DETAILED ACTION
This action is in response to the amendments and remarks filed 03/23/2026. Claims 1-4, 6-11, and 13-17 are pending and have been examined.
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 .
Priority
Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 02/03/2026 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Objections
Claim 17 is objected to because of the following informalities: “determining split points of the artificial intelligence model based respective products of the output data sizes and corresponding weight values” is improper grammar. Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 7 and 14 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 7 discloses the limitation “wherein, when a processing result of a split model that has not yet been completely processed by the first electronic apparatus is received from the second electronic apparatus while the plurality of split models are processed by the first electronic apparatus”. It’s unclear whether the “processing result” hasn’t been completely processed, or the “split model” hasn’t been completely processed, or both. Thus, the scope of the claim is rendered indefinite. This rejection is applicable to substantially similar claim 14. This limitation is interpreted as stating that the processing of the processing result is not yet completed by the first electronic apparatus.
Claim 7 additionally discloses the limitation “processing is sequentially performed from a split model that receives the received processing result as an input, from among the plurality of split models, based on the received processing result”. It’s unclear whether the split model is performing sequential processing, the split model is being processed sequentially, or sequential processing is following from the split model in some manner. This rejection is applicable to substantially similar claim 14. This limitation is interpreted as performing processing using a split model that receives the processing result, sequentially following some previous processing.
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 1-3, 6-10, and 13-16 are rejected under 35 U.S.C. 101 because the claimed inventions are directed to non-statutory subject matter without significantly more.
Claim 1
Step 1: The claim recites “A method”, and is therefore directed to the statutory category of process
Step 2A Prong 1: The claim recites the following judicial exception(s)
obtaining performance information of a second electronic apparatus: This can be performed as a mental process. One can merely observe the performance of a second electronic apparatus.
splitting the artificial intelligence model into a plurality of split models: This can be performed as a mental process. One can mentally partition the model into a set of subsets.
estimating a processing time required for the second electronic apparatus to execute each of the plurality of split models based on the performance information: This can be performed as a mental process. One can merely guess how long it would take for the second electronic apparatus to execute each model based on previously observed performance information on the apparatus.
based on the estimated processing time, determining at least one split model to be executed by the second electronic apparatus, from among the plurality of split models: This can be performed as a mental process. One can merely determine the split model with the fastest observed or estimated processing time to be executed.
updating the performance information of the second electronic apparatus based on the information about the time consumed: This can be performed as a mental process. One can merely update their mental performance information by observing the time consumption information.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the following additional element(s)
A method of performing, by a first electronic apparatus, distributed processing on an artificial intelligence model: This is mere instruction to perform the recited judicial exceptions on generic computer hardware (MPEP 2106.05(f)).
requesting the second electronic apparatus to execute the at least one split model to enable distributed processing of the artificial intelligence model: This is mere instruction to use a judicial exception to enable distributed processing in a generic manner (MPEP 2106.05(f)).
receiving, from the second electronic apparatus, a result of a processing of the at least one split model, and information about a time consumed to process the at least one split model by the second electronic apparatus: This amounts to mere data reception and is insignificant extra-solution activity (MPEP 2106.05(g)).
determining, based on the updated performance information, whether to request processing of a split model that is not processed yet to the second electronic apparatus: This is mere instruction to apply a judicial exception in a generic manner (MPEP 2106.05(f)).
Step 2B: The following additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
A method of performing, by a first electronic apparatus, distributed processing on an artificial intelligence model: This is mere instruction to perform the recited judicial exceptions on generic computer hardware (MPEP 2106.05(f)).
requesting the second electronic apparatus to execute the at least one split model to enable distributed processing of the artificial intelligence model: This is mere instruction to use a judicial exception to enable distributed processing in a generic manner (MPEP 2106.05(f)).
receiving, from the second electronic apparatus, a result of a processing of the at least one split model, and information about a time consumed to process the at least one split model by the second electronic apparatus: This is an instance of receiving data over a network, a limitation known to be well-understood, routine, and conventional (MPEP 2106.05(d) II. i.)
determining, based on the updated performance information, whether to request processing of a split model that is not processed yet to the second electronic apparatus: This is mere instruction to apply a judicial exception in a generic manner (MPEP 2106.05(f)).
Claim 2
Step 1: The claim recites a process, as in claim 1
Step 2A Prong 1: The claim recites the following further judicial exception(s)
wherein the splitting the artificial intelligence model comprises: splitting the artificial intelligence model into the plurality of split models based on the performance information of the second electronic apparatus: Splitting the AI model into a plurality of split models can still be performed as a mental process. One can mentally partition the model into a set of subsets based on observed performance of the second electronic apparatus.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the additional element(s)
Step 2B: The additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
Claim 3
Step 1: The claim recites a process, as in claim 1
Step 2A Prong 1: The claim recites the following further judicial exception(s)
wherein the splitting of the artificial intelligence model comprises: identifying at least one layer to be split from a plurality of layers that are included in the artificial intelligence model; and splitting the artificial intelligence model into the plurality of split models each of which comprises the identified at least one layer as an output layer: Splitting the AI model into a plurality of split models can still be performed as a mental process. One can mentally partition the model into a set of subsets, each subset’s output layer being identified from the AI model.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the additional element(s)
Step 2B: The additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
Claim 6
Step 1: The claim recites a process, as in claim 1
Step 2A Prong 1: The claim recites the following judicial exception(s)
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the further additional element(s)
sequentially operating, by the first electronic apparatus, the plurality of split models including the at least one split model that is requested to be operated by the second electronic apparatus: This is mere instruction to operate the split models in a generic manner (MPEP 2106.05(f)).
