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 .
This action is in response to amendments filed January 30th, 2026, in which claims 1-5, 8-9, and 11-13 have been amended. No claims have been cancelled nor added. The amendments have been entered, and claims 1-5, 8-9, and 11-13 are currently pending in the case. Claims 1, 12 and 13 are independent claims.
The information disclosure statement (IDS) submitted on February 19, 2026 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Rejections - 35 USC § 101
35 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 1-13 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1:
Step 1: Claim 1 is directed to [a] computer implemented method, therefore it falls under the statuary category of a process.
Step 2A Prong 1: The claim recites, in part:
“the quality of the training data comprising a determination of a balance of the training data at the clinical site regarding different output classifications…” this encompasses the mental determination of a balance of observed training data regarding observe red output classifications.
“updating the parameter…, based on the received local updates to the parameter and the received metadata by combining the local updates to the parameter to determine an update… by weighting each local update according to the respective metadata such that local updates resulting from more balanced training data are given more weight than local updates resulting from less balanced training data.” This encompasses the mental update of an observed parameter based on observed local updates and weighting the update based on observed metadata. Further, this limitation is a mathematical concept.
Step 2A Prong 2: The judicial exception is not integrated into a practical application; the remaining limitations of the claim are as follows: “of the global model”, “in the global model”, “to the global model” these limitations are an additional element that generally links the use of the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(h). “sending information to the plurality of clinical sites to enable each of the plurality of clinical sites to create a local copy of the global model”, “receiving, from each of the plurality of clinical sites, i) a local update to a parameter in the global model…and ii) metadata indicating a quality of the training data at the respective clinical site” these limitations are an additional element that amounts to adding insignificant extra-solution activity to the judicial exception. See MPEP § 2106.05(g). “train the respective local copy of the global model on training data at the respective clinical site”, “obtained by training the local copy of the global model on the training data at the respective clinical site” the limitations are an additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP § 2106.05(f)(2).
Step 2B: The additional elements “train the respective local copy of the global model on training data at the respective clinical site”, “obtained by training the local copy of the global model on the training data at the respective clinical site”, taken individually and in combination, do not provide an inventive concept of significantly more than the abstract idea itself for the reasons set forth in step 2A prong 2 above. Further, “sending information to the plurality of clinical sites to enable each of the plurality of clinical sites to create a local copy of the global model”, “receiving, from each of the plurality of clinical sites, i) a local update to a parameter in the global model…and ii) metadata indicating a quality of the training data at the respective clinical site” these limitations are an additional element that amounts to adding insignificant extra-solution activity to the judicial exception. See MPEP § 2106.05(g). Furthermore the additional element is directed to receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d. Therefore, the claim is ineligible.
Regarding claim 2, the rejection of claim 1 is incorporated and further:
Step 2A Prong 1: The claim recites, in part:
“determining a parameter for the global model according to:
a Global Parameter =
(
α
1
*
w
1
+
α
2
*
w
3
+
α
3
*
w
3
+
…
+
α
N
*
w
N
)
/
(
α
1
+
α
2
+
α
3
+
…
+
α
N
)
;
WN comprises the local update to the parameter in the global model as determined by the nth clinical site, and
α
N
comprises a real number in the range
0
≤
α
N
≤
1
; and
wherein the value of
α
N
is determined from the metadata associated with the update to the parameter in the global model determined by the nth clinical site.” This limitation is a mathematical concept.
Step 2A Prong 2: The claim does not recite any additional limitations, thus does not further recite any additional elements that integrates the judicial exception into a practical application or amount to significantly more.
Regarding claim 3, the rejection of claim 1 is incorporated and further:
Step 2A Prong 1: The claim recites, in part:
“the metadata provides an indication of a performance of the respective local copy of the global model after the training, for one or more subsets of training data at the respective clinical site having a common characteristic that is expected to influence model error.” This encompasses the mental determination of a performance of an observed local copy of a model after training.
Step 2A Prong 2: The claim does not recite any additional limitations, thus does not further recite any additional elements that integrates the judicial exception into a practical application or amount to significantly more.
Regarding claim 4, the rejection of claim 3 is incorporated and further:
Step 2A Prong 1: a continuation of the abstract idea identified in the parent claim.
Step 2A Prong 2: The judicial exception is not integrated into a practical application; the remaining limitations of the claim are as follows: “the medical data comprises computed tomography, CT, scans; and wherein the metadata comprises an indication of the performance of the local copy of the global model when classifying CT scans of different radiation dosage” the limitation is an additional element that generally links the use of the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(h).
Step 2B: The additional elements, taken individually and in combination, do not provide an inventive concept of significantly more than the abstract idea itself for the reasons set forth in step 2A prong 2 above. Therefore, the claim is ineligible.
Regarding claim 5, the rejection of claim 3 is incorporated and further:
Step 2A Prong 1: The claim recites, in part:
“wherein the metadata comprises an indication of the performance of the global model when segmenting full images of the anatomical feature and/or partial images of the anatomical feature” a continuation of the abstract idea identified in the parent claim.
Step 2A Prong 2: The judicial exception is not integrated into a practical application; the remaining limitations of the claim are as follows: “segmenting the medical image to obtain a segmentation of an anatomical feature in the medical imaging data” the limitation is an additional element that amounts to adding insignificant extra-solution activity to the judicial exception. See MPEP § 2106.05(g). “the medical data comprises a medical image” the limitation is an additional element that generally links the use of the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(h). “the global model is for use in” the limitation is an additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP § 2106.05(f)(2).
Step 2B: The additional elements “the medical data comprises a medical image”, “the global model is for use in”, taken individually and in combination, do not provide an inventive concept of significantly more than the abstract idea itself for the reasons set forth in step 2A prong 2 above. Further, “segmenting the medical image to obtain a segmentation of an anatomical feature in the medical imaging data” the limitation is an additional element that amounts to adding insignificant extra-solution activity to the judicial exception. Furthermore the additional element is well‐understood, routine, and conventional as taught by activity supported under Berkheimer. Sheller et al. (“Multi-Institutional Deep Learning Modeling Without Sharing Patient Data: A Feasibility Study on Brain Tumor Segmentation”, Sheller et al., 23 June 2019) (hereinafter “Sheller”), teaches that segmentation of medical images using various topologies is well‐understood, routine, and conventional. Sheller, page 4, section 2.4, ¶1 “Since its introduction in 2015, U-Net has quickly become one of the standard deep learning topologies for image segmentation and has been instrumental in creating prediction models for segmenting nerves in ultrasound images, lungs in CT scans, and even interference in radio telescopes.”. See MPEP § 2106.05(d). Therefore, the claim is ineligible.
