214, 216 according to a prediction map 220 to generate one or more feature maps 226, 228, 230 FMj. In at least one embodiment, an update 222 is software instructions that, when executed, adjust information contained in a pseudolabel 210 according to information about objects contained in a prediction map 220.) [A training framework update, i.e., a machine learning system, generates prediction maps, i.e., prediction data. Like the prediction data, the prediction map represents an intermediate output of the machine learning model.]
“generating pseudo-reference data based on a gradient of a predetermined function evaluated with the prediction data;” (Xu at 0072: 

    PNG
    media_image1.png
    239
    471
    media_image1.png
    Greyscale
) [The pseudo labels are updated, i.e., generating pseudo-reference data, based on the gradient descent of the backpropagation algorithm of the local loss which evaluates the prediction map, i.e., prediction data, and pseudolabels.]
“generating loss data based on the pseudo-reference data and the prediction data;” (Xu at 0072: 
    PNG
    media_image1.png
    239
    471
    media_image1.png
    Greyscale
.) [The pseudo labels are updated, i.e., generating pseudo-reference data, based on the gradient descent of the backpropagation algorithm of the local loss which evaluates the prediction map and pseudolabels.]
“updating parameter data of the machine learning system based on the loss data; and” (Xu at 0072: A model 218 M is trained (by backpropagation 240) by a training framework, in an embodiment, using at least a preliminary context loss function according to pseudolabels 210, generated by said training framework from weak annotations 238, to be updated 222 by a prediction map 220 X. See also 0073 In at least one embodiment, each locally updated model Mj is used, by a training framework, to determine updated feature maps 224 from a prediction map 220 X and pseudolabels 210 in each pseudolabel group 212, 214, 216.)
“performing, via the machine learning system, a task in the second domain after the parameter has been updated.” (Xu at 0072: In at least one embodiment, a model 218 is data values and software instructions that, when executed, predict a segmentation boundary between background and foreground objects in an input image 202. In at least one embodiment, a model 218 is a 3D U-Net. In at least one embodiment, a model 218 is any other type of neural network further described herein. A model 218 M is trained (by backpropagation 240) by a training framework…) [The trained model, 218, is used to predict boundary between background and foreground objects, i.e., a task in the second domain after the parameter data has been updated.]
Regarding claim 16, Xu recites “The system of claim 15,” and Xu further recites “wherein: the machine learning system is a classifier configured to perform the task of generating output data that classifies input data;” (Xu at 0061: In at least one embodiment, supervision comprises input information that describes one or more aspects of training data 104, such as objects, features, or styles, or a classification for said training data 104, to assist training one or more untrained neural networks 108 by a training framework 106. In at least one embodiment, supervision is strong, wherein input information provides direct identification of an object, feature, style, or other aspect of an item, such as an image, in training data 104.)
And Xu further recites “and the predetermined function is a loss function relating to the task.” (Xu at 0072: In at least one embodiment, a model 218 is a 3D U-Net. In at least one embodiment, a model 218 is any other type of neural network further described herein. A model 218 M is trained (by backpropagation 240) by a training framework, in an embodiment, using at least a preliminary context loss function according to pseudolabels 210, generated by said training framework from weak annotations 238, to be updated 222 by a prediction map 220 X. In at least one embodiment, a context loss is a measurement comprising a distance between objects or locations of objects (a context) in one or more pseudolabels 210 and a predicted segmentation boundary from a prediction map 220 X.)
Regarding claim 18, Xu recites “The system of claim 15,” and Xu further recites “wherein the parameter data is updated using a scaled gradient of the loss data.” (Xu at 0072: 
    PNG
    media_image3.png
    325
    583
    media_image3.png
    Greyscale
) [The loss function is computed for each pseudo-label in each pseudo-label group. This loss function is used to update the model via gradient descent. In this process, the system scales the gradient by a weight of y = 1 in order to control the impact of each pseudo-label on the overall model, i.e., the parameter data, the prediction maps, are updated using a scaled gradient of the loss data.]
