DETAILED ACTION
Notice of Pre-AIA or AIA Status
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
2. Claims 1-20 are pending.
Information Disclosure Statement
3. The information disclosure statement (IDS) submitted on 4/17/2025 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Drawings
4. The drawings have been reviewed and are accepted as being in compliance with the provisions of 37 CFR 1.121.
Priority
5. Acknowledgment is made of applicant’s claim for foreign priority under 35 U.S.C. 119 (a)-(d). The certified copy has been obtained by the office on May 6 2025, by the Digital Access Service (DAS).
Claim Rejections - 35 USC § 112
6. 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.
7. Regarding claims 1-20, the phrase "to be" renders the claim indefinite because it is unclear whether the limitation(s) following the phrase are part of the claimed invention. See MPEP § 2173.05(d).
Claims 1-20 recite “an operator”, “…obtaining a format mapping scheme corresponding to the operator”, “selecting a startup operator from operators “, “the operator based on both the data format and the operator”, “both the data format and the operator”, “selecting a particular operator from the operators”, “the historical operator”..., etc. The examiner requests clarification of how many operators are there, if there is one and then modified, or what is the functionality of “an operator” and “a particular operator”. For the purpose of this office action, the Examiner will interpret the operators as functions/instructions of the neural network model.
Claim Rejections - 35 USC § 101
8. 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.
9. Claims 1-20 are rejected under 35 U.S.C. 101 as being directed to an abstract idea without significantly more.
Step 1: Claim 1 recites “A method for…”; the claim recites a series of steps and therefore is a process. Claim 12 recites “An apparatus…” therefore the claim is a machine. Claim 17 recites “A non-transitory computer-readable storage medium …”, obtaining an indication of a data format, an initial model file, a data sub-format, attribute, a final model file and a processing result, , therefore the claim is a manufacture.
Step 2A Prong One: Claims 1, 12, and 17 recite the limitations "obtaining" and specifically “obtaining an indication of a data format of input data of a neural network model; obtaining an initial model file for the neural network model, the initial model file indicating an operator of the neural network model; obtaining, by processing circuitry, a data sub-format of input data of the operator based on both the data format and the operator; obtaining attribute information of the processing circuitry; obtaining a final model file for the processing circuitry…” These limitations are processes that, under their broadest reasonable interpretation, cover performance of the limitation in the mind, but for the recitation of generic computer components.
That is, other than reciting a "processor of a computer", nothing in the claim element precludes the step from practically being performed in a human mind or with the aid of pen and paper. For example, “determining” in the context of this claim encompasses a user mentally, and with the aid of pen and paper, grouping and evaluating data, if the token is inexistent then store it.
If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind, then it falls within the “Mental Processes” grouping of abstract ideas (concepts performed in the human mind including an observation, evaluation, judgment, and opinion).
Step 2A Prong Two: The judicial exception is not integrated into a practical application. The claim recites the additional elements “based on compiling the initial model file according to both the attribute information and the data sub-format.” And The “compiling” based on a determination of the “processing result”, this limitation is a mere generic transmission and presentation of collected and analyzed data (MPEP 2106.05(g).
A claim to "collecting information, analyzing it, and displaying certain results of the collection and analysis," where the data analysis steps are recited at a high level of generality such that they are considered insignificant extra-solution activity. Electric Power Group v. Alstom, S.A., 830 F.3d 1350, 1353-54, 119 USPQ2d 1739, 1741-42 (Fed. Cir. 2016).
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The insignificant extra-solution activities listed above, including “compiling” and “running” and “the processing circuitry”, these limitations are generic transmission and presentation of collected and analyzed data. The limitations performed by a “neural network model” being a tool; generates a combination of data or “output” based on data and update them based on comparison of data ( it is recognized by the courts as well-understood, routine, and conventional activities when they are claimed in a merely generic manner. (see MPEP 2106.04(a)(2). There are no additional elements that amount to significantly more than the above-identified judicial exception (abstract idea).
Koninklijke KPN N.V. v.Gemalto M2M GmbH, 942 F.3d 1143, 1149 (Fed. Cir.2019) (quoting Affinity Labs of Tex., LLC v. DIRECTV, LLC, 838 F.3d 1253, 1257 (Fed. Cir. 2016)). In the context of software patents (which includes machine learning patents), the step-one inquiry determines “whether the claims focus on ‘the specific asserted improvement in computer capabilities . . . or, instead, on a process that qualifies as an abstract idea for which computers are invoked merely as a tool.’” Id. (alteration in original) (quoting Finjan, Inc. v. Blue Coat Sys., Inc., 879 F.3d 1299, 1303 (Fed. Cir. 2018)).