Step 2B: The further additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
sequentially operating, by the first electronic apparatus, the plurality of split models including the at least one split model that is requested to be operated by the second electronic apparatus: This is mere instruction to operate the split models in a generic manner (MPEP 2106.05(f)).
Claim 7
Step 1: The claim recites a process, as in claim 6
Step 2A Prong 1: The claim recites the following judicial exception(s)
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the further additional element(s)
wherein, when a processing result of a split model that has not yet been completely processed by the first electronic apparatus is received from the second electronic apparatus while the plurality of split models are processed by the first electronic apparatus, processing is sequentially performed from a split model that receives the received processing result as an input, from among the plurality of split models, based on the received processing result: This is mere instruction to operate the split models in a generic manner (MPEP 2106.05(f)).
Step 2B: The further additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
wherein, when a processing result of a split model that has not yet been completely processed by the first electronic apparatus is received from the second electronic apparatus while the plurality of split models are processed by the first electronic apparatus, processing is sequentially performed from a split model that receives the received processing result as an input, from among the plurality of split models, based on the received processing result: This is mere instruction to operate the split models in a generic manner (MPEP 2106.05(f)).
Claim 8
Step 1: The claim recites “A first electronic apparatus”, and is therefore directed to the statutory category of machine
Step 2A Prong 1: The claim recites the following judicial exception(s)
obtain performance information of a second electronic apparatus: This can be performed as a mental process. One can merely observe the performance of a second electronic apparatus.
split the artificial intelligence model into a plurality of split models: This can be performed as a mental process. One can mentally partition the model into a set of subsets.
estimate a processing time required for the second electronic apparatus to execute each of the plurality of split models based on the performance information: This can be performed as a mental process. One can merely guess how long it would take for the second electronic apparatus to execute each model based on previously observed performance information on the apparatus.
based on the estimated processing time, determine at least one split model to be executed by the second electronic apparatus, from among the plurality of split models: This can be performed as a mental process. One can merely determine the split model with the fastest observed or estimated processing time to be executed.
update the performance information of the second electronic apparatus based on the information about the time consumed: This can be performed as a mental process. One can merely update their mental performance information by observing the time consumption information.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the following additional element(s)
a memory storing the artificial intelligence model; a communication interface; and at least one processor configured to: This is mere instruction to perform the recited judicial exceptions on generic computer hardware (MPEP 2106.05(f)).
transmit, to the second electronic apparatus via the communication interface, a request for executing the at least one: This amounts to mere data transmission and is insignificant extra-solution activity (MPEP 2106.05(g)).
receive, from the second electronic apparatus, a result of a processing of the at least one split model, and information about a time consumed to process the at least one split model by the second electronic apparatus: This amounts to mere data reception and is insignificant extra-solution activity (MPEP 2106.05(g)).
determine, based on the updated performance information, whether to request processing of a split model that is not processed yet to the second electronic apparatus: This is mere instruction to apply a judicial exception in a generic manner (MPEP 2106.05(f)).
Step 2B: The following additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
a memory storing the artificial intelligence model; a communication interface; and at least one processor configured to: This is mere instruction to perform the recited judicial exceptions on generic computer hardware (MPEP 2106.05(f)).
transmit, to the second electronic apparatus via the communication interface, a request for executing the at least one: This is an instance of sending data over a network, a limitation known to be well-understood, routine, and conventional (MPEP 2106.05(d) II. i.)
receive, from the second electronic apparatus, a result of a processing of the at least one split model, and information about a time consumed to process the at least one split model by the second electronic apparatus: This is an instance of receiving data over a network, a limitation known to be well-understood, routine, and conventional (MPEP 2106.05(d) II. i.)
determine, based on the updated performance information, whether to request processing of a split model that is not processed yet to the second electronic apparatus: This is mere instruction to apply a judicial exception in a generic manner (MPEP 2106.05(f)).
Claims 9-10 & 13-14
Step 1: Claims 9-10 & 13-14 recite a machine, as in claim 8.
Step 2A Prong 1: Claims 9-10 & 13-14 recite the same judicial exception(s) as claims 2-3 & 6-7, respectively.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through any additional elements. The analysis of claims 9-10 & 13-14 at this step mirrors that of claims 2-3 & 6-7, respectively, with the exception that claims 9-10 & 13-14 are directed to “a memory storing the artificial intelligence model; at least one processor configured to:”, said processor performing operations mirroring those of claims 2-3 & 6-7. This is a mere instruction to apply the exceptions using generic computer equipment (MPEP 2106.05(f)).
Step 2B: The additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s). The analysis of claims 9-10 & 13-14 at this step mirrors that of claims 2-3 & 6-7, with the exception that claims 9-10 & 13-14 are directed to “a memory storing the artificial intelligence model; at least one processor configured to:”, said processor performing operations mirroring those of claims 2-3 & 6-7. This is a mere instruction to apply the exceptions using generic computer equipment (MPEP 2106.05(f)).
Claim 15
Step 1: The claim recites a process, as in claim 1
Step 2A Prong 1: The claim recites no further judicial exception(s)
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the further additional element(s)
A non-transitory computer-readable recording medium having recorded thereon a program for executing the method of claim 1: This is mere instruction to execute the judicial exceptions of claim 1 with generic computer hardware (MPEP 2106.05(f)).