Regarding claim 8, the rejection of claim 1 is incorporated and further:
Step 2A Prong 1: The claim recites, in part:
“determining, for a test medical image, a first region of the test image used… to perform the task on the test medical image” this encompasses the mental determination of a first region of an observed medical image used to perform as task.
“determining, for the test medical image, a second region of the test image used… to perform the task on the test medical image” this encompasses the mental determination of a second region of an observed medical image used to perform as task.
“comparing the first region of the test image to the second region of the test image to determine a measure of model drift” this encompasses the mental comparison of a first and second region of an observed image to determine a model drift.
Step 2A Prong 2: The judicial exception is not integrated into a practical application; the remaining limitations of the claim are as follows: “by the global model”, “by the updated global model” these limitations are an additional element that generally links the use of the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(h).
Step 2B: The additional elements, taken individually and in combination, do not provide an inventive concept of significantly more than the abstract idea itself for the reasons set forth in step 2A prong 2 above. Therefore, the claim is ineligible.
Regarding claim 9, the rejection of claim 1 is incorporated and further:
Step 2A Prong 1: a continuation of the abstract idea identified in the parent claim.
Step 2A Prong 2: The judicial exception is not integrated into a practical application; the remaining limitations of the claim are as follows: “Repeating the steps of...updating, for a subset of the training data at each respective clinical site that was classified by the global model with a certainty below a threshold certainty level” the limitation is an additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP § 2106.05(f)(2). “sending, receiving” the limitation is an additional element that amounts to adding insignificant extra-solution activity to the judicial exception. See MPEP § 2106.05(g).
Step 2B: The additional elements, taken individually and in combination, do not provide an inventive concept of significantly more than the abstract idea itself for the reasons set forth in step 2A prong 2 above. Further, “sending, receiving” the limitation is an additional element that amounts to adding insignificant extra-solution activity to the judicial exception. See MPEP § 2106.05(g). Furthermore the additional element is directed to receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d. Therefore, the claim is ineligible.
Regarding claim 11, the rejection of claim 1 is incorporated and further:
Step 2A Prong 1: a continuation of the abstract idea identified in the parent claim.
Step 2A Prong 2: The judicial exception is not integrated into a practical application; the remaining limitations of the claim are as follows: “the global model comprises a neural network model and the parameter comprises a weight or a bias in the neural network model” the limitation is an additional element that generally links the use of the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(h).
Step 2B: The additional elements, taken individually and in combination, do not provide an inventive concept of significantly more than the abstract idea itself for the reasons set forth in step 2A prong 2 above. Therefore, the claim is ineligible.
Regarding claim 12:
Step 1: Claim 1 is directed to [a]n apparatus, therefore it falls under the statuary category of a machine.
Step 2A Prong 1: The claim recites, in part:
“the quality of the training data comprising a determination of a balance of the training data at the clinical site regarding different output classifications of the global model” this encompasses the mental determination of a balance of observed training data.
“update the parameter in the global model, based on the received local updates to the parameter and the received metadata by combining the local updates to the parameter to determine an update to the global model by weighting each local update according to the respective metadata such that local updates resulting from more balanced training data are given more weight than local updates resulting from less balanced training data.” This encompasses the mental update of an observed parameter based on observed local updates and weighting the update based on observed metadata. Further, this limitation is a mathematical concept.
Step 2A Prong 2: The judicial exception is not integrated into a practical application; the remaining limitations of the claim are as follows: “a memory comprising instruction data representing a set of instructions; and a processor configured to communicate with the memory and to execute the set of instructions” the limitation is an additional element that generally links the use of the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(h). “the set of instructions, when executed by the processor, cause the processor to”, “train the respective local copy of the global model on training data at the respective clinical site”, “obtained by training the local copy of the global model on the training data at the respective clinical site” the limitations are an additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP § 2106.05(f)(2). “send information to the plurality of clinical sites to enable each of the plurality of clinical sites to create a local copy of the global model”, “receive, from each of the plurality of clinical sites, i) a local update to a parameter in the global model…and ii) metadata indicating a quality of the training data at the respective clinical site” these limitations are an additional element that amounts to adding insignificant extra-solution activity to the judicial exception. See MPEP § 2106.05(g).
Step 2B: The additional elements “a memory comprising instruction data representing a set of instructions; and a processor configured to communicate with the memory and to execute the set of instructions”, “the set of instructions, when executed by the processor, cause the processor to”, “train the respective local copy of the global model on training data at the respective clinical site”, “obtained by training the local copy of the global model on the training data at the respective clinical site”, taken individually and in combination, do not provide an inventive concept of significantly more than the abstract idea itself for the reasons set forth in step 2A prong 2 above. Further, “send information to the plurality of clinical sites to enable each of the plurality of clinical sites to create a local copy of the global model”, “receive, from each of the plurality of clinical sites, i) a local update to a parameter in the global model…and ii) metadata indicating a quality of the training data at the respective clinical site” these limitations are an additional element that amounts to adding insignificant extra-solution activity to the judicial exception. See MPEP § 2106.05(g). Furthermore the additional element is directed to receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d. Therefore, the claim is ineligible.
Regarding claim 13:
Step 1: Claim 1 is directed to [a] non-transitory computer readable medium, therefore it falls under the statuary category of a manufacture.
Step 2A Prong 1: The claim recites, in part:
“the quality of the training data comprising a determination of a balance of the training data at the clinical site regarding different output classifications of the global model” this encompasses the mental determination of a balance of observed training data.
“update the parameter in the global model, based on the received local updates to the parameter and the received metadata by combining the local updates to the parameter to determine an update to the global model by weighting each local update according to the respective metadata such that local updates resulting from more balanced training data are given more weight than local updates resulting from less balanced training data.” This encompasses the mental update of an observed parameter based on observed local updates and weighting the update based on observed metadata. Further, this limitation is a mathematical concept.