Claim Rejection - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.

The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary.  Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 2, 4, 5, 6, 7, 19 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Xu in view of Murez, et al. (US 11620527 B2) (hereafter referred to as “Murez”).
Regarding claim 1, Xu recites “A computer-implemented method for adapting a machine learning system that is trained with training data in a first domain to operate with sensor data in a second domain, the computer-implemented method comprising:” (Xu at 0084: A training framework performs region grow to specify foreground and background points in a weak supervision technique 304, in an embodiment, as follows, for commonly used weak annotation methods in medical image analysis:… See also 0068) [The weak annotated image set is the source domain or first domain, the unlabeled image set is the target domain or second domain.]
“obtaining the sensor data from the second domain;” (Xu at 0068: In at least one embodiment, a pseudolabel 210 is data comprising classes for unlabeled data, such as an input image 202, as if said classes were true labels. In at least one embodiment a pseudolabel 210 comprises information about an input image 202 that is less specific than a true label in that said information is predicted and not known as true. For example, in an embodiment, a pseudolabel 210 can indicate that a region of an input image 202 comprises a kidney, while a label directly indicates which pixels in an input image 202 correspond to a kidney. In another example, a pseudolabel 210 indicates that a region of an input image 202 comprises something medically interesting, while a label indicates exactly which pixels correspond to that medically interesting thing, and what that medically interesting thing is, in an embodiment.) [The unlabeled data is the second domain, such as an image comprises a kidney, which is obtained from a sensor.]
“generating, via the machine learning system, prediction data based on the sensor data;” (Xu at 0072: In at least one embodiment, a training framework updates 222 each pseudolabel 210 in each pseudolabel group 212, 214, 216 according to a prediction map 220 to generate one or more feature maps 226, 228, 230 FMj. In at least one embodiment, an update 222 is software instructions that, when executed, adjust information contained in a pseudolabel 210 according to information about objects contained in a prediction map 220.) [A training framework update, i.e., a machine learning system, generates prediction maps, i.e., prediction data. Like the prediction data, the prediction map represents an intermediate output of the machine learning model.]
“generating pseudo-reference data based on a gradient of a predetermined function evaluated with the prediction data;” (Xu at 0072: 

    PNG
    media_image1.png
    239
    471
    media_image1.png
    Greyscale
) [The pseudo labels are updated, i.e., generating pseudo-reference data, based on the gradient descent of the backpropagation algorithm of the local loss which evaluates the prediction map, i.e., prediction data, and pseudo labels.]
“generating loss data based on the pseudo-reference data and the prediction data;” (Xu at 0072: 
    PNG
    media_image1.png
    239
    471
    media_image1.png
    Greyscale
) [The pseudo labels are updated, i.e., generating pseudo-reference data, based on the gradient descent of the backpropagation algorithm of the local loss, i.e., generated loss data, which evaluates the prediction map and pseudolabels, i.e., based on the pseudo-reference data and the prediction.]
“updating parameter data of the machine learning system based on the loss data;” (Xu at 0072: A model 218 M is trained (by backpropagation 240) by a training framework, in an embodiment, using at least a preliminary context loss function according to pseudolabels 210, generated by said training framework from weak annotations 238, to be updated 222 by a prediction map 220 X. See also 0073 In at least one embodiment, each locally updated model Mj is used, by a training framework, to determine updated feature maps 224 from a prediction map 220 X and pseudolabels 210 in each pseudolabel group 212, 214, 216.)
“performing, via the machine learning system, a task in the second domain after the parameter data has been updated; and” (Xu at 0072: In at least one embodiment, a model 218 is data values and software instructions that, when executed, predict a segmentation boundary between background and foreground objects in an input image 202. In at least one embodiment, a model 218 is a 3D U-Net. In at least one embodiment, a model 218 is any other type of neural network further described herein. A model 218 M is trained (by backpropagation 240) by a training framework…) [The trained model, 218, is used to predict boundary between background and foreground objects, i.e., a task in the second domain after the parameter data has been updated.]