Step 2B: The further additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
A non-transitory computer-readable recording medium having recorded thereon a program for executing the method of claim 1: This is mere instruction to execute the judicial exceptions of claim 1 with generic computer hardware (MPEP 2106.05(f)).
Claim 16
Step 1: The claim recites a process, as in claim 1
Step 2A Prong 1: The claim recites the following further judicial exception(s)
wherein the performance information comprises at least one of: a performance of a processor included in the second electronic apparatus, a number of processes currently being processed by the second electronic apparatus, a remaining battery level of the second electronic apparatus, or a memory size of the second electronic apparatus: Obtaining performance information of a second electronic apparatus can still be performed as a mental process. One can merely observe the performance of a second electronic apparatus in terms of processors, battery level, or memory consumption.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the additional element(s)
Step 2B: The additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
Claim Rejections - 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 1-4, 6-11, 13-14 & 16 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Kang et al. (Neurosurgeon: Collaborative Intelligence Between the Cloud and Mobile Edge, published 2017, ASPLOS '17: Proceedings of the Twenty-Second International Conference on Architectural Support for Programming Languages and Operating Systems Pages 615 – 629), hereafter referred to as Kang.
Regarding claim 1, Kang discloses [a] method of performing, by a first electronic apparatus, distributed processing on an artificial intelligence model, the method comprising:
obtaining performance information of a second electronic apparatus: “At Deployment – Neurosurgeon profiles the mobile device (second electronic apparatus) and the server (first electronic apparatus) to generate performance prediction models for the spectrum of DNN layer types” (Kang, page 621, right column, paragraph 2). Note that ‘server’ and ‘cloud’ are used somewhat interchangeably within Kang.
splitting the artificial intelligence model into a plurality of split models:
“Neurosurgeon, a system to intelligently partition DNN (artificial intelligence model) computation between the mobile and cloud.” (Kang, page 616, right column, paragraph 5)
PNG
media_image1.png
349
908
media_image1.png
Greyscale
(Kang, page 622, Figure 10). The AI model is partitioned into two split models, one on the mobile device and one on the server.
estimating a processing time required for the second electronic apparatus to execute each of the plurality of split models based on the performance information; based on the estimated processing time, determining at least one split model to be executed by the second electronic apparatus, from among the plurality of split models:
“We observe that for each layer type, there is a large latency variation across layer configurations. Thus, to construct the prediction model for each layer type, we vary the configurable parameters of the layer and measure the latency and power consumption (performance information) for each configuration. Using these profiles, we establish a regression model for each layer type to predict the latency (processing time) and power of the layer based on its configuration.” (Kang, page 622, left column, paragraph 4)
“Neurosurgeon extracts each layer’s type and configuration (
L
i
) and uses the regression models to predict the latency of executing layer
L
i
on mobile (second electronic apparatus) (
T
M
i
) and cloud (
T
C
i
), while taking into consideration of current datacenter load level (K). Line 13 estimates the power of executing layer Li on the mobile device (
P
M
i
) and line 14 calculates the wireless data transfer latency (
T
U
i
) based on the latest wireless network bandwidth.” (Kang, page 622, right column, paragraph 6)
“Neurosurgeon then selects the best partition point. The candidate points are after each layer. Lines 16 and 18 evaluate the performance when partitioning at each candidate point and select the point for either best end-to-end latency or best mobile energy consumption.” (Kang, page 623, left column, paragraph 2). The best partition point results in a split model to be executed by the second electronic apparatus.
PNG
media_image2.png
534
555
media_image2.png
Greyscale
(Kang, page 623, left column, Algorithm 1). As shown above on line 16, the performance of the mobile device (second electronic apparatus) is measured on all possible split models (one measurement for each partition
PNG
media_image3.png
66
95
media_image3.png
Greyscale
).
requesting the second electronic apparatus to execute the at least one split model to enable distributed processing of the artificial intelligence model:
“the steps are as follows: … Neurosurgeon selects the best partition point, optimizing for best end-to-end latency or best mobile energy consumption; 4) Neurosurgeon executes the DNN, partitioning work between the mobile (second electronic apparatus) and cloud” (Kang, page 622, left column, paragraph 2)
“Partitioning computation after a specific layer means executing the DNN on the mobile (second electronic apparatus) up to that layer (at least one split model), transferring the output of that layer to the cloud via wireless network, and executing the remaining layers in the cloud.” (Kang, page 619, right column, paragraph 3)
receiving, from the second electronic apparatus, a result of a processing of the at least one split model, and information about a time consumed to process the at least one split model by the second electronic apparatus; updating the performance information of the second electronic apparatus based on the information about the time consumed:
“Neurosurgeon extracts each layer’s type and configuration (
L
i
) and uses the regression models to predict the latency (processing result) of executing layer
L
i
on mobile (second electronic apparatus) (
T
M
i
) and cloud (
T
C
i
), while taking into consideration of current datacenter load level (K). Line 13 estimates the power of executing layer Li on the mobile device (
P
M
i
) and line 14 calculates the wireless data transfer latency (
T
U
i
) based on the latest wireless network bandwidth.” (Kang, page 622, right column, paragraph 6)
PNG
media_image2.png
534
555
media_image2.png
Greyscale
(Kang, page 623, left column, Algorithm 1). As shown above on line 16, the latency (time) of the mobile device (second electronic apparatus) is measured on all possible split models (one measurement for each partition
PNG
media_image3.png
66
95
media_image3.png
Greyscale
). This performance information is stored, at least insofar as storing the minimum model partition cost.
determining, based on the updated performance information, whether to request processing of a split model that is not processed yet to the second electronic apparatus:
“Neurosurgeon then selects the best partition point. The candidate points are after each layer. Lines 16 and 18 (calculations based on updated performance information) evaluate the performance when partitioning at each candidate point and select the point for either best end-to-end latency or best mobile energy consumption.” (Kang, page 623, left column, paragraph 2)
“Partitioning computation after a specific layer means executing the DNN on the mobile (second electronic apparatus) up to that layer (second apparatus split model), transferring the output of that layer to the cloud (first electronic apparatus) via wireless network, and executing the remaining layers in the cloud” (Kang, page 619, right column, paragraph 3).