Step 2A Prong 2: The judicial exception is not integrated into a practical application; the remaining limitations of the claim are as follows: “execution by a suitable computer or processor, causes the computer or processor to”, “train the respective local copy of the global model on training data at the respective clinical site”, “obtained by training the local copy of the global model on the training data at the respective clinical site” the limitations are an additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP § 2106.05(f)(2). “send information to the plurality of clinical sites to enable each of the plurality of clinical sites to create a local copy of the global model”, “receive, from each of the plurality of clinical sites, i) a local update to a parameter in the global model…and ii) metadata indicating a quality of the training data at the respective clinical site” these limitations are an additional element that amounts to adding insignificant extra-solution activity to the judicial exception. See MPEP § 2106.05(g).
Step 2B: The additional elements “execution by a suitable computer or processor, causes the computer or processor to”, “train the respective local copy of the global model on training data at the respective clinical site”, “obtained by training the local copy of the global model on the training data at the respective clinical site”, taken individually and in combination, do not provide an inventive concept of significantly more than the abstract idea itself for the reasons set forth in step 2A prong 2 above. Further, “send information to the plurality of clinical sites to enable each of the plurality of clinical sites to create a local copy of the global model”, “receive, from each of the plurality of clinical sites, i) a local update to a parameter in the global model…and ii) metadata indicating a quality of the training data at the respective clinical site” these limitations are an additional element that amounts to adding insignificant extra-solution activity to the judicial exception. See MPEP § 2106.05(g). Furthermore the additional element is directed to receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d. Therefore, the claim is ineligible.
Claim Rejections - 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.
Claims 1, 3, 9 and 11-13 are rejected under 35 U.S.C. § 103 as being unpatentable over Dirk (EP3528179A1) in view of Xie et al. (“Multi-Center Federated Learning”, Xie et al., 3 May 2020) (hereinafter “Xie”) in further view of Duan et al. (“Astraea: Self-Balancing Federated Learning for Improving Classification Accuracy of Mobile Deep Learning Applications”, Duan et al., 10 February 2020) (hereinafter “Duan”).
Regarding claim 1:
Dirk teaches [a] computer implemented method of training a model to perform a task on medical data using a distributed machine learning process (Dirk, claim 2 “A method as in claim 1 wherein the first neural network and the first set of medical data are located remotely to the central processor;”) whereby a global model is updated based on training performed on local copies of the global model (Dirk, ¶6 “The values from the first and second neural networks may be combined and the combined values sent back to the first neural network to enable the first neural network to update its weights based on the combined learning experiences of both the first and second neural networks. Essentially therefore, the central processor acts to collate weight values and thus learning insights from a plurality of local instances of a neural network, without the medical data used to train the local neural networks needing to be transferred or combined into a single database” further, Dirk, ¶23 “Such a copy of the neural network model, accessible to the central processor, is referred to herein as a central neural network model. The central processor may update the parameters of the central neural network model with any updates to the weights. In this way, the central neural network model may comprise a central copy of the current “best” weights from the combining the weights of the first, second (and any subsequent) neural network models.” the central neural network model can be considered the a global model in light of the specification, pages 1-2, lines 32-1 “The central server holds a “global” or central copy of the model and may send 114 information about the global model, e.g. such as parameters enabling a local copy of the model to be created, to each clinical site.” ) at a plurality of clinical sites (Dirk, ¶49-50 “In this embodiment, each neural network 304, 306 and 308 is associated with one or more imaging devices at different clinical sites.”), wherein the global model is for use in predicting a classification for the medical data, wherein the medical data comprises a medical image (Dirk, ¶41 “In some embodiments, the step 202 of training the first neural network on a first set of medical data comprises training the first neural network to classify or locate an object in a medical image.”), the method comprising:
sending information to the plurality of clinical sites to enable each of the plurality of clinical sites to create a local copy of the global model and train the respective local copy of the global model on training data at the respective clinical site (Dirk, ¶49-50 “In this embodiment, each neural network 304, 306 and 308 is associated with one or more imaging devices at different clinical sites. The imaging devices update a local image database that contains clinical images and meta-data (e.g. annotations/classifications provided by a human) collected on the individual medical images and this data forms the respective sets of medical data 310, 312 and 314.
At the beginning of the learning process, an initial copy of the same neural network model (e.g. a copy of the central model described above with respect to method 100), including model parameters p (e.g. the weights for the model) is sent from the central parameter server 302 to the three local instances.”);
receiving, from each of the plurality of clinical sites, i) a local update to a parameter in the global model obtained by training the local copy of the global model on the training data at the respective clinical site (Dirk, ¶52 “Once trained, the first, second and third neural networks 304, 306, 308 then each send a first parameter p indicative of a respective weight in each of the neural networks to the central parameter server 302, according to the step 204 of the method 200 as described above with respect to Figure 2.”)
Dirk does not teach “updating the parameter in the global model, based on the received local updates to the parameter and the received metadata by combining the local updates to the parameter to determine an update to the global model by weighting each local update according to the respective metadata such that local updates resulting from more balanced training data are given more weight than local updates resulting from less balanced training data”
However, Xie teaches updating the parameter in the global model, based on the received local updates to the parameter and the received metadata by combining the local updates to the parameter to determine an update to the global model by weighting each local update according to the respective metadata such that local updates resulting from more balanced training data are given more weight than local updates resulting from less balanced training data (Xie, page 1-2, section 1, ¶2 “An aggregation algorithm is responsible for averaging the many local models’ parameters, weighted by the size of training data on each device. Compared with conventional distributed machine learning, federated Learning is robust against unbalanced and non-IID (Independent and Identically Distributed) data distributions, which is the defining characteristic of modern AI products for mobile devices.”).
Dirk and Xie are analogous art because both references concern methods for federated learning. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dirk’s federated learning method to incorporate the weighting taught by Xie. The motivation for doing so would have been to tackle non-IID (unbalanced) data in federated learning as stated in Xie, page 13, conclusion, ¶1 “In this work, we propose a novel multi-center federated learning framework to tackle the non-IID challenge of the federated setting. This proposed method can efficiently capture the multiple hidden distributions of numerous devices or users.”.