However, Xu does not explicitly recite “controlling an actuator based on the task performed in the second domain.” 
On the other hand, Murez recites “controlling an actuator based on the task performed in the second domain.” (Murez at cl. 13, ln. 20-26: Thus, the annotations can then be utilized to cause an automatic operation related to controlling a component of the autonomous vehicle.)
Xu and Murez4 are analogous arts in the field of machine learning in the field of endeavor of domain adaptation. A person skilled in the art, before the filing date of the present application, would be able to modify Xu with Murez controlling an actuator based on the task performed in the second domain with the motivation being “(cl. 13, ln 10-9: [t]he invention according to embodiments of the present disclosure is of particular value to fully autonomous navigation systems for vehicle manufacturers. TS2 will significantly reduce the amount of annotated real-world training data needed to train their perception and sensing algorithms. Furthermore, thanks to its domain agnostic feature extraction capability, TS2 produces more robust results when navigating in novel or unseen conditions, such as a new city or in rare weather conditions (e.g., snow, fog, rain).”
Regarding claim 2, Xu in view of Murez recites “The computer-implemented method of claim 1,” and Xu further recites “wherein the machine learning system is a classifier configured to perform the task of generating output data that classifies input data.” (Xu at 0061: In at least one embodiment, supervision comprises input information that describes one or more aspects of training data 104, such as objects, features, or styles, or a classification for said training data 104, to assist training one or more untrained neural networks 108 by a training framework 106. In at least one embodiment, supervision is strong, wherein input information provides direct identification of an object, feature, style, or other aspect of an item, such as an image, in training data 104.)
Regarding claim 4, Xu in view of Murez recites “The computer-implemented method of claim 1,” and Xu further recites “wherein the predetermined function is a loss function relating to the task performed by the machine learning system.” (Xu at 0072: In at least one embodiment, a model 218 is a 3D U-Net. In at least one embodiment, a model 218 is any other type of neural network further described herein. A model 218 M is trained (by backpropagation 240) by a training framework, in an embodiment, using at least a preliminary context loss function according to pseudolabels 210, generated by said training framework from weak annotations 238, to be updated 222 by a prediction map 220 X. In at least one embodiment, a context loss is a measurement comprising a distance between objects or locations of objects (a context) in one or more pseudolabels 210 and a predicted segmentation boundary from a prediction map 220 X.) [The context loss function measures the distance between objects in the pseudo labels and the prediction map. This loss is used to update the pseudo-labels, improving the accuracy of the model, i.e., related to the task performed by the machine learning system.]
Regarding claim 5, Xu in view of Murez recites “The computer-implemented method of claim 4,” and Murez recites “wherein the loss function is a cross-entropy loss function, a squared loss function, a hinge loss function, a tangent loss function, a polyloss function, or a logistic loss function.” (Murez at cl. 10, ln. 3-10: 
    PNG
    media_image4.png
    158
    407
    media_image4.png
    Greyscale
) [The loss function comprises a cross entropy loss function.]
A person skilled in the art, before the filing date of the present application, would be modify Xu with Murez to recite wherein the loss function is a cross-entropy loss function, a squared loss function, a hinge loss function, a tangent loss function, a polyloss function, or a logistic loss function with the motivation being “(cl. 2, ln. 46-47) [a] cross entropy loss function that is defined as a number of correct classifications of the discriminator is optimized.”
Regarding claim 6, Xu in view of Murez recites “The computer-implemented method of claim 1,” and Xu further recites “wherein the parameter data is updated using a scaled gradient of the loss data.” (Xu at 0072: 
    PNG
    media_image3.png
    325
    583
    media_image3.png
    Greyscale
) [The loss function is computed for each pseudo-label in each pseudo-label group. This loss function is used to update the model via gradient descent. In this process, the system scales the gradient by a weight of y = 1 in order to control the impact of each pseudo-label on the overall model, i.e., the parameter data, the prediction maps, are updated using a scaled gradient of the loss data.]