Generic use of AI without other parameters, such as “improving the mathematical algorithm or making machine learning better,” is abstract. Recentive, 134 F.4th at 1213. A claim that merely restricts an abstract idea to a particular field or environment is still directed to the abstract idea. See Recentive, 134 F.4th at 1213; GoTV, 2026 WL 346200, at *8; Intell. Ventures I LLC v. Capital One Bank (USA), 792 F.3d 1363, 1366 (Fed. Cir. 2015). In Recentive, we determined that a claim that was directed towards applying machine learning to a “new field of use” was directed to an abstract idea because “the only thing the claims disclose about the use of machine learning is that machine learning is used in a new environment.” Recentive, 134 F.4th at 1213. Rensselaer concedes that, under Recentive, claims that require nothing more than application of artificial intelligence to a new environment are directed to an abstract idea
As per Claim 2, The claim recites the additional limitations: wherein the obtaining the data sub-format comprises: obtaining a format mapping scheme corresponding to the operator; and obtaining the data sub-format of the input data of the operator based on mapping the data format to the data sub-format through the format mapping scheme.
This additional limitation elaborates on the abstract idea described above where the process/script results based on running a file, “based on the format” “sub-format”, for which a person can have a list of processes to apply, within the file, each identified with a tag/identifier for which the person can examine and select the process they are looking for, which is again a mental process.
As per Claim 3, The claim recites the additional limitations: wherein the obtaining the data sub-format comprises: selecting a startup operator from operators; obtaining a data sub-format of output data of the startup operator based on both the startup operator and the data format as a data sub-format of input data of the startup operator; based on a next operator of the startup operator existing in the operators, with the data sub-format of the output data of the startup operator as the data format and the next operator as the startup operator, returning to obtaining the data sub-format of output data of the startup operator; and based on no next operator of the startup operator existing in the operators, determining the data sub-format of the input data of the operator according to the data sub-format of the input data of the startup operator. This additional limitation elaborates on the abstract idea described above where the process/script results based on running a file, “initial model file” then select operators, as needed, for which a person can have a list of processes/instructions to apply, within the file, each identified with a tag/identifier for which the person can examine and select the process they are looking for, which is again a mental process.
As per Claim 4, The claim recites the additional limitations: wherein the obtaining the data sub-format of input data of the operator comprises: selecting a particular operator from the operators; determining a historical operator preceding the particular operator;
obtaining a data sub-format of input data of the particular operator based on both the data format and the historical operator; and determining the data sub-format of the input data of the operator according to the data sub-format of the input data of the particular operator. This additional limitation elaborates on the abstract idea described
above where the process/script results based on running a file, determination of data formats and operators, for which a person can have a list of processes to apply, within the file, each identified with a tag/identifier for which the person can examine and select the process they are looking for, which is again a mental process.
As per Claim 5, The claim recites the additional limitations: wherein the obtaining the data sub-format of the input data of the particular operator comprises: determining a conversion model parameter according to a model parameter of the historical operator; and obtaining the data sub-format of the input data of the particular operator based on both the data format and the conversion model parameter. This additional limitation elaborates on the abstract idea described above where the process/script results based on running a file, “input data” based on formats, for which a person can have a list of processes to apply, within the file, each identified with a tag/identifier for which the person can examine and select the process they are looking for, which is again a mental process.
As per Claim 6, The claim recites the additional limitations: wherein a plurality of processors are provided in the processing circuitry, and obtaining the final model file for the processing circuitry comprises: obtaining an original computation graph, wherein the original computation graph comprises a node that represents the operator; partitioning, according to the attribute information, the original computation graph into a subgraph matching each processor; and obtaining a final model file of each processor based on compiling the initial model file according to the attribute information, the data sub-format, and the subgraph. This additional limitation elaborates on the abstract idea described above where the process/script results based on running a file, “final model file”, for which a person can have a list of processes to apply, within the file, each identified with a tag/identifier for which the person can examine and select the process they are looking for, which is again a mental process.