Dirk in view of Xie does not teach “ii) metadata indicating a quality of the training data at the respective clinical site, the quality of the training data comprising a determination of a balance of the training data at the clinical site regarding different output classifications of the global model”
However, Duan teaches ii) metadata indicating a quality of the training data at the respective clinical site, the quality of the training data comprising a determination of a balance of the training data (Duan, page 6, col 1, Algorithm 3, line 7 “k ←arg mini DKL(Pm + Pi||Pu), i ∈Sclient” further, “To measure the extent of partial equilibrium, we using Kullback-Leibler divergence between Pm + Pk and Pu, where Pm, Pk, Pu means the probability distributions of mediator, rescheduling client, and uniform distribution, respectively.”) at the clinical site regarding different output classifications of the global model (Duan, page 5, col 2, Algorithm 2, line 2 “Calculate the data size of each class C1,…, CN, and the mean
C
-
.”)
Dirk in view of Xie and Duan are analogous art because both references concern methods for federated learning. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dirk/Xie’s federated learning method to incorporate the balance metadata taught by Duan. The motivation for doing so would have been to tackle non-IID (unbalanced) data in federated learning as stated in Duan, page 11, col 1, Abstract, ¶1 “In this paper, we demonstrate that the imbalanced distributed training data will cause accuracy degradation in FL. To counter this problem, we build a self-balancing federated learning framework call Astraea, which alleviates the imbalances”.
Regarding claim 3:
Dirk in view of Xie in further view of Duan teaches [t]he method of claim 1, wherein the metadata provides an indication of a performance of the respective local copy of the global model after the training, for one or more subsets of training data at the respective clinical site (Dirk, ¶51 “Each of the first, second and third neural networks 304, 306, 308 are then trained on first, second and third sets of medical data 310, 312 and 314 respectively, according to the step 202 of the method 200 shown in Figure 2, using the locally available training data as input and the meta-data as reference output.”) having a common characteristic that is expected to influence model error (Xie, page 1-2, section 1, ¶2 “An aggregation algorithm is responsible for averaging the many local models’ parameters, weighted by the size of training data on each device. Compared with conventional distributed machine learning, federated Learning is robust against unbalanced and non-IID (Independent and Identically Distributed) data distributions, which is the defining characteristic of modern AI products for mobile devices.” Here, the size of the training data can be considered a common characteristic that is expected to influence model error).
It would have been obvious to combine the teachings of Dirk and Xie for the reasons set forth in connection with claim 1 above.
Regarding claim 9:
Dirk in view of Xie in further view of Duan teaches [t]he method of claim 1 further comprising: Repeating the steps of sending, receiving, and updating, for a subset of the training data at each respective clinical site that was classified by the global model with a certainty below a threshold certainty level (Dirk, ¶34 “In some embodiments the method further comprises repeating the steps of receiving 102, combining 104 and sending 106 until a quality criterion associated with the first neural network is satisfied. For example, these steps may be repeated until the first neural network reaches a threshold accuracy. The accuracy may be determined, for example, by determining the percentage of correct classifications made by the first neural network on a quality control set of example medical data (e.g. a set of unseen medical data used specifically for quality control purposes).”).
Regarding claim 11:
Dirk in view of Xie in further view of Duan teaches [t]he method of claim 1, wherein the global model comprises a neural network model and the parameter comprises a weight or a bias in the neural network model (Dirk, ¶5 “The method comprises receiving a first parameter indicative of a weight in the first neural network, the first parameter being obtained during training of the first neural network on a first set of medical data.” It is noted the claim recites alternative language, and Dirk teaches at least one of the alternatives.).
Regarding claim 12:
Dirk teaches [a]n apparatus for training a model to perform a task on medical data using a distributed machine learning process (Dirk, claim 2 “A method as in claim 1 wherein the first neural network and the first set of medical data are located remotely to the central processor;”) whereby a global model is updated based on training performed on local copies of the global model (Dirk, ¶6 “The values from the first and second neural networks may be combined and the combined values sent back to the first neural network to enable the first neural network to update its weights based on the combined learning experiences of both the first and second neural networks. Essentially therefore, the central processor acts to collate weight values and thus learning insights from a plurality of local instances of a neural network, without the medical data used to train the local neural networks needing to be transferred or combined into a single database” further, Dirk, ¶23 “Such a copy of the neural network model, accessible to the central processor, is referred to herein as a central neural network model. The central processor may update the parameters of the central neural network model with any updates to the weights. In this way, the central neural network model may comprise a central copy of the current “best” weights from the combining the weights of the first, second (and any subsequent) neural network models.” the central neural network model can be considered the a global model in light of the specification, pages 1-2, lines 32-1 “The central server holds a “global” or central copy of the model and may send 114 information about the global model, e.g. such as parameters enabling a local copy of the model to be created, to each clinical site.” ) at a plurality of clinical sites (Dirk, ¶49-50 “In this embodiment, each neural network 304, 306 and 308 is associated with one or more imaging devices at different clinical sites.”)) at a plurality of clinical sites (Dirk, ¶49-50 “In this embodiment, each neural network 304, 306 and 308 is associated with one or more imaging devices at different clinical sites.”), wherein the global model is for use in predicting a classification for the medical data, wherein the medical data comprises a medical image (Dirk, ¶41 “In some embodiments, the step 202 of training the first neural network on a first set of medical data comprises training the first neural network to classify or locate an object in a medical image.”), the apparatus comprising:
a memory comprising instruction data representing a set of instructions; and a processor configured to communicate with the memory and to execute the set of instructions, wherein the set of instructions, when executed by the processor, cause the processor to (Dirk, claim 13 “A system (400) configured for training a first neural network to classify medical data, the system comprising:
a memory (404) comprising instruction data representing a set of instructions;
a processor (402) configured to communicate with the memory and to execute the set of instructions, wherein the set of instructions, when executed by the processor, cause the processor to”):
send information to the plurality of clinical sites to enable each of the plurality of clinical sites to create a local copy of the global model (Dirk, ¶23 “Such a copy of the neural network model, accessible to the central processor, is referred to herein as a central neural network model. The central processor may update the parameters of the central neural network model with any updates to the weights. In this way, the central neural network model may comprise a central copy of the current “best” weights from the combining the weights of the first, second (and any subsequent) neural network models.” the central neural network model can be considered the a global model in light of the specification, pages 1-2, lines 32-1 “The central server holds a “global” or central copy of the model and may send 114 information about the global model, e.g. such as parameters enabling a local copy of the model to be created, to each clinical site.” ) at a plurality of clinical sites (Dirk, ¶49-50 “In this embodiment, each neural network 304, 306 and 308 is associated with one or more imaging devices at different clinical sites.”) and train the respective local copy of the global model on training data at the respective clinical site (Dirk, ¶49-50 “In this embodiment, each neural network 304, 306 and 308 is associated with one or more imaging devices at different clinical sites. The imaging devices update a local image database that contains clinical images and meta-data (e.g. annotations/classifications provided by a human) collected on the individual medical images and this data forms the respective sets of medical data 310, 312 and 314.