Regarding claim 7, Xu in view of Murez recites “The computer-implemented method of claim 1,” and, although Xu recites sensors (Xu at 0136) it is not explicitly clear that these sensors are in the same embodiment, therefore, in the interest of compact prosecution, Murez further recites “wherein the sensor data includes digital image data or digital audio data obtained from one or more sensors.” (Murez at cl. 7, ln. 55-6: …which maps an input image obtained from a target domain sensor (element 804) into a feature space…) [The input image, i.e., digital image data, is obtained from a target domain sensor, i.e., one or more sensors.]
The motivation rationale used to modify Xu with Murez in claim 1 is similarly applicable to claim 7.
Regarding claim 19, Xu recites “The system of claim 15,” however Xu does not recites “further comprising: an image sensor or a microphone;”
“wherein the sensor data includes digital image data from the image sensor or digital audio data obtained from the microphone.”
On the other hand, Murez recites “further comprising: an image sensor or a microphone;” (Murez at cl. 7, ln. 52-6: FIG. 8 depicts an image processing system (element 800) comprising a convolutional neural network (CNN). CNNs are made of two parts: a deep feature extractor module (element 802), which maps an input image obtained from a target domain sensor (element 804) into a feature space… See also Fig. 8.) 
Murez further recites “wherein the sensor data includes digital image data from the image sensor or digital audio data obtained from the microphone.” (Murez at cl. 7, ln. 55-6: …which maps an input image obtained from a target domain sensor (element 804) into a feature space…)
The motivation rationale used to modify Xu with Murez in claim 1 is similarly applicable to claim 19.
Regarding claim 20, Xu recites “The system of claim 15,” however Xu does not recite “further comprising: an actuator, wherein, the processor is configured to generate control data based on the task performed by the machine learning system with respect to other sensor data in the second domain, and the actuator is controlled based on the control data.”
 On the other hand, Murez recites “further comprising: an actuator, wherein, the processor is configured to generate control data based on the task performed by the machine learning system with respect to other sensor data in the second domain, and the actuator is controlled based on the control data.” (Murez at cl. 13, ln. 20-7: The annotations for the target image domain obtained by the TS2 framework can be used for detection and recognition of objects, such as vehicles, pedestrians, and traffic signs, under different weather conditions (e.g., rain, snow, fog) and lighting conditions (e.g., low light, bright light). Thus, the annotations can then be utilized to cause an automatic operation related to controlling a component of the autonomous vehicle.)
The motivation rationale used to modify Xu with Murez in claim 1 is similarly applicable to claim 20.
Claim 3 is rejected under 35 U.S.C. 103 as being unpatentable over Xu and Murez in further view of Hu, et al. (US 20230244924 A1) (hereafter referred to as “Hu”).
Regarding claim 3, Xu in view of Murez recites “The computer-implemented method of claim 1,” however neither Xu nor Murez recites “wherein the machine learning system is trained in the first domain using the same predetermined function that is used to generate the pseudo-reference data in the second domain.”
On the other hand, Hu recites “wherein the machine learning system is trained in the first domain using the same predetermined function that is used to generate the pseudo-reference data in the second domain.” (Hu at 0046: The pseudo-labeled dataset may then be used to train a student network 214 using the loss function (L) as shown by Equation (5)… See also Hu at 0044: The human-annotated dataset Ol may also be exploited to train the teacher network 206 (which may be represented as θ.sub.1) using a conventional loss function (L) for object detection, where L may be composed by the classification loss and regression loss for bounding box prediction.) [The teacher network, i.e., the machine learning system generating the pseudo-reference data in the second domain, and the student network, i.e., the machine learning system is trained in the first domain, are both trained with L, i.e., using the same predetermined function.]
Xu, Murez and Hu are analogous arts in the field of machine learning in the field of endeavor of domain adaptation. A person skilled in the art, before the filing date of the present application, would modify Xu and Murez with Hu to recite wherein the machine learning system is trained in the first domain using the same predetermined function that is used to generate the pseudo-reference data in the second domain with the motivation being “(0044) [t]he human-annotated dataset Ol may also be exploited to train the teacher network 206 (which may be represented as θ.sub.1) using a conventional loss function (L) for object detection, where L may be composed by the classification loss and regression loss for bounding box prediction[,]” i.e., to compare the outputs of both models.