As per Claim 7 , The claim recites the additional limitations: wherein the obtaining the final model file of each processor comprises: determining an initial data type of input data of the subgraph according to the original computation graph;
determining a final data type of the input data of the subgraph according to the attribute
information; and obtaining the final model file of each processor based on compiling the initial model file according to the attribute information, the data sub-format, the initial data type, and the final data type. This additional limitation elaborates on the abstract idea described above where the process/script results based on running a file, “initial model file”, for which a person can have a list of processes to apply, within the file, each identified with a tag/identifier for which the person can examine and select the process they are looking for, which is again a mental process.
As per Claim 8 , The claim recites the additional limitations: wherein the obtaining the final model file of each processor based on compiling the initial model file according to the attribute information, the data sub-format, the initial data type, and the final data type comprises: obtaining a data type conversion operator based on the initial data type being different from the final data type;
obtaining a first adjusted subgraph based on adjusting the subgraph according to the data type conversion operator; and obtaining the final model file of each processor based on compiling the initial model file according to the attribute information, the data sub-format, and the first adjusted subgraph. This additional limitation elaborates on the abstract idea described above where the process/script results based on running a file, “initial model file”, for which a person can have a list of processes to apply, within the file, each identified with a tag/identifier for which the person can examine and select the process they are looking for, which is again a mental process.
As per Claim 9 , The claim recites the additional limitations: wherein the obtaining the final model file of each processor based on compiling the initial model file according to the attribute information, the data sub-format, and the first adjusted subgraph comprises: determining a type of an operator in the subgraph; obtaining a second adjusted subgraph based on adjusting the operator in the subgraph
based on the type of the operator in the subgraph being a preset type; and
obtaining the final model file of each processor based on compiling the initial model file
according to the attribute information, the data sub-format, and the second adjusted subgraph. This additional limitation elaborates on the abstract idea described
above where the process/script results based on running a file, “initial model file”, for which a person can have a list of processes to apply, within the file, each identified with a tag/identifier for which the person can examine and select the process they are looking for, which is again a mental process.
As per Claim 10 , The claim recites the additional limitations: wherein the obtaining the processing result comprises: determining a data format of the data to be processed of the neural network model; obtaining adjusted data to be processed based on adjusting the data format of the data to be processed based on the data format of the data to be processed not matching the data sub-format in the final model file; and obtaining the processing result based on running the final model file with the adjusted data to be processed. This additional limitation elaborates on the abstract idea described
above where the process/script results based on running a file, for which a person can have a list of processes to apply, within the file, each identified with a tag/identifier for which the person can examine and select the process they are looking for, which is again a mental process.
As per Claim 11 , The claim recites the additional limitations: wherein the obtaining the
processing result comprises: obtaining a candidate processing result of the data to be processed based on running the final model file with the data to be processed;
determining an output data format of the candidate processing result; and
obtaining the processing result based on adjusting the output data format of the candidate processing result according to the final model file. This additional limitation elaborates on the abstract idea described above where the process/script results based on running a file, for which a person can have a list of processes to apply, within the file, each identified with a tag/identifier for which the person can examine and select the process they are looking for, which is again a mental process.
As per Claims 12-20, being the apparatus and non-transitory medium claims corresponding to the system claims 1-11 respectively and rejected under the same reason set forth in connection of the rejections of Claims 1-11.
Claim Rejections - 35 USC § 103
10. In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
11. 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.
12. Claim(s) 1-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over WANG et al (US 2023/0259774), in view of YANG et al (US 2023/0315410), hereinafter “WANG” and “YANG” respectively.