Examiner’s note: Kang’s method determines the optimal split of the network to be performed on the second apparatus vs. the first. In other words, it determines which split model should be processed on the second apparatus before transferring processing to the first.
Kang relates to partitioning neural network layers across distributed hardware and is analogous to the claimed invention.
Regarding claim 2, the rejection of claim 1 in view of Kang is incorporated. Kang further discloses a method, wherein the splitting the artificial intelligence model comprises: splitting the artificial intelligence model into the plurality of split models based on the performance information of the second electronic apparatus:
“At Deployment – Neurosurgeon profiles the mobile device (second electronic apparatus) and the server to generate performance prediction models for the spectrum of DNN layer types” (Kang, page 621, right column, paragraph 2).
“the steps are as follows: 1) Neurosurgeon analyzes and extracts the DNN architecture’s layer types and configurations; 2) the system uses the stored layer performance prediction models to estimate the latency and energy consumption for executing each layer on the mobile and cloud; 3) with these predictions, combined with the current wireless connection bandwidth and datacenter load level, Neurosurgeon selects the best partition point, optimizing for best end-to-end latency or best mobile energy consumption; 4) Neurosurgeon executes the DNN, partitioning work between the mobile (second electronic apparatus) and cloud” (Kang, page 622, left column, paragraph 2)
“Partitioning computation after a specific layer means executing the DNN on the mobile up to that layer (second apparatus split model), transferring the output of that layer to the cloud via wireless network, and executing the remaining layers (first apparatus split model) in the cloud.” (Kang, page 619, right column, paragraph 3)
Regarding claim 3, the rejection of claim 1 in view of Kang is incorporated. Kang further discloses a method, wherein the splitting of the artificial intelligence model comprises: identifying at least one layer to be split from a plurality of layers that are included in the artificial intelligence model; and splitting the artificial intelligence model into the plurality of split models each of which comprises the identified at least one layer as an output layer:
PNG
media_image1.png
349
908
media_image1.png
Greyscale
(Kang, page 622, Figure 10).
“Partitioning computation after a specific layer means executing the DNN on the mobile up to that layer (second apparatus split model), transferring the output of that layer to the cloud via wireless network, and executing the remaining layers (first apparatus split model) in the cloud.” (Kang, page 619, right column, paragraph 3). Two output layers are split from each other. The first is the layer at the partition point, which becomes the output of the second apparatus split model. The second is the original output layer of the model, which becomes the output of the first apparatus split model.
Regarding claim 4, the rejection of claim 3 in view of Kang is incorporated. Kang further discloses a method, wherein the splitting of the artificial intelligence model comprises identifying the at least one layer for splitting the artificial intelligence model based on at least one of:
a size of data output from each of the plurality of layers, a data rate between the first electronic apparatus and the second electronic apparatus:
“line 14 calculates the wireless data transfer latency (
T
U
i
) (data rate) based on the latest wireless network bandwidth.” (Kang, page 622, right column, paragraph 6)
“Neurosurgeon then selects the best partition point. The candidate points are after each layer. Lines 16 and 18 evaluate the performance when partitioning at each candidate point and select the point for either best end-to-end latency or best mobile energy consumption.” (Kang, page 623, left column, paragraph 2).
PNG
media_image2.png
534
555
media_image2.png
Greyscale
(Kang, page 623, left column, Algorithm 1). The data transfer latency (data rate) is a variable used to calculate estimated split performance in lines 16 and 18. As shown in lines 4 and 14, data transfer latency is calculated using the data size at each layer.
“For a forward pass through a DNN, the output of a layer is the input to the next layer.” (Kang, page 617, left column, paragraph 1). The size of one layer’s input is identical to that of the previous layer’s output.
whether each of the plurality of layers is to be processed in an accelerated manner by the second electronic apparatus:
“Partitioning computation after a specific layer means executing the DNN on the mobile up to that layer, transferring the output of that layer to the cloud via wireless network, and executing the remaining layers in the cloud (second electronic apparatus)” (Kang, page 619, right column, paragraph 3)
“We use Caffe [18], an actively developed open-source deep learning library, for the mobile and server platform. For the mobile CPU, we use OpenBLAS [19], a NEONvectorized matrix multiplication library and use the 4 cores available. For both GPUs, we use cuDNN [20], an optimized NVIDIA library that accelerates key layers in Caffe, and use Caffe’s CUDA implementations for rest of the layers.” (Kang, page 617, right column, paragraph 2)
Regarding claim 6, the rejection of claim 1 in view of Kang is incorporated. Kang discloses a method, further comprising: sequentially operating, by the first electronic apparatus, the plurality of split models including the at least one split model that is requested to be operated by the second electronic apparatus: “Partitioning computation after a specific layer means executing the DNN on the mobile (second electronic apparatus) up to that layer (at least one split model), transferring the output of that layer to the cloud (first electronic apparatus) via wireless network, and executing the remaining layers in the cloud” (Kang, page 619, right column, paragraph 3). The cloud (first electronic apparatus) operates the at least one split model of the mobile (second electronic apparatus) by transferring its output(s) to the next split model.