Claims 8, 9, 11, 12, 13, 14 are rejected under 35 U.S.C. 103 as being unpatentable over Xu, et al., (US 20220027672 A1) (hereafter referred to as “Xu”) in view of Wei, et al., (US 20110281253 A1) (hereafter referred to as “Wei”) in further view of Murez, et al. (US 11620527 B2) (hereafter referred to as “Murez”).
Regarding claim 8, Xu recites  “A computer-implemented method for test-time adaptation of a machine learning system from a source domain to a target domain, the machine learning system having been trained with training data of the source domain, the computer-implemented method comprising” (Xu at 0112: In at least one embodiment, semi-supervised learning may be used, which is a technique in which in training dataset 802 includes a mix of labeled and unlabeled data. In at least one embodiment, training framework 804 may be used to perform incremental learning, such as through transferred learning techniques. In at least one embodiment, incremental learning enables trained neural network 808 to adapt to new dataset 812 without forgetting knowledge instilled within trained neural network 808 during initial training.) [By generating pseudo-labels from the partially labeled or unlabeled data, the model is adapting new data distribution without explicit supervision, i.e., the model is learning and updating itself based on the new data it encounters, i.e., time-test adaptation.]
“obtaining sensor data from the target domain;” (Xu at 0068: In at least one embodiment, a pseudolabel 210 is data comprising classes for unlabeled data, such as an input image 202, as if said classes were true labels. In at least one embodiment a pseudolabel 210 comprises information about an input image 202 that is less specific than a true label in that said information is predicted and not known as true. For example, in an embodiment, a pseudolabel 210 can indicate that a region of an input image 202 comprises a kidney, while a label directly indicates which pixels in an input image 202 correspond to a kidney. In another example, a pseudolabel 210 indicates that a region of an input image 202 comprises something medically interesting, while a label indicates exactly which pixels correspond to that medically interesting thing, and what that medically interesting thing is, in an embodiment.) [The unlabeled data is the second domain, such as an image comprises a kidney, which is obtained from a sensor.]
“generating, via the machine learning system, prediction data based on the sensor data;” (Xu at 0072: In at least one embodiment, a training framework updates 222 each pseudolabel 210 in each pseudolabel group 212, 214, 216 according to a prediction map 220 to generate one or more feature maps 226, 228, 230 FMj. In at least one embodiment, an update 222 is software instructions that, when executed, adjust information contained in a pseudolabel 210 according to information about objects contained in a prediction map 220.) [A training framework update, i.e., a machine learning system, generates prediction maps, i.e., prediction data. Like the prediction data, the prediction map represents an intermediate output of the machine learning model.]
Xu recites “generating loss data based on a gradient of a predetermined function evaluated with the prediction data;” (Xu at 0072: 

    PNG
    media_image1.png
    239
    471
    media_image1.png
    Greyscale
) [The pseudo labels are updated, i.e., generating pseudo-reference data, based on the gradient descent of the backpropagation algorithm of the local loss which evaluates the prediction map and pseudolabels.]
Xu further recites “updating parameter data of the machine learning system based on the loss data;” (Xu at 0072: A model 218 M is trained (by backpropagation 240) by a training framework, in an embodiment, using at least a preliminary context loss function according to pseudolabels 210, generated by said training framework from weak annotations 238, to be updated 222 by a prediction map 220 X. See also 0073 In at least one embodiment, each locally updated model Mj is used, by a training framework, to determine updated feature maps 224 from a prediction map 220 X and pseudolabels 210 in each pseudolabel group 212, 214, 216.)