At the beginning of the learning process, an initial copy of the same neural network model (e.g. a copy of the central model described above with respect to method 100), including model parameters p (e.g. the weights for the model) is sent from the central parameter server 302 to the three local instances.”);
receive, from each of the plurality of clinical sites, i) a local update to a parameter in the global model obtained by training the local copy of the global model on the training data at the respective clinical site (Dirk, ¶52 “Once trained, the first, second and third neural networks 304, 306, 308 then each send a first parameter p indicative of a respective weight in each of the neural networks to the central parameter server 302, according to the step 204 of the method 200 as described above with respect to Figure 2.”)
Dirk does not teach “update the parameter in the global model, based on the received local updates to the parameter and the received metadata by combining the local updates to the parameter to determine an update to the global model by weighting each local update according to the respective metadata such that local updates resulting from more balanced training data are given more weight than local updates resulting from less balanced training data.”
update the parameter in the global model, based on the received local updates to the parameter and the received metadata by combining the local updates to the parameter to determine an update to the global model by weighting each local update according to the respective metadata such that local updates resulting from more balanced training data are given more weight than local updates resulting from less balanced training data (Xie, page 1-2, section 1, ¶2 “An aggregation algorithm is responsible for averaging the many local models’ parameters, weighted by the size of training data on each device. Compared with conventional distributed machine learning, federated Learning is robust against unbalanced and non-IID (Independent and Identically Distributed) data distributions, which is the defining characteristic of modern AI products for mobile devices.”).
Dirk and Xie are analogous art because both references concern methods for federated learning. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dirk’s federated learning method to incorporate the balance weighting taught by Xie. The motivation for doing so would have been to tackle non-IID (unbalanced) data in federated learning as stated in Xie, page 13, conclusion, ¶1 “In this work, we propose a novel multi-center federated learning framework to tackle the non-IID challenge of the federated setting. This proposed method can efficiently capture the multiple hidden distributions of numerous devices or users.”.
Dirk in view of Xie does not teach “ii) metadata indicating a quality of the training data at the respective clinical site, the quality of the training data comprising a determination of a balance of the training data at the clinical site regarding different output classifications of the global model”
However, Duan teaches ii) metadata indicating a quality of the training data at the respective clinical site, the quality of the training data comprising a determination of a balance of the training data (Duan, page 6, col 1, Algorithm 3, line 7 “k ←arg mini DKL(Pm + Pi||Pu), i ∈Sclient” further, “To measure the extent of partial equilibrium, we using Kullback-Leibler divergence between Pm + Pk and Pu, where Pm, Pk, Pu means the probability distributions of mediator, rescheduling client, and uniform distribution, respectively.”) at the clinical site regarding different output classifications of the global model (Duan, page 5, col 2, Algorithm 2, line 2 “Calculate the data size of each class C1,…, CN, and the mean
C
-
.”)
Dirk in view of Xie and Duan are analogous art because both references concern methods for federated learning. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dirk/Xie’s federated learning method to incorporate the balance metadata taught by Duan. The motivation for doing so would have been to tackle non-IID (unbalanced) data in federated learning as stated in Duan, page 11, col 1, Abstract, ¶1 “In this paper, we demonstrate that the imbalanced distributed training data will cause accuracy degradation in FL. To counter this problem, we build a self-balancing federated learning framework call Astraea, which alleviates the imbalances”.
Regarding claim 13:
Dirk teaches [a] non-transitory computer readable medium, storing computer readable code that, on execution by a suitable computer or processor, causes the computer or processor to (Dirk, claim 15 “A computer program product comprising a non-transitory computer readable medium, the computer readable medium having computer readable code embodied therein, the computer readable code being configured such that, on execution by a suitable computer or processor, the computer or processor is caused to perform the method of any one of claims 1 to 12.”):
send information to the plurality of clinical sites to enable each of the plurality of clinical sites to create a local copy of the global model and train the respective local copy of the global model on training data at the respective clinical site (Dirk, ¶49-50 “In this embodiment, each neural network 304, 306 and 308 is associated with one or more imaging devices at different clinical sites. The imaging devices update a local image database that contains clinical images and meta-data (e.g. annotations/classifications provided by a human) collected on the individual medical images and this data forms the respective sets of medical data 310, 312 and 314.
At the beginning of the learning process, an initial copy of the same neural network model (e.g. a copy of the central model described above with respect to method 100), including model parameters p (e.g. the weights for the model) is sent from the central parameter server 302 to the three local instances.”);
receive, from each of the plurality of clinical sites, i) a local update to a parameter in the global model obtained by training the local copy of the global model on the training data at the respective clinical site (Dirk, ¶52 “Once trained, the first, second and third neural networks 304, 306, 308 then each send a first parameter p indicative of a respective weight in each of the neural networks to the central parameter server 302, according to the step 204 of the method 200 as described above with respect to Figure 2.”)
Dirk does not teach “update the parameter in the global model, based on the received local updates to the parameter and the received metadata by combining the local updates to the parameter to determine an update to the global model by weighting each local update according to the respective metadata such that local updates resulting from more balanced training data are given more weight than local updates resulting from less balanced training data.”
However, Xie teaches update the parameter in the global model, based on the received local updates to the parameter and the received metadata by combining the local updates to the parameter to determine an update to the global model by weighting each local update according to the respective metadata such that local updates resulting from more balanced training data are given more weight than local updates resulting from less balanced training data (Xie, page 1-2, section 1, ¶2 “An aggregation algorithm is responsible for averaging the many local models’ parameters, weighted by the size of training data on each device. Compared with conventional distributed machine learning, federated Learning is robust against unbalanced and non-IID (Independent and Identically Distributed) data distributions, which is the defining characteristic of modern AI products for mobile devices.”).
Dirk and Xie are analogous art because both references concern methods for federated learning. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dirk’s federated learning method to incorporate the balance weighting taught by Xie. The motivation for doing so would have been to tackle non-IID (unbalanced) data in federated learning as stated in Xie, page 13, conclusion, ¶1 “In this work, we propose a novel multi-center federated learning framework to tackle the non-IID challenge of the federated setting. This proposed method can efficiently capture the multiple hidden distributions of numerous devices or users.”.