As per Claim 1, WANG discloses:
A data processing method, comprising: obtaining an indication of a data format of input data of a neural network model; (Par [0007] and par [0063], “…a deep learning model can be conveniently constructed to perform conversion and description between different formats of the deep learning model, iterative optimization, and flexible deployment.”) obtaining an initial model file for the neural network model, the initial model file indicating an operator of the neural network model; (Par [0007], “S1, parsing an input model file so as to acquire topological structure information of a neural network: parsing a user-defined function body including input and output parameters, and compiling the function body into a computation graph composed of logical computation expressions, wherein the logical computation expressions are referred to as operators for short” the “input model file” being the ”initial model file” as claimed) obtaining, by processing circuitry, a data sub-format of input data of the operator based on both the data format and the operator; obtaining attribute information of the processing circuitry; (Par [0031], “Preferably, the specific process of the step S23 is: inferring shape and data type information of the input and output logical tensors of the current logical operator by using the meta attributes of distribution information of each operator.”) obtaining a final model file for the processing circuitry based on compiling the initial model file according to both the attribute information and the data sub-format; (Par [0063], “…a deep learning model can be conveniently constructed to perform conversion and description between different formats of the deep learning model, iterative optimization, and flexible deployment.” And par [0076], “…parsing an input model file so as to acquire topological structure information of a neural network: parsing a user-defined function body including input and output parameters, and compiling the function body into a computation graph composed of logical computation expressions, wherein the logical computation expressions are referred to as operators for short” and par [0076], “…input and output parameters, and compiling the function body into a computation graph composed of logical computation expressions, wherein the logical computation expressions are referred to as operators for short” the “output” being the final model file as claimed and Figures 5-7) and obtaining a processing result based on running the final model file, by the processing circuitry, with data to be processed by the neural network model. (Par [0026] Preferably, the meta attributes of each operator in the logical computation graph in the step S22 refer to a mapping relationship between the input and output tensors of each logical operator and the input and output tensors of each physical operator, and the meta attributes are a list of valid attributes included in each operator, to describe a mapping relationship between the input and output tensors of one logical operator and the input and output tensors of multiple physical operators on respective devices.” And see Figures 3-4)
WANG discloses different formats, not “sub-format” as claimed)
YANG discloses the above claimed features as follows: (Par [0072], “PEF—processor-executable format—a file format suitable for configuring a configurable data processor.” And Par [0086], “The configuration file can include configuration data for the CGR array and CGR units in the CGR array, and link the computation graph to the CGR array. Execution of the configuration file by CGR processor 110 causes the CGR array (s) to implement the user algorithms and functions in the dataflow graph.” And see figures 3 and 6; The PEF processes different data formats.
Therefore, it would have been obvious to a person of ordinary skill in the art at the effective filing date to incorporate the teachings of YANG specifically inferring a different format into the method of WANG to take advantage on applying the respective acquired information to create a pertaining configuration. The modification would have been obvious because one of the ordinary skills in the art would implement obtaining and determining mappings of neural network operators and data flow to CGRP processing and/or memory elements, and/or configurations of CGRP processing and/or memory elements, to determine alternative mappings (See YANG, par [0039]).
As per Claim 2, the rejection of Claim 1 is incorporated and WANG further discloses: wherein the obtaining the data sub-format comprises: obtaining a format mapping scheme corresponding to the operator; (Par [0026], “Preferably, the meta attributes of each operator in the logical computation graph in the step S22 refer to a mapping relationship between the input and output tensors of each logical operator and the input and output tensors of each physical operator, and the meta attributes are a list of valid attributes included in each operator, to describe a mapping relationship between the input and output tensors of one logical operator and the input and output tensors of multiple physical operators on respective devices.” And Par [0063], “…a deep learning model can be conveniently constructed to perform conversion and description between different formats of the deep learning model, iterative optimization, and flexible deployment.” And see Figures 3-4) and obtaining the data sub-format of the input data of the operator based on mapping the data format to the data sub-format through the format mapping scheme. (Par [0063], “…a deep learning model can be conveniently constructed to perform conversion and description between different formats of the deep learning model, iterative optimization, and flexible deployment.” And Par [0120], “S721, constructing a mapping table of mutually exclusive memories, traversing all memories, acquiring a start point and an end point of a life cycle of each memory, reserving mutually exclusive points of two memories with overlapping life cycles, and constructing a hash table of mutually exclusive memories” and see Figures 1 and 3-4).
WANG discloses different formats, not “sub-format” as claimed)
YANG discloses the above claimed features as follows: (Par [0072], “PEF—processor-executable format—a file format suitable for configuring a configurable data processor.” And Par [0086], “The configuration file can include configuration data for the CGR array and CGR units in the CGR array, and link the computation graph to the CGR array. Execution of the configuration file by CGR processor 110 causes the CGR array (s) to implement the user algorithms and functions in the dataflow graph.” And see figures 3 and 6; The PEF processes different data formats.
Therefore, it would have been obvious to a person of ordinary skill in the art at the effective filing date to incorporate the teachings of YANG specifically inferring a different format into the method of WANG to take advantage on applying the respective acquired information to create a pertaining configuration. The modification would have been obvious because one of the ordinary skills in the art would implement obtaining and determining mappings of neural network operators and data flow to CGRP processing and/or memory elements, and/or configurations of CGRP processing and/or memory elements, to determine alternative mappings (See YANG, par [0039]).