“performing, via the machine learning system, a task in the second domain after the parameter data has been updated; and” (Xu at 0072: In at least one embodiment, a model 218 is data values and software instructions that, when executed, predict a segmentation boundary between background and foreground objects in an input image 202. In at least one embodiment, a model 218 is a 3D U-Net. In at least one embodiment, a model 218 is any other type of neural network further described herein. A model 218 M is trained (by backpropagation 240) by a training framework…) [The trained model, 218, is used to predict boundary between background and foreground objects, i.e., a task in the second domain after the parameter data has been updated.]
However Xu does not recite:
“based on a negative convex conjugate of a predetermined function applied to a gradient of the predetermined function,”
“controlling an actuator based on the task performed in the target domain.”
On the other hand, Wei recites based on a negative convex conjugate of a predetermined function applied to a gradient of the predetermined function (Wei at 0031: The conjugate gradient methods may be the best suitable for global optimizations, especially when the objective function to be maximized, here the learning efficacy p(T1, T2, . . . Tm), is a linear or concave function in the convex body defined by the constraints (9) and (10), that is, the optimization problem is convex programming, in which case a conjugate gradient algorithm is guaranteed to converge to the optimal solution within m iteration steps of line search and search direction update…) [A concave conjugate is functionally equivalent negative convex conjugate.]
Xu and Wei are analogous arts in the field of machine learning in the field of endeavor labeling of datasets. A person skilled in the art, before the filing date of the present application, would be motivated to modify Xu with Wei to recite based on a negative convex conjugate of a predetermined function applied to a gradient of the predetermined function with the motivation being “(0031) conjugate gradient algorithm is guaranteed to converge to the optimal solution within m iteration steps of line search and search direction update, each of which involves a gradient calculation and a few objective function evaluations.”
On the other hand, Murez recites “controlling an actuator based on the task performed in the target domain.”
Xu, Wei and Murez5 are analogous arts in the field of machine learning in the field of endeavor of labeling of datasets. A person skilled in the art, before the filing date of the present application, would be able to modify Xu and Wei with Murez controlling an actuator based on the task performed in the second domain with the motivation being “(cl. 13, ln 10-9: [t]he invention according to embodiments of the present disclosure is of particular value to fully autonomous navigation systems for vehicle manufacturers. TS2 will significantly reduce the amount of annotated real-world training data needed to train their perception and sensing algorithms. Furthermore, thanks to its domain agnostic feature extraction capability, TS2 produces more robust results when navigating in novel or unseen conditions, such as a new city or in rare weather conditions (e.g., snow, fog, rain).”
Regarding claim 9, Xu, Wei and Murez recites “The computer-implemented method of claim 8,” and Xu recites “wherein the machine learning system is a classifier configured to perform the task of generating output data that classifies input data.” (Xu at 0061: In at least one embodiment, supervision comprises input information that describes one or more aspects of training data 104, such as objects, features, or styles, or a classification for said training data 104, to assist training one or more untrained neural networks 108 by a training framework 106. In at least one embodiment, supervision is strong, wherein input information provides direct identification of an object, feature, style, or other aspect of an item, such as an image, in training data 104.)
Regarding claim 11, Xu, Wei and Murez recites “The computer-implemented method of claim 8,” and Xu further recites “wherein the predetermined function is a loss function relating to the task performed by the machine learning system.” (Xu at 0072: In at least one embodiment, a model 218 is a 3D U-Net. In at least one embodiment, a model 218 is any other type of neural network further described herein. A model 218 M is trained (by backpropagation 240) by a training framework, in an embodiment, using at least a preliminary context loss function according to pseudolabels 210, generated by said training framework from weak annotations 238, to be updated 222 by a prediction map 220 X. In at least one embodiment, a context loss is a measurement comprising a distance between objects or locations of objects (a context) in one or more pseudolabels 210 and a predicted segmentation boundary from a prediction map 220 X.) [The context loss function measures the distance between objects in the pseudo labels and the prediction map. This loss is used to update the pseudo-labels, improving the accuracy of the model, i.e., related to the task performed by the machine learning system.]