And does not teach “ii) metadata indicating a quality of the training data at the respective clinical site, the quality of the training data comprising a determination of a balance of the training data at the clinical site regarding different output classifications of the global model”
However, Duan teaches ii) metadata indicating a quality of the training data at the respective clinical site, the quality of the training data comprising a determination of a balance of the training data (Duan, page 6, col 1, Algorithm 3, line 7 “k ←arg mini DKL(Pm + Pi||Pu), i ∈Sclient” further, “To measure the extent of partial equilibrium, we using Kullback-Leibler divergence between Pm + Pk and Pu, where Pm, Pk, Pu means the probability distributions of mediator, rescheduling client, and uniform distribution, respectively.”) at the clinical site regarding different output classifications of the global model (Duan, page 5, col 2, Algorithm 2, line 2 “Calculate the data size of each class C1,…, CN, and the mean
C
-
.”)
Dirk in view of Xie Duan are analogous art because both references concern methods for federated learning. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dirk/Xie’s federated learning method to incorporate the balance metadata taught by Duan. The motivation for doing so would have been to tackle non-IID (unbalanced) data in federated learning as stated in Duan, page 11, col 1, Abstract, ¶1 “In this paper, we demonstrate that the imbalanced distributed training data will cause accuracy degradation in FL. To counter this problem, we build a self-balancing federated learning framework call Astraea, which alleviates the imbalances”.
Claims 2 and 5 are rejected under 35 U.S.C. § 103 as being unpatentable over Dirk in view of Xie in view of Duan in further view of Roy et al. (“BrainTorrent: A Peer-to-Peer Environment for Decentralized Federated Learning”, Roy et al., 16 May 2019) (hereinafter “Roy”).
Regarding claim 2:
Dirk in view of Xie in further view of Duan teaches [t]he method of claim 1
However Dirk in view of Xie in further view of Duan does not teach “wherein the step of combining the local updates to the parameter to determine the update to the global model comprises: determining a parameter for the global model according to:
a Global Parameter =
(
α
1
*
w
1
+
α
2
*
w
3
+
α
3
*
w
3
+
…
+
α
N
*
w
N
)
/
(
α
1
+
α
2
+
α
3
+
…
+
α
N
)
;
WN comprises the local update to the parameter in the global model as determined by the nth clinical site, and
α
N
comprises a real number in the range
0
≤
α
N
≤
1
; and
wherein the value of
α
N
is determined from the metadata associated with the update to the parameter in the global model determined by the nth clinical site”
However, Roy teaches wherein the step of combining the local updates to the parameter to determine the update to the global model comprises: determining a parameter for the global model according to:
a Global Parameter =
(
α
1
*
w
1
+
α
2
*
w
3
+
α
3
*
w
3
+
…
+
α
N
*
w
N
)
/
(
α
1
+
α
2
+
α
3
+
…
+
α
N
)
(Roy, page 3, section 2.1, ¶1 “Next, all the clients send these partially trained parameters to the central server S, which aggregates them by weighted averaging
w
s
=
Σ
i
a
i
a
w
ⅈ
” and further Roy, page 3, section 2 “In a general setting, data from all centers are pooled together
a
=
Σ
i
a
i
at a common server, where a common model is learned that is distributed across all the centers for usage” Here, the weighted averaging equation is equivalent to the global parameter determination equation as the numerator is a sum of (a-i * wi) and the denominator is a sum of a-i);
WN comprises the local update to the parameter in the global model (Roy, page 3, section 2.1, ¶1 “Let the weight parameter list for all clients be indicated by {W1 , . . . ,WN}.”) as determined by the nth clinical site (Roy, page 2, ¶3 “The design is motivated to fulfill the above mentioned requirements for a group of medical centers to collaborate.”), and
α
N
comprises a real number in the range
0
≤
α
N
≤
1
; and
wherein the value of
α
N
is determined from the metadata associated with the update to the parameter in the global model determined by the nth clinical site (Roy, page 3, section 2.1 “The multiplicative factor is computed as the fraction of the total data belonging to a client. The rationale is to emphasize clients with more training data.” Here, the multiplicative factor
a
i
a
can be considered
α
N
and
a
i
a
w
i
l
l
a
l
w
a
y
s
r
e
s
u
l
t
i
n
a
v
a
l
u
e
[
0,1
]
).
Dirk in view of Xie in further view of Duan and Roy are analogous art because both references concern methods for federated learning in medical environments. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dirk/Xie/Duan’s federated learning system to incorporate the weighted averaging taught by Roy. The motivation for doing so would have been to better perform irrespective of the number of clients. As stated in Roy, page 6, ¶2 “We observe that irrespective of the number of clients, BrainTorrent outperforms FLS for both, average Dice score over clients and Dice score for aggregated model”.
Regarding claim 5:
Dirk in view of Xie in further view of Duan teaches [t]he method of claim 3
Dirk in view of Xie in further view of Duan does not teach “wherein the medical data comprises a medical image and the global model is for use in segmenting the medical image to obtain a segmentation of an anatomical feature in the medical imaging data; and
wherein the metadata comprises an indication of the performance of the global model when segmenting full images of the anatomical feature and/or partial images of the anatomical feature.”
However, Roy teaches wherein the medical data comprises a medical image and the global model is for use in segmenting the medical image to obtain a segmentation of an anatomical feature in the medical imaging data (Roy, page 3, ¶1 “In contrast to that, we introduce a new peer-to-peer FL approach, and tackle the more challenging task of whole-brain segmentation with 20 classes with severe class-imbalance.” Here, the 20 classes can be considered anatomical features); and
wherein the metadata comprises an indication of the performance of the global model when segmenting full images of the anatomical feature and/or partial images of the anatomical feature (Roy, page 7, ¶1 “We take a closer look at the segmentation performance of the client-specific models.” And further, Roy, page 7, ¶1 “Also, as lower bound analysis, we train client-specific models with only 2 scans, referred to as ‘only client’ models and report their performance on the same validation set.”).
Dirk in view of Xie in further view of Duan and Roy are analogous art because both references concern methods for federated learning in medical environments. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dirk/Xie/Duan’s federated learning system to incorporate the weighted averaging taught by Roy. The motivation for doing so would have been to incorporate segmentation with a large number of classes with severe class-imbalance. As stated in Roy, page 3, ¶1 “In contrast to that, we introduce a new peer-to-peer FL approach, and tackle the more challenging task of whole-brain segmentation with 20 classes with severe class-imbalance.”.