Regarding claim 7, the rejection of claim 6 in view of Kang is incorporated. Kang further discloses a method, wherein, when a processing result of a split model that has not yet been completely processed by the first electronic apparatus is received from the second electronic apparatus while the plurality of split models are processed by the first electronic apparatus, processing is sequentially performed from a split model that receives the received processing result as an input from among the plurality of split models, based on the received processing result: “Partitioning computation after a specific layer means executing the DNN on the mobile (second electronic apparatus) up to that layer (second apparatus split model), transferring the output (processing result of a split model that has not yet been completely processed by the first electronic apparatus) of that layer to the cloud (first electronic apparatus) via wireless network, and executing the remaining layers (first apparatus split model) in the cloud” (Kang, page 619, right column, paragraph 3). The output is received by the cloud (first apparatus) while it begins processing its own split model, using the output received (processing result) as input. Processing on the cloud’s split model is performed sequentially after that of the mobile’s split model.
Regarding claim 8, Kang discloses [a] first electronic apparatus configured to perform distributed processing on an artificial intelligence model, the first electronic apparatus comprising:
a memory storing the artificial intelligence model; a communication interface; and at least one processor configured to:
PNG
media_image4.png
180
551
media_image4.png
Greyscale
(Kang, page 617, left column, Table 1). The server (first electronic apparatus) contains a CPU (processor), and memory storing the artificial intelligence model, as made clear in mappings below.
obtain performance information of a second electronic apparatus to perform a distributed processing on the artificial intelligence model:
“Using this insight, we design Neurosurgeon, a lightweight scheduler to automatically partition DNN computation between mobile devices and datacenters at the granularity of neural network layers.” (Kang, page 615, left column, paragraph 3)
“At Deployment – Neurosurgeon profiles the mobile device (second electronic apparatus) and the server (first electronic apparatus) to generate performance prediction models for the spectrum of DNN layer types” (Kang, page 621, right column, paragraph 2). Note that ‘server’ and ‘cloud’ are used somewhat interchangeably within Kang.
“we implement our client-server interface using Thrift [33], an open source flexible RPC interface for inter-process communication” (Kang, page 623, right column, paragraph 1)
split the artificial intelligence model into a plurality of split models
“Neurosurgeon, a system to intelligently partition DNN (artificial intelligence model) computation between the mobile and cloud.” (Kang, page 616, right column, paragraph 5)
PNG
media_image1.png
349
908
media_image1.png
Greyscale
(Kang, page 622, Figure 10). The AI model is partitioned into two split models, one on the mobile device and one on the server.
estimate a processing time required for the second electronic apparatus to execute each of the plurality of split models based on the performance information; based on the estimated processing time, determine at least one split model to be executed by the second electronic apparatus from among the plurality of split models:
“We observe that for each layer type, there is a large latency variation across layer configurations. Thus, to construct the prediction model for each layer type, we vary the configurable parameters of the layer and measure the latency and power consumption (performance information) for each configuration. Using these profiles, we establish a regression model for each layer type to predict the latency (processing time) and power of the layer based on its configuration.” (Kang, page 622, left column, paragraph 4)
“Neurosurgeon extracts each layer’s type and configuration (
L
i
) and uses the regression models to predict the latency of executing layer
L
i
on mobile (second electronic apparatus) (
T
M
i
) and cloud (
T
C
i
), while taking into consideration of current datacenter load level (K). Line 13 estimates the power of executing layer Li on the mobile device (
P
M
i
) and line 14 calculates the wireless data transfer latency (
T
U
i
) based on the latest wireless network bandwidth.” (Kang, page 622, right column, paragraph 6)
“Neurosurgeon then selects the best partition point. The candidate points are after each layer. Lines 16 and 18 evaluate the performance when partitioning at each candidate point and select the point for either best end-to-end latency or best mobile energy consumption.” (Kang, page 623, left column, paragraph 2). The best partition point results in a split model to be executed by the second electronic apparatus.
PNG
media_image2.png
534
555
media_image2.png
Greyscale
(Kang, page 623, left column, Algorithm 1). As shown above on line 16, the performance of the mobile device (second electronic apparatus) is measured on all possible split models (one measurement for each partition
PNG
media_image3.png
66
95
media_image3.png
Greyscale
).
transmit, to the second electronic apparatus via the communication interface, a request for executing the at least one split model to enable distributed processing of the artificial intelligence model:
“the steps are as follows: … Neurosurgeon selects the best partition point, optimizing for best end-to-end latency or best mobile energy consumption; 4) Neurosurgeon executes the DNN, partitioning work between the mobile (second electronic apparatus) and cloud” (Kang, page 622, left column, paragraph 2)
“Partitioning computation after a specific layer means executing the DNN on the mobile (second electronic apparatus) up to that layer (at least one split model), transferring the output of that layer to the cloud via wireless network, and executing the remaining layers in the cloud.” (Kang, page 619, right column, paragraph 3) The communication interface between the mobile device and cloud is inherent in this data transfer.