As per Claim 3, the rejection of Claim 1 is incorporated and WANG further discloses: wherein the obtaining the data sub-format comprises: selecting a startup operator from operators; (Par [0063], “…a deep learning model can be conveniently constructed to perform conversion and description between different formats of the deep learning model, iterative optimization, and flexible deployment.” And Par [0102-0111], “…when the deep learning framework generates a distributed physical execution graph, an attribute list of legitimate distribution information allowed by each operator is firstly inferred; and an attribute with a minimum transmission overhead is selected from the attribute list as a distribution policy for this training, and is used for guiding a deep learning framework compiler to generate the most efficient execution graph. FIG. 4 shows meta attributes of each operator in the logical computation graph.” And see figures 1 and 3-5) obtaining a data sub-format of output data of the startup operator based on both the startup operator and the data format as a data sub-format of input data of the startup operator; (Par [0063], “…a deep learning model can be conveniently constructed to perform conversion and description between different formats of the deep learning model, iterative optimization, and flexible deployment.” See figures 1 and 3-5) based on a next operator of the startup operator existing in the operators, with the data sub-format of the output data of the startup operator as the data format and the next operator as the startup operator, returning to obtaining the data sub-format of output data of the startup operator; (Par [0063] and See figures 3-5) and based on no next operator of the startup operator existing in the operators, determining the data sub-format of the input data of the operator according to the data sub-format of the input data of the startup operator. (Par [0063] and See figures 3-5).
WANG discloses different formats, not “sub-format” as claimed)
YANG discloses the above claimed features as follows: (Par [0072], “PEF—processor-executable format—a file format suitable for configuring a configurable data processor.” And Par [0086], “The configuration file can include configuration data for the CGR array and CGR units in the CGR array, and link the computation graph to the CGR array. Execution of the configuration file by CGR processor 110 causes the CGR array (s) to implement the user algorithms and functions in the dataflow graph.” And see figures 3 and 6; The PEF processes different data formats.
Therefore, it would have been obvious to a person of ordinary skill in the art at the effective filing date to incorporate the teachings of YANG specifically inferring a different format into the method of WANG to take advantage on applying the respective acquired information to create a pertaining configuration. The modification would have been obvious because one of the ordinary skills in the art would implement obtaining and determining mappings of neural network operators and data flow to CGRP processing and/or memory elements, and/or configurations of CGRP processing and/or memory elements, to determine alternative mappings (See YANG, par [0039]).
As per Claim 4, the rejection of Claim 2 is incorporated and WANG further discloses: wherein the obtaining the data sub-format of input data of the operator comprises: selecting a particular operator from the operators; determining a historical operator preceding the particular operator; (Par [0063] and See figures 3-5) obtaining a data sub-format of input data of the particular operator based on both the data format and the historical operator; (Par [0063] and See figures 3-5)
and determining the data sub-format of the input data of the operator according to the data sub-format of the input data of the particular operator. (Par [0063] and See figures 3-5)
WANG discloses different formats, not “sub-format” as claimed)
YANG discloses the above claimed features as follows: (Par [0072], “PEF—processor-executable format—a file format suitable for configuring a configurable data processor.” And Par [0086], “The configuration file can include configuration data for the CGR array and CGR units in the CGR array, and link the computation graph to the CGR array. Execution of the configuration file by CGR processor 110 causes the CGR array (s) to implement the user algorithms and functions in the dataflow graph.” And see figures 3 and 6; The PEF processes different data formats.
Therefore, it would have been obvious to a person of ordinary skill in the art at the effective filing date to incorporate the teachings of YANG specifically inferring a different format into the method of WANG to take advantage on applying the respective acquired information to create a pertaining configuration. The modification would have been obvious because one of the ordinary skills in the art would implement obtaining and determining mappings of neural network operators and data flow to CGRP processing and/or memory elements, and/or configurations of CGRP processing and/or memory elements, to determine alternative mappings (See YANG, par [0039]).