Regarding claim 12, “The computer-implemented method of claim 11,” “wherein the loss function is a cross- entropy loss function, a squared loss function, a hinge loss function, a tangent loss function, a polyloss function, or a logistic loss function.” (Murez at cl. 10, ln. 3-10: 
    PNG
    media_image4.png
    158
    407
    media_image4.png
    Greyscale
) [The loss function comprises a cross entropy loss function.]
A person skilled in the art, before the filing date of the present application, would be modify Xu and Wei with Murez to recite wherein the loss function is a cross-entropy loss function, a squared loss function, a hinge loss function, a tangent loss function, a polyloss function, or a logistic loss function with the motivation being “(cl. 2, ln. 46-47) [a] cross entropy loss function that is defined as a number of correct classifications of the discriminator is optimized.”
Regarding claim 13, Xu, Wei and Murez recites “The computer-implemented method of claim 8,” and Wei further recites “wherein the parameter data is updated using a scaled gradient of the loss data.” (Xu at 0072: 
    PNG
    media_image3.png
    325
    583
    media_image3.png
    Greyscale
) [The loss function is computed for each pseudo-label in each pseudo-label group. This loss function is used to update the model via gradient descent. In this process, the system scales the gradient by a weight of y = 1 in order to control the impact of each pseudo-label on the overall model, i.e., the parameter data, the prediction maps, are updated using a scaled gradient of the loss data.]
Regarding claim 14, Xu, Wei and Murez recites “The computer-implemented method of claim 8,” and, although Xu recites sensors (Xu at 0136) it is not explicitly clear that these sensors are in the same embodiment, therefore, in the interest of compact prosecution, Murez further recites “wherein the sensor data includes digital image data or digital audio data obtained from one or more sensors.” (Murez at cl. 7, ln. 55-6: …which maps an input image obtained from a target domain sensor (element 804) into a feature space…) [The input image, i.e., digital image data, is obtained from a target domain sensor, i.e., one or more sensors.]
The motivation rationale used to modify Xu and Wei with Murez in claim 8 is similarly applicable to claim 14.
 Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable in view of Xu and Wei and Murez in further view of Hu.
Regarding claim 10, Xu, Wei and Murez recites “The computer-implemented method of claim 8,” however neither Xu, Wei nor Murez recite “wherein the machine learning system is trained in the source domain using the same predetermined function.” 
On the other hand, Hu recites wherein the machine learning system is trained in the source domain using the same predetermined function. (Hu at 0046: The pseudo-labeled dataset may then be used to train a student network 214 using the loss function (L) as shown by Equation (5)… See also Hu at 0044: The human-annotated dataset Ol may also be exploited to train the teacher network 206 (which may be represented as θ.sub.1) using a conventional loss function (L) for object detection, where L may be composed by the classification loss and regression loss for bounding box prediction.) [The teacher network, i.e., the machine learning system generating the pseudo-reference data in the second domain, and the student network, i.e., the machine leaning system is trained in the first domain, are both trained with L, i.e., using the same predetermined function.]
Xu, Wei, Murez and Hu are analogous arts in the field of machine learning in the field of endeavor of labeling of datasets. A person skilled in the art, before the filing date of the present application, would modify Xu, Wei and Murez with Hu to recite wherein the machine learning system is trained in the first domain using the same predetermined function that is used to generate the pseudo-reference data in the second domain with the motivation being “(0044) [t]he human-annotated dataset Ol may also be exploited to train the teacher network 206 (which may be represented as θ.sub.1) using a conventional loss function (L) for object detection, where L may be composed by the classification loss and regression loss for bounding box prediction[,]” i.e., to compare the outputs of both models.
Claim 17 is rejected under 35 U.S.C. 103 as being unpatentable over Xu in further view of Hu.
Regarding claim 17, Xu recites “The system of claim 15,” Xu does not recite “wherein the machine learning system is trained in the first domain using the same predetermined function that is used to generate the pseudo-reference data in the second domain.”