receive, from the second electronic apparatus via the communication interface, a result of a processing of the at least one split model, and information about a time consumed to process the at least one split model by the second electronic apparatus; update the performance information of the second electronic apparatus based on the information about the time consumed:
“Neurosurgeon extracts each layer’s type and configuration (
L
i
) and uses the regression models to predict the latency (processing result) of executing layer
L
i
on mobile (second electronic apparatus) (
T
M
i
) and cloud (
T
C
i
), while taking into consideration of current datacenter load level (K). Line 13 estimates the power of executing layer Li on the mobile device (
P
M
i
) and line 14 calculates the wireless data transfer latency (
T
U
i
) based on the latest wireless network bandwidth.” (Kang, page 622, right column, paragraph 6)
PNG
media_image2.png
534
555
media_image2.png
Greyscale
(Kang, page 623, left column, Algorithm 1). As shown above on line 16, the latency (time) of the mobile device (second electronic apparatus) is measured on all possible split models (one measurement for each partition
PNG
media_image3.png
66
95
media_image3.png
Greyscale
). This performance information is stored, at least insofar as storing the minimum model partition cost.
determine, based on the updated performance information, whether to request processing of a split model that is not processed yet to the second electronic apparatus:
“Neurosurgeon then selects the best partition point. The candidate points are after each layer. Lines 16 and 18 (calculations based on updated performance information) evaluate the performance when partitioning at each candidate point and select the point for either best end-to-end latency or best mobile energy consumption.” (Kang, page 623, left column, paragraph 2)
“Partitioning computation after a specific layer means executing the DNN on the mobile (second electronic apparatus) up to that layer (second apparatus split model), transferring the output of that layer to the cloud (first electronic apparatus) via wireless network, and executing the remaining layers in the cloud” (Kang, page 619, right column, paragraph 3).
Examiner’s note: Kang’s method determines the optimal split of the network to be performed on the second apparatus vs. the first. In other words, it determines which split model should be processed on the second apparatus before transferring processing to the first.
Kang relates to partitioning neural network layers across distributed hardware and is analogous to the claimed invention.
The analysis of claims 9-11 & 13-14 mirrors that of claims 2-4 & 6-7, with the exception that claims 9-11 & 13-14 are directed to generic computer hardware which executes the methods of claims 2-4 & 6-7. This generic hardware is taught by Kang, as discussed regarding claim 8. Thus, claims 9-11 & 13-14 are rejected under the same rationales used for claims 2-4 & 6-7, respectively.
Regarding claim 16, the rejection of claim 1 in view of Kang is incorporated. Kang further discloses a method, wherein the performance information comprises at least one of: a performance of a processor included in the second electronic apparatus, a number of processes currently being processed by the second electronic apparatus, a remaining battery level of the second electronic apparatus, or a memory size of the second electronic apparatus: “Neurosurgeon extracts each layer’s type and configuration (
L
i
) and uses the regression models to predict the latency (performance of a processor) of executing layer
L
i
on mobile (second electronic apparatus) (
T
M
i
) and cloud (
T
C
i
), while taking into consideration of current datacenter load level (K). Line 13 estimates the power of executing layer Li on the mobile device (
P
M
i
) and line 14 calculates the wireless data transfer latency (
T
U
i
) based on the latest wireless network bandwidth.” (Kang, page 622, right column, paragraph 6)
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.
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.
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Kang et al. (Neurosurgeon: Collaborative Intelligence Between the Cloud and Mobile Edge, published 2017, ASPLOS '17: Proceedings of the Twenty-Second International Conference on Architectural Support for Programming Languages and Operating Systems Pages 615 – 629), hereafter referred to as Kang, and further in view of Chen et al. (TRAINING GIANT NEURAL NETWORKS USING PIPELINE PARALLELISM, filed 8/10/2020, US 2021/0042620 A1), hereafter referred to as Chen.
Regarding claim 15, the rejection of claim 1 in view of Kang is incorporated. While Kang fails to disclose the further limitations of the claim, Chen discloses [a] non-transitory computer-readable recording medium having recorded thereon a program for executing the method of claim 1: “Implementations of the subject matter described in this specification can be implemented as one or more computer programs, e.g., one or more modules of computer program instructions, encoded on a computer program carrier, for execution by, or to control the operation of, data processing apparatus. The carrier may be a tangible non-transitory computer storage medium” (Chen, [0137])
Chen relates to partitioning layers of a neural network to execute on a distributed computing system and is analogous to the claimed invention. Kang teaches a method of partitioning neural networks on distributed hardware. The claimed invention improves upon this method by storing it in the form of instructions on computer hardware. Chen teaches non-transitory computer hardware, applicable to Kang. A person of ordinary skill in the art would have recognized that storing Kang’s method as computer instructions on Chen’s hardware would lead to the predictable result of the method being executable by a computing system, and would improve the known device by allowing it to be performed with real data (MPEP 2143 I. (D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results).
Allowable Subject Matter
Claim 17 would be allowable over the prior art of record if rewritten or amended to overcome the objections set forth in this Office action.