As per Claim 5, the rejection of Claim 4 is incorporated and WANG further discloses: wherein the obtaining the data sub-format of the input data of the particular operator comprises: determining a conversion model parameter according to a model parameter of the historical operator; (Par [0063] and See figures 3-5)
and obtaining the data sub-format of the input data of the particular operator based on both the data format and the conversion model parameter. (Par [0063] and See figures 3-5)
WANG discloses different formats, not “sub-format” as claimed)
YANG discloses the above claimed features as follows: (Par [0072], “PEF—processor-executable format—a file format suitable for configuring a configurable data processor.” And Par [0086], “The configuration file can include configuration data for the CGR array and CGR units in the CGR array, and link the computation graph to the CGR array. Execution of the configuration file by CGR processor 110 causes the CGR array (s) to implement the user algorithms and functions in the dataflow graph.” And see figures 3 and 6; The PEF processes different data formats.
Therefore, it would have been obvious to a person of ordinary skill in the art at the effective filing date to incorporate the teachings of YANG specifically inferring a different format into the method of WANG to take advantage on applying the respective acquired information to create a pertaining configuration. The modification would have been obvious because one of the ordinary skills in the art would implement obtaining and determining mappings of neural network operators and data flow to CGRP processing and/or memory elements, and/or configurations of CGRP processing and/or memory elements, to determine alternative mappings (See YANG, par [0039]).
As per Claim 6, the rejection of Claim 1 is incorporated and WANG further discloses: wherein a plurality of processors are provided in the processing circuitry, and obtaining the final model file for the processing circuitry comprises: obtaining an original computation graph, wherein the original computation graph comprises a node that represents the operator; (Par [0026], “Preferably, the meta attributes of each operator in the logical computation graph in the step S22 refer to a mapping relationship between the input and output tensors of each logical operator and the input and output tensors of each physical operator, and the meta attributes are a list of valid attributes included in each operator, to describe a mapping relationship between the input and output tensors of one logical operator and the input and output tensors of multiple physical operators on respective devices.” And par [0032] Preferably, the specific process of the step S31 is: traversing each logical node of the logical computation graph, and generating physical computation nodes with a topological order according to the physical layout information of each logical node” and [0074], “…parsing a topological structure of an input deep learning neural network model, and a generation method and apparatus capable of converting a neural network computation-oriented topological structure into an intermediate representation constituted based on computation expressions between tensors”) partitioning, according to the attribute information, the original computation graph into a subgraph matching each processor; and obtaining a final model file of each processor based on compiling the initial model file according to the attribute information, the data sub-format, and the subgraph. . (Par [0138], “…when the deep learning framework generates a distributed physical execution graph, an attribute list of legitimate distribution information allowed by each operator is firstly inferred; and an attribute with a minimum transmission overhead is selected from the attribute list as a distribution policy for this training, and is used for guiding a deep learning framework compiler to generate the most efficient execution graph.” And par [0139], the output being the “final model file” as claimed).
As per Claim 7, the rejection of Claim 6 is incorporated and WANG further discloses: wherein the obtaining the final model file of each processor comprises: determining an initial data type of input data of the subgraph according to the original computation graph; (Par [0074], “…parsing a topological structure of an input deep learning neural network model, and a generation method and apparatus capable of converting a neural network computation-oriented topological structure into an intermediate representation constituted based on computation expressions between tensors”) determining a final data type of the input data of the subgraph according to the attribute information; and obtaining the final model file of each processor based on compiling the initial model file according to the attribute information, the data sub-format, the initial data type, and the final data type. (Par [0138], “…when the deep learning framework generates a distributed physical execution graph, an attribute list of legitimate distribution information allowed by each operator is firstly inferred; and an attribute with a minimum transmission overhead is selected from the attribute list as a distribution policy for this training, and is used for guiding a deep learning framework compiler to generate the most efficient execution graph.” And par [0139], the output being the “final model file” as claimed).
WANG discloses different formats, not “sub-format” as claimed)
YANG discloses the above claimed features as follows: (Par [0072], “PEF—processor-executable format—a file format suitable for configuring a configurable data processor.” And Par [0086], “The configuration file can include configuration data for the CGR array and CGR units in the CGR array, and link the computation graph to the CGR array. Execution of the configuration file by CGR processor 110 causes the CGR array (s) to implement the user algorithms and functions in the dataflow graph.” And see figures 3 and 6; The PEF processes different data formats.
Therefore, it would have been obvious to a person of ordinary skill in the art at the effective filing date to incorporate the teachings of YANG specifically inferring a different format into the method of WANG to take advantage on applying the respective acquired information to create a pertaining configuration. The modification would have been obvious because one of the ordinary skills in the art would implement obtaining and determining mappings of neural network operators and data flow to CGRP processing and/or memory elements, and/or configurations of CGRP processing and/or memory elements, to determine alternative mappings (See YANG, par [0039]).