Claim 4 is rejected under 35 U.S.C. § 103 as being unpatentable over Dirk in view of Xie in view of Duan in further view of Yang et al. (“Low-Dose CT Image Denoising Using a Generative Adversarial Network With Wasserstein Distance and Perceptual Loss”, Yang et al., June 2018) (hereinafter “Yang”).
Regarding claim 4:
Dirk in view of Xie in further view of Duan teaches [t]he method of claim 3, wherein the medical data comprises computed tomography, CT, scans (Dirk, ¶21 “Examples of medical data include medical images (e.g.+ x-ray images, ultrasound images, CT images, MR images or optical images of the exterior of the patient etc.), medical records, test results, acquisition parameters, manually entered user data, anthropometric patient data (e.g. patient size, weight), or data from additional sensors (e.g. optical 3D range cameras), and/or any other types of medical data or medical notes.”);
Dirk in view of Xie in further view of Duan does not teach “wherein the metadata comprises an indication of the performance of the local copy of the global model when classifying CT scans of different radiation dosage.”
However, Yang teaches wherein the metadata comprises an indication of the performance of the local copy of the global model when classifying CT scans of different radiation dosage (Yang, page 8, col 2, section F, ¶1 “Since VGG network is trained on natural images, it may cause concerns on how well it performs on CT image feature extraction. Thus, we displayed two feature maps of normal dose and quarter dose images and their absolute difference in Fig. 9. The feature map contains 512 small images of size 32 × 32. We organize these small images into a 32 × 16 array. Each small image emphasizes a feature of the original CT image, i.e. boundaries, edges, or whole structures.”).
Dirk in view of Xie in further view of Duan and Yang are analogous art because both references concern methods for neural networks operating on CT images. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dirk/Xie/Duan’s federated learning system to incorporate the performance of various doses taught by Yang. The motivation for doing so would have been to solve over-fitting issues and generate images with reduced noise as stated in Yang, page 9, col 1, ¶3 “Our experiment results with real clinical images have shown that the proposed WGAN-VGG network can effectively solve the well-known over-smoothing problem and generate images with reduced noise and increased contrast for improved lesion detection.”.
Claim 8 is rejected under 35 U.S.C. § 103 as being unpatentable over Dirk in view of Xie in view of Duan in further view of Adebayo et al. (“Sanity Checks for Saliency Maps”, Adebayo et al., 28 Oct 2018) (hereinafter “Adebayo”) in further view of Karimireddy et al. (“SCAFFOLD: Stochastic Controlled Averaging for Federated Learning”, Karimireddy et al., 17 Feb 2020) (hereinafter “Karimireddy”).
Regarding claim 8:
Dirk in view of Xie in further view of Duan teaches [t]he method of claim 1, wherein the medical data comprises a medical image, the method further comprising:
Dirk in view of Xie in further view of Duan does not teach “preceding the steps of sending, receiving, and updating: determining, for a test medical image, a first region of the test image used by the global model to perform the task on the test medical image; and
following the steps of sending, receiving, and updating: determining, for the test medical image, a second region of the test image used by the updated global model to perform the task on the test medical image;”
However, Adebayo teaches preceding the steps of sending, receiving, and updating: determining, for a test medical image, a first region (Adebayo, page 15, figure 7 “C) Guided Backprop explanation of sample A for an Inception v4 model with completely random weights.” Here the explanation sample can be considered the first region) of the test image used by the global model to perform the task on the test medical image (Adebayo, page 15, figure 7 “A) a skeletal radiogram from the pediatric bone age challenge organized by the radiological society of north America (RSNA). Given several thousand radiographs, challenge participants are tasked with building models to predict the age (in months) of the patient.” The skeletal radiogram A) can be considered the first region of the test image); and
following the steps of sending, receiving, and updating: determining, for the test medical image, a second region of the test image used by the updated global model to perform the task on the test medical image (Adebayo, page 15, figure 7 “B) Guided Backprop explanation of sample A for an Inception v4 model trained on the radiograms” Here, the trained model can be considered an updated mode, and the explanation sample can be considered the first region);
Dirk in view of Xie in further view of Duan and Adebayo are analogous art because both references concern methods for neural network classification including with medical data. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dirk/Xie/Duan’s federated learning system to incorporate the saliency regions taught by Adebayo. The motivation for doing so would have been to serve as a sanity check. As stated in Adebayo, page 9, section 6, ¶1 “We envision these methods to serve as sanity checks in the design of new model explanations.”
Dirk in view of Xie in further view of Yang does not teach “comparing the first region of the test image to the second region of the test image to determine a measure of model drift.”
However, Karimireddy teaches comparing the first region of the test image to the second region of the test image to determine a measure of model drift (Karimireddy, page 1, col 2, ¶3 “As a solution, we propose a new Stochastic Controlled Averaging algorithm (SCAFFOLD) which tries to correct for this client-drift. Intuitively, SCAFFOLD estimates the update direction for the server model (c) and the update direction for each client ci. The difference (c − ci) is then an estimate of the client-drift which is used to correct the local update.”).
Dirk in view of Xie in further view of Adebayo and Karimireddy are analogous art because both references concern methods for federated learning. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dirk/Xie/Duan/Adebayo’s federated learning system to incorporate the model drift measure taught by Karimireddy. The motivation for doing so would have been to overcome heterogeneity and converg quickly, as stated in Karimireddy, page 1, col 2, ¶3 “The difference (c − ci) is then an estimate of the client-drift which is used to correct the local update. This strategy successfully overcomes heterogeneity and converges in significantly fewer rounds of communication.”
Response to Arguments
Applicant's arguments filed January 30th, 2026 (hereinafter “Remarks”) have been fully considered but they are not persuasive.
Applicant’s arguments regarding the 35 U.S.C. 112(b) rejections of the previous office action have been fully considered, and are persuasive. The rejections have been withdrawn due to claim amendments.
Applicant’s arguments regarding the 35 U.S.C. § 103 rejections have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Rejections under 35 U.S.C. § 101:
Argument 1:
“It is not practical to monitor for, and determine the balance of, training data regarding different output classifications of the global model for different sites in ones mind, a task made further impractical when considering the training data involves medical images.” (Remarks, page 2).