Regarding claim 17, “determining a weight value for each of the identified top n number of layers based on an order of execution of the identified top n number of layers; and determining split points of the artificial intelligence model based respective products of the output data sizes and corresponding weight values for the identified top n number of layers” is not taught by the prior art of record. The closest prior arts of record are Kang et al. (Neurosurgeon: Collaborative Intelligence Between the Cloud and Mobile Edge, published 2017, ASPLOS '17: Proceedings of the Twenty-Second International Conference on Architectural Support for Programming Languages and Operating Systems Pages 615 – 629), hereafter referred to as Kang, Teerapittayanon et al. (BranchyNet: Fast Inference via Early Exiting from Deep Neural Networks, published 2016, 2016 ICPR pp. 4-8), hereafter referred to as BranchyNet, and Teerapittayanon et al. (Distributed Deep Neural Networks over the Cloud, the Edge and End Devices, published 2017, 2017 IEEE ICDCS pp. 328-339), hereafter referred to as DDNN.
Kang discloses
identifying a top n number of layers having smallest output data sizes among a plurality of layers included in the artificial intelligence model:
“Data Size Variations – The right bars (dark-colored) in Figure 5 shows the size of each layer’s output data, which is also the input to the next layer” (Kang, page 619, left column, paragraph 1)
PNG
media_image2.png
534
555
media_image2.png
Greyscale
(Kang, page 623, left column, Algorithm 1). In line 4, it’s shown that the data size of each layer is measured, inherently including the smallest output data sizes.
determining a weight value for each of the identified top n number of layers based on an order of execution of the identified top n number of layers; and determining split points of the artificial intelligence model based respective products of the output data sizes and corresponding weight values for the identified top n number of layers: “Neurosurgeon then selects the best partition point (split point). The candidate points are after each layer. Lines 16 and 18 evaluate the performance when partitioning at each candidate point and select the point for either best end-to-end latency (weight value) or best mobile energy consumption (weight value)” (Kang, page 623, left column, paragraph 2). As seen in lines 16 and 18 of Algorithm 1, the weight of a potential split at a particular layer comprises sums of consumption metrics up to and including the potential split layer. Thus, these calculations are based on an order of execution of the identified layers.
Kang does not disclose multiplying output data sizes and corresponding weight values, or determining split points based on such a calculation.
The combination of BranchyNet and DDNN discloses:
determining a weight value for each of the identified top n number of layers based on an order of execution of the identified top n number of layers; and determining split points of the artificial intelligence model based respective products of the output data sizes and corresponding weight values for the identified top n number of layers:
(DDNN) “In DDNN, exit points are placed at physical boundaries (e.g., between the last NN layer on an end device and the first NN layer in the next higher layer of the distributed computing hierarchy such as the edge or the cloud).” (DDNN, page 330, left column, paragraph 4)
(DDNN) “To train the DDNN we form a joint optimization problem as minimizing a weighted sum (product) of the loss functions of each exit (split point):
PNG
media_image5.png
102
431
media_image5.png
Greyscale
where N is the total number of exit points and
w
n
(weight value) is the associated weight of each exit (identified layer)” (DDNN, page 332, left column, paragraph 1)
(DDNN) “DDNN leverages our earlier work on BranchyNet [3] which allows early exit points to be placed in a DNN.” (DDNN, page 329, left column, paragraph 4)
(BranchyNet) “When selecting the weight of each branch, we observed that giving more weight to early branches improves the accuracy of the later branches due to the added regularization.” (BranchyNet, page 2467, right column, paragraph 4). Weight values proportional to branch point in the network are determined proportionally to the order of execution.
The combination of BranchyNet and DDNN does not disclose identifying a top number of layers with small output sizes, determining a split point based off said small output size layers, or multiplying weights and output sizes of the layers.
A combination of Kang, BranchyNet, and DNN would not have been obvious without the use of impermissible hindsight. Therefore, claim 17 is not obvious over the prior art of record.
Response to Arguments
The following responses address arguments and remarks made in the instant remarks dated 03/23/2026.
Priority
Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55.
Objections
In light of the instant amendments, some previous objections to the claims have been withdrawn. However, some previous objections to the claims are maintained and objections to the new claims have been made.
112 Rejections
In light of the instant amendments, previous rejections under 35 U.S.C. 112(b) have been withdrawn for cancelled claims. However, rejections for claims 7 and 14 are maintained under this statute.
101 Rejections
On pages 9-11 of the instant remarks, the Applicant argues that the claimed invention represents an improvement in a technical field:
“Applicant respectfully submits that the claims are patent eligible because they represent
an improvement in a technical field. The specification identifies the following technical problem
in paras. [0003 ]-[0005]:
…
The claims address the above technical problem by providing a system for performing
distributed processing on an artificial intelligence (AI) model. (Spec. at para. [0006]). For
example, independent claim 1 recites ‘estimating a processing time required for the second
electronic apparatus to execute each of the plurality of split models based on the performance
information; based on the estimated processing time, determining at least one split model to be
executed by the second electronic apparatus, from among the plurality of split models; and
requesting the second electronic apparatus to execute the at least one split model to enable
distributed processing of the artificial intelligence model.’ As another example, dependent claim
4 recites ‘wherein the splitting of the artificial intelligence model comprises identifying the at
least one layer for splitting the artificial intelligence model based on at least one of a size of data
output from each of the plurality of layers, a data rate between the first electronic apparatus and
the second electronic apparatus, and whether each of the plurality of layers is to be processed in
an accelerated manner by the second electronic apparatus.’”