On the other hand, Hu recites “wherein the machine learning system is trained in the first domain using the same predetermined function that is used to generate the pseudo-reference data in the second domain.” (Hu at 0046: The pseudo-labeled dataset may then be used to train a student network 214 using the loss function (L) as shown by Equation (5)… See also Hu at 0044: The human-annotated dataset Ol may also be exploited to train the teacher network 206 (which may be represented as θ.sub.1) using a conventional loss function (L) for object detection, where L may be composed by the classification loss and regression loss for bounding box prediction.) [The teacher network, i.e., the machine learning system generating the pseudo-reference data in the second domain, and the student network, i.e., the machine learning system is trained in the first domain, are both trained with L, i.e., using the same predetermined function.]
Xu and Hu are analogous arts in the field of machine learning in the field of endeavor of domain adaptation. A person skilled in the art, before the filing date of the present application, would modify Xu with Hu to recite wherein the machine learning system is trained in the first domain using the same predetermined function that is used to generate the pseudo-reference data in the second domain with the motivation being “(0044) [t]he human-annotated dataset Ol may also be exploited to train the teacher network 206 (which may be represented as θ.sub.1) using a conventional loss function (L) for object detection, where L may be composed by the classification loss and regression loss for bounding box prediction[,]” i.e., to compare the outputs of both models.
Examiner’s Note
	Examiner cites particular columns, paragraphs, figures and line numbers in the references as applied to the claims above for the convenience of the applicant. Although the specified citations are representative of the teachings in the art and are applied to the specific limitations within the individual claim, other passages and figures may apply as well. It is respectfully requested that, in preparing responses, the applicant fully consider the references in their entirety as potentially teaching all or part of the claimed invention, as well as the context of the passage as taught by the prior art or disclosed by the examiner.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to EVAN LEHRER whose telephone number is (703) 756-1466. The examiner can normally be reached Mon-Thr. 7AM-5PM EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, David Yi can be reached on (571) 270-7519 
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000. 
/E.L./Examiner, Art Unit 2126

/DAVID YI/Supervisory Patent Examiner, Art Unit 2126                                                                                                                                                                                                                                                                                                                                                                                                         





    
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
        
            
    

    
        1 Applicant associates claim 18 with “during generation of pseudo-reference data” however claim 18 is associated with the step of “updating parameter data” limitation/step.
        2 Examiner notes that the applicants’ specification list several other fields of use such as a drill bit in a drill, a nozzle to spray chemicals or a locking mechanism in a door. Regardless, this limitation recites a field of use (see MPEP 2106.05(h)).
        3 Examiner notes that the applicants’ specification list several other fields of use such as a drill bit in a drill, a nozzle to spray chemicals or a locking mechanism in a door. Regardless, this limitation recites a field of use (see MPEP 2106.05(h)).
        4 For the purposes of compact prosecution, Murez also recites, at minimum, a computer-implemented method for adapting a machine learning system that is trained with training data in a first domain to operate with sensor data in a second domain, (Murez at 27: The method according to embodiments of the present disclosure includes two unique ways to improve the domain adaptation performance by re-designing the feature extractors that are learned… In short, an adversarial network (FIG. 4, module 412) is trained to distinguish the features coming from domain X from that of domain Y… See also 26: “target domain sensor”) and “a second domain” (Murez at 5: The feature extractor extracts features from both source domain (e.g., domain ‘A’) images and target domain images (e.g., domain ‘B’).) [The target data or Domain B is functionally equivalent to a second domain.]
        
        5 For the purposes of compact prosecution, Murez also recites, at minimum, a computer-implemented method for adapting a machine learning system that is trained with training data in a first domain to operate with sensor data in a second domain, (Murez at 27: The method according to embodiments of the present disclosure includes two unique ways to improve the domain adaptation performance by re-designing the feature extractors that are learned… In short, an adversarial network (FIG. 4, module 412) is trained to distinguish the features coming from domain X from that of domain Y… See also 26: “target domain sensor”) and “a second domain” (Murez at 5: The feature extractor extracts features from both source domain (e.g., domain ‘A’) images and target domain images (e.g., domain ‘B’).) [The target data or Domain B is functionally equivalent to a second domain.]