As per Claim 8, the rejection of Claim 7 is incorporated and WANG further discloses: wherein the obtaining the final model file of each processor based on compiling the initial model file according to the attribute information, the data sub-format, the initial data type, and the final data type comprises: obtaining a data type conversion operator based on the initial data type being different from the final data type; (Par [0019-0020], “S62, applying for a required second-level memory from the first-level memory by each subgraph execution engine during operation according to a memory offset corresponding to a memory size required by each subgraph, wherein the second-level memory is a sub-memory of the first-level memory” and Figures 4-5)
obtaining a first adjusted subgraph based on adjusting the subgraph according to the data type conversion operator; and (Par [0019-0020], “S62, applying for a required second-level memory from the first-level memory by each subgraph execution engine during operation according to a memory offset corresponding to a memory size required by each subgraph, wherein the second-level memory is a sub-memory of the first-level memory” and Figures 4-5) obtaining the final model file of each processor based on compiling the initial model file according to the attribute information, the data sub-format, and the first adjusted subgraph. (Par [0138], “…when the deep learning framework generates a distributed physical execution graph, an attribute list of legitimate distribution information allowed by each operator is firstly inferred; and an attribute with a minimum transmission overhead is selected from the attribute list as a distribution policy for this training, and is used for guiding a deep learning framework compiler to generate the most efficient execution graph.” And par [0139], the output being the “final model file” as claimed).
WANG discloses different formats, not “sub-format” as claimed)
YANG discloses the above claimed features as follows: (Par [0072], “PEF—processor-executable format—a file format suitable for configuring a configurable data processor.” And Par [0086], “The configuration file can include configuration data for the CGR array and CGR units in the CGR array, and link the computation graph to the CGR array. Execution of the configuration file by CGR processor 110 causes the CGR array (s) to implement the user algorithms and functions in the dataflow graph.” And see figures 3 and 6; The PEF processes different data formats.
Therefore, it would have been obvious to a person of ordinary skill in the art at the effective filing date to incorporate the teachings of YANG specifically inferring a different format into the method of WANG to take advantage on applying the respective acquired information to create a pertaining configuration. The modification would have been obvious because one of the ordinary skills in the art would implement obtaining and determining mappings of neural network operators and data flow to CGRP processing and/or memory elements, and/or configurations of CGRP processing and/or memory elements, to determine alternative mappings (See YANG, par [0039]).
As per Claim 9, the rejection of Claim 8 is incorporated and WANG further discloses: wherein the obtaining the final model file of each processor based on compiling the initial model file according to the attribute information, the data sub-format, and the first adjusted subgraph comprises: determining a type of an operator in the subgraph; (Par [0019-0020], “S62, applying for a required second-level memory from the first-level memory by each subgraph execution engine during operation according to a memory offset corresponding to a memory size required by each subgraph, wherein the second-level memory is a sub-memory of the first-level memory;”) obtaining a second adjusted subgraph based on adjusting the operator in the subgraph based on the type of the operator in the subgraph being a preset type; (Par [0062]-[0063], “a first-level memory is a one-time allocated memory before the physical computation graph generation unit constructs and infers the physical computation graph; a second-level memory is a required memory allocated from the first-level memory by a subgraph execution engine of the physical computation graph during operation; and a third-level memory is a memory that is allocated from the second-level memory and required by the input and output tensor edges of each computation node inside a subgraph of the physical computation graph.”)
and obtaining the final model file of each processor based on compiling the initial model file according to the attribute information, the data sub-format, and the second adjusted subgraph. (Par [0138], “…when the deep learning framework generates a distributed physical execution graph, an attribute list of legitimate distribution information allowed by each operator is firstly inferred; and an attribute with a minimum transmission overhead is selected from the attribute list as a distribution policy for this training, and is used for guiding a deep learning framework compiler to generate the most efficient execution graph.” And par [0139], the output being the “final model file” as claimed).
WANG discloses different formats, not “sub-format” as claimed)
YANG discloses the above claimed features as follows: (Par [0072], “PEF—processor-executable format—a file format suitable for configuring a configurable data processor.” And Par [0086], “The configuration file can include configuration data for the CGR array and CGR units in the CGR array, and link the computation graph to the CGR array. Execution of the configuration file by CGR processor 110 causes the CGR array (s) to implement the user algorithms and functions in the dataflow graph.” And see figures 3 and 6; The PEF processes different data formats.