In response to the Applicant’s argument that the claimed invention improves upon existing technology, the Examiner partly disagrees. The improvement of a claimed invention must be sufficiently detailed, as noted in MPEP 2106.05(a): “If it is asserted that the invention improves upon conventional functioning of a computer, or upon conventional technology or technological processes, a technical explanation as to how to implement the invention should be present in the specification. That is, the disclosure must provide sufficient details such that one of ordinary skill in the art would recognize the claimed invention as providing an improvement. The specification need not explicitly set forth the improvement, but it must describe the invention such that the improvement would be apparent to one of ordinary skill in the art … After the examiner has consulted the specification and determined that the disclosed invention improves technology, the claim must be evaluated to ensure the claim itself reflects the disclosed improvement in technology. Intellectual Ventures I LLC v. Symantec Corp., 838 F.3d 1307, 1316, 120 USPQ2d 1353, 1359 (Fed. Cir. 2016) (patent owner argued that the claimed email filtering system improved technology by shrinking the protection gap and mooting the volume problem, but the court disagreed because the claims themselves did not have any limitations that addressed these issues). That is, the claim must include the components or steps of the invention that provide the improvement described in the specification.”
The Examiner finds that the argued improvements are represented in claims 4, 11, and 17 of the amended claims. However, the Examiner disagrees that these improvements are represented in the independent or other dependent claims. The argued improvement is not solely partitioning processing between two computers, but rather partitioning network layers in a manner that maximizes performance while harnessing both the benefits of privacy preservation on primary hardware and higher processing speed on remote processing servers. Broadly determining partitions “based on the estimated processing time” or “based on the updated performance information” as claimed in the independent claims fails to capture either the details of how this improvement works (partitioning layers based on data transfer between apparatuses) or its intended goal (partitioning to maximize overall performance).
Thus, the rejections of claims 1-3, 6-10, and 13-15 are maintained on these grounds. Additionally, new claim 16 is found to be rejected under 35 U.S.C. 101. See the 101 rejections section for more detail.
In light of claim 4 representing the argued improvements, The Examiner suggests bringing the details of claim 4 and claim 3 (which claim 4 depends on) into the language of the independent claims.
102 / 103 Rejections
On page 13 of the instant remarks, the Applicant argues that Kang fails to disclose the amended limitations of the independent claims:
“Accordingly, Kang discloses a single step for generating a performance prediction model
for a mobile device and a single determination step for partitioning work between the mobile
device and the cloud. Kang, however, does not disclose performing multiple determination steps
for a single artificial intelligence model including a determination based on updated information,
let alone, "receiving, from the second electronic apparatus, a result of a processing of the at least
one split model, and information about a time consumed to process the at least one split model by
the second electronic apparatus; updating the performance information of the second electronic
apparatus based on the information about the time consumed; and determining, based on the
updated performance information, whether to request processing of a split model that is not
processed yet to the second electronic apparatus," as recited by claim 1.”
In response to applicant's arguments above, it is noted that the amended claims do not recite temporal limitations specifying the order of limitations executed, nor limitations specifying which recited processes are mutually exclusive to one another. The Examiner contends that the steps of performing neurosurgeon recited by Kang are commensurate with the claim language.
Regarding “receiving, from the second electronic apparatus, a result of a processing of the at least one split model, and information about a time consumed to process the at least one split model by the second electronic apparatus”, Kang discloses retrieving processing results of the split model (latencies / processing time) for second electronic apparatus (Kang, page 622, right column, paragraph 6 & page 623, left column, Algorithm 1).
Regarding “updating the performance information of the second electronic apparatus based on the information about the time consumed”, Kang discloses updated calculated performance information for the second electronic apparatus based on the latencies (Kang, page 623, left column, Algorithm 1).
Regarding “determining, based on the updated performance information, whether to request processing of a split model that is not processed yet to the second electronic apparatus”, Kang discloses using the calculated performance estimates of each split model to determine where the model should be split between processing of the two apparatuses (Kang, page 623, left column, paragraph 2 & page 619, right column, paragraph 3).
Similar reasoning is applicable to substantially similar independent claim 8. No rejections are withdrawn on these grounds.
On page 14 of the instant remarks, the Applicant argues that new claims 16 and 17 are patentable over the cited references:
“New Claims
Applicant has added new claims 16 and 17 for the Examiner's consideration. Support for
these claims is found at least in paragraphs [0061 ], [0088], [0089], and [0 147]-[0149] of the
present specification.
Applicant respectfully submits that the new claims are patentable due to their
dependencies upon claim I as well as for additional distinct features recited therein. Therefore,
consideration and allowance of the claims are respectfully requested.”
Regarding the Applicant’s arguments above, the Examiner respectfully disagrees.
Claim 16 is found to be anticipated by Kang (Kang, page 622, right column, paragraph 6; see 102 rejections section for more detail).
Claim 17 would be allowable over the prior art due to its further limitations if amended to overcome current claim objections. See the allowable subject matter section for more detail.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Li et al. (Edge AI: On-Demand Accelerating Deep Neural Network Inference via Edge Computing, published 2019, arXiv:1910.05316v1) discloses a method of dynamically partitioning a neural network across a mobile and edge network device with layer-level granularity
Shi et al. (Privacy-Aware Edge Computing Based on Adaptive DNN Partitioning, published 2019, 2019 IEEE Global Communications Conference (GLOBECOM) Pages 1 – 6) discloses a method of dynamically partitioning a neural network across multiple devices based on performance requirements
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 Aaron P Gormley whose telephone number is (571)272-1372. The examiner can normally be reached Monday - Friday 12:00 PM - 8:00 PM 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, Michelle T Bechtold can be reached at (571) 431-0762. 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.
/AG/Examiner, Art Unit 2148
/MICHELLE T BECHTOLD/Supervisory Patent Examiner, Art Unit 2148
Prosecution Insights