Examiners Response:
Examiner respectfully disagrees, the MPEP states “Nor do the courts distinguish between claims that recite mental processes performed by humans and claims that recite mental processes performed on a computer. As the Federal Circuit has explained, “[c]ourts have examined claims that required the use of a computer and still found that the underlying, patent-ineligible invention could be performed via pen and paper or in a person’s mind.” Versata Dev. Group v. SAP Am., Inc., 793 F.3d 1306, 1335, 115 USPQ2d 1681, 1702 (Fed. Cir. 2015). See also Intellectual Ventures I LLC v. Symantec Corp., 838 F.3d 1307, 1318, 120 USPQ2d 1353, 1360 (Fed. Cir. 2016) (‘‘[W]ith the exception of generic computer-implemented steps, there is nothing in the claims themselves that foreclose them from being performed by a human, mentally or with pen and paper.’’); Mortgage Grader, Inc. v. First Choice Loan Servs. Inc., 811 F.3d 1314, 1324, 117 USPQ2d 1693, 1699 (Fed. Cir. 2016) (holding that computer-implemented method for “anonymous loan shopping” was an abstract idea because it could be “performed by humans without a computer”). Mental processes recited in claims that require computers are explained further below with respect to point C.” See MPEP § 2106.04(a)(2)(III). A person could observe medical images and determine a balance of data regarding classifications. For instance a doctor could observe CT scans and, after determining the diseases present in each scan, conclude based on the total number of each disease observed that the data was balanced or imbalanced.
Argument 2:
“For instance, give the layers of a machine learning model, how does one update the parameters based on the expected large volume of data from multiple site in one's mind or pad and paper? It is simply not practical. Accordingly, Applicant submits that the claims are patent eligible under Step 2A, Prong One.” (Remarks, pages 2-3).
Examiners Response:
Examiner respectfully disagrees, the MPEP states ““claiming the improved speed or efficiency inherent with applying the abstract idea on a computer” does not integrate a judicial exception into a practical application or provide an inventive concept. Intellectual Ventures I LLC v. Capital One Bank (USA), 792 F.3d 1363, 1367, 115 USPQ2d 1636, 1639 (Fed. Cir. 2015).” See MPEP § 2106.5(f). Here, the updating of parameters can be considered the abstract ideas of mathematical calculations and a mental process because, as currently claimed, they do not require a large volume of data, and can therefore can be practically performed in the human mind. Therefore the claims are rejected under 35 U.S.C. § 101.
Argument 3:
“It is clear from above that the claimed embodiment circumvents barriers to data sharing across multiple sites, improves model quality by avoiding model drift, and improves model accuracy by lowering bias. The specification describes how these improvements are achieved, namely by “considering appropriate metadata while merging the weights”. In fact, the claims are not “merely claiming the idea of a solution or outcome”, but rather, describing “a particular solution to a problem or a particular way to achieve a desired outcome”.” (Remarks, page 4).
Examiners Response:
Examiner respectfully disagrees, the applicant merely uses a computer to perform processes which can be performed by a human mind. An improvement through considering appropriate metadata while merging the weights may be an improvement in an abstract idea, but not an improvement in computer capabilities or to improve an existing technology. A person could consider observed appropriate metadata while merging the weights in their head or using pen and paper. Further, improvements to data sharing across multiple sites is an additional element that amounts to adding insignificant extra-solution activity to the judicial exception. See MPEP § 2106.05(g). The courts have found the additional element to be directed to receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d. See MPEP § 2106.05(d)/(II).
Argument 4:
“In the Ex Parte Desjardins panel decision, referenced in the USPTO email of September 26, 2025, the panel opined as follows in finding in favor of patent eligibility…Similarly, the claims of the present application particularly describe the particular way of achieving the desired outcome of improving the machine learning model.” (Remarks, page 4).
Examiners Response:
Examiner respectfully disagrees, the MPEP states “An inventive concept “cannot be furnished by the unpatentable law of nature (or natural phenomenon or abstract idea) itself.” Genetic Techs. v. Merial LLC, 818 F.3d 1369, 1376, 118 USPQ2d 1541, 1546 (Fed. Cir. 2016).” See MPEP § 2106.05(I). here, unlike in Ex Parte Desjardins, the improvements are directed to an abstract idea and rely upon additional elements that amount to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP § 2106.05(f)(2). “As the Supreme Court explained in Alice Corp., mere physical or tangible implementation of an exception is not in itself an inventive concept and does not guarantee eligibility: “The fact that a computer “necessarily exist[s] in the physical, rather than purely conceptual, realm,” is beside the point. There is no dispute that a computer is a tangible system (in § 101 terms, a “machine”), or that many computer-implemented claims are formally addressed to patent-eligible subject matter. But if that were the end of the § 101 inquiry, an applicant could claim any principle of the physical or social sciences by reciting a computer system configured to implement the relevant concept. Such a result would make the determination of patent eligibility “depend simply on the draftsman’s art,” Flook, supra, at 593, 98 S. Ct. 2522, 57 L. Ed. 2d 451, thereby eviscerating the rule that “‘[l]aws of nature, natural phenomena, and abstract ideas are not patentable,’” Myriad, 133 S. Ct. 1289, 186 L. Ed. 2d 124, 133).” Alice Corp., 573 U.S. at 224, 110 USPQ2d at 1983-84 (alterations in original).” See MPEP § 2106.05(I)(A).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Xiao et al. (“An Experimental Study of Class Imbalance in Federated Learning”, Xiao et al., 28 Mar 2022) discloses two new metrics to define class imbalance– the global class imbalance degree (MID) and the local difference of class imbalance among clients (WCS).
Wang et al. (“Addressing Class Imbalance in Federated Learning”, Wang et al., 15 Dec 2020) discloses a monitoring scheme that can infer the composition of training data for each FL round, and design a new loss function — Ratio Loss to mitigate the impact of the imbalance.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 JACOB Z SUSSMAN MOSS whose telephone number is (571) 272-1579. The examiner can normally be reached Monday - Friday, 9 a.m. - 5 p.m. ET.
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, Kakali Chaki can be reached on (571) 272-3719. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
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.
/J.S.M./Examiner, Art Unit 2122
/KAKALI CHAKI/Supervisory Patent Examiner, Art Unit 2122
Prosecution Insights