Therefore, it would have been obvious to a person of ordinary skill in the art at the effective filing date to incorporate the teachings of YANG specifically inferring a different format into the method of WANG to take advantage on applying the respective acquired information to create a pertaining configuration. The modification would have been obvious because one of the ordinary skills in the art would implement obtaining and determining mappings of neural network operators and data flow to CGRP processing and/or memory elements, and/or configurations of CGRP processing and/or memory elements, to determine alternative mappings (See YANG, par [0039]).
As per Claim 10, the rejection of Claim 1 is incorporated and WANG further discloses: wherein the obtaining the processing result comprises: determining a data format of the data to be processed of the neural network model; (Par [0063], “…a deep learning model can be conveniently constructed to perform conversion and description between different formats of the deep learning model, iterative optimization, and flexible deployment.”) obtaining adjusted data to be processed based on adjusting the data format of the data to be processed based on the data format of the data to be processed not matching the data sub-format in the final model file; and obtaining the processing result based on running the final model file with the adjusted data to be processed. (Par [0138], “…when the deep learning framework generates a distributed physical execution graph, an attribute list of legitimate distribution information allowed by each operator is firstly inferred; and an attribute with a minimum transmission overhead is selected from the attribute list as a distribution policy for this training, and is used for guiding a deep learning framework compiler to generate the most efficient execution graph.” And par [0139], the output being the “final model file” as claimed).
As per Claim 11, the rejection of Claim 10 is incorporated and WANG further discloses: wherein the obtaining the processing result comprises: obtaining a candidate processing result of the data to be processed based on running the final model file with the data to be processed; (Par [0104-0107], “In the step S3, the physical computation graph describes the full view of the entire runtime computation graph, and the process of constructing the physical computation graph is a process of generating the physical…” and see Figures 3-5) determining an output data format of the candidate processing result; and obtaining the processing result based on adjusting the output data format of the candidate processing result according to the final model file. (Par [0138], “…when the deep learning framework generates a distributed physical execution graph, an attribute list of legitimate distribution information allowed by each operator is firstly inferred; and an attribute with a minimum transmission overhead is selected from the attribute list as a distribution policy for this training, and is used for guiding a deep learning framework compiler to generate the most efficient execution graph.” And par [0139], the output being the “final model file” as claimed).
As per Claims 12-20, being the apparatus and non-transitory medium claims corresponding to the system claims 1-11 respectively and rejected under the same reason set forth in connection of the rejections of Claims 1-11 and further WANG discloses: (Par [0178]).
Conclusion
13. The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure.
Stevens; Luis F. (US-20210157858-A1) relates to a neural network model to be quickly trained to perform as accurately as possible given the subject domain, image characteristics, and/or types of images analyzed by the neural network. In some implementations, the training method employs two or more training epochs, wherein in each epoch a neural network model is trained to recognize at least a predefined percentage of images in a training set with subject characteristics that are more likely than not to belong to predefined classes with respect to that subject characteristic.
Ren; Bin (US-20220413862-A1), relates to describe a method for accelerating deep neural networks comprising the steps of: classifying operators into one of five high-level abstract types; performing a mapping type analysis for each paired input and input; classifying mapping types into one of three classes; providing a computational graph; rewriting said computational graph; and generating and optimizing the fusion code.
SHAN; GANG (US-20250156159-A1), relates to neural networks, discontinuity of data in memory is often caused by view-class operators, this disclosure proposes a scheme of constructing a view-class operator subgraph based on view-class operators in a computing graph. This view-class operator subgraph may then support the subsequent efficient moving of memory data from discontiguous to contiguous.
Farkash; Ariel ( US-12254110-B2), relate to DNN ecosystem ONNX (Open Neural Network Exchange), and finds that the mapping relation between (each) input and output of each operator is critical to determine both the profitability and correct implementation of fusion optimization. Moreover, it is possible for us to classify all operators into five high-level abstract types based on the relationship between input elements and output elements.
14. Any inquiry concerning this communication or earlier communications from the examiner should be directed to ANGELICA RUIZ whose telephone number is (571)270-3158. The examiner can normally be reached M-F 10:00 am to 6:00 pm.
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, Boris Gorney can be reached at (571) 270-5626. 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.
/ANGELICA RUIZ/Primary Examiner, Art Unit 2154 April 3, 2026
Prosecution Insights