DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claims 1-20 are presented for examination.
Response to Amendment
Applicant’s amendment has obviated many, but not all, of the objections to the specification, drawings, and claims given in the last Office action. To the extent that an objection and/or rejection appears in both this Office action and the previous action, that objection and/or rejection is maintained. To the extent that the objection and/or rejection appears only in the previous Office action, that objection and/or rejection is withdrawn.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on July 24, 2026 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Specification
The lengthy specification has not been checked to the extent necessary to determine the presence of all possible minor errors. Applicant’s cooperation is requested in correcting any errors of which applicant may become aware in the specification.
The disclosure is objected to because of the following informalities: “in response to that … is” in paragraphs 7-8 and 15-16 should be “in response to … being” and “in response to that … correspond” in paragraph 155 should be “in response to … corresponding”. Also, the appropriate article should precede “current test” in paragraphs 5-6, 8, 13-14, 16, 22, 24, 62, 83, 85-87, 103, 120, 155, 157, 182, and 184.
Appropriate correction is required.
Claim Objections
Claims 2-3 are objected to because of the following informalities: “before” should be “wherein before”.
Claim 8 is objected to because of the following informalities: the meaning of the abbreviation “TNN” should be spelled out.
Claims 4, 13, and 18 are objected to because of the following informalities: “packeged” (two instances per claim) should be “packaged”.
Claims 5, 14, and 19 are objected to because of the following informalities: “current test” should read “a current test” for the first recitation and “the current test” in subsequent recitations.
Appropriate correction is required.
Claim Rejections - 35 USC § 103
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claims 1, 7, and 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over Cui et al. (US 20200042362) (“Cui”) in view of Verma et al. (US 20220067450) (“Verma”).
Regarding claim 1, Cui discloses “[a] method for testing a deep learning model, the method being applied to an edge device (Cui Fig. 6, client systems 610), and the method comprising:
acquiring, by the edge device, a deep learning model to be deployed from a local server or a cloud server (self-adaptive batch dataset partitioning control method optimizes load balancing among a set of accelerator resources for a distributed deep learning model task [i.e., a deep learning model is acquired] – Cui, paragraph 43; see also Fig. 4, ref. char. 400 (disclosing that a hybrid set of accelerator devices is provisioned for a distributed deep learning model training process by the worker nodes [edge devices which receive the model]), Fig. 1 (showing that these accelerators are each located on a worker node and that the deep learning model is initially stored on a deep learning computing platform [cloud server]));
acquiring, by the edge device, an acceleration instruction specified by a user (service request received from client systems [edge devices] can include user-specified conditions and demands for executing a given job; for example, a service request may specify a specific type/model of accelerator device, a desired number of accelerator devices, etc. – Cui, paragraph 73), and accelerating, by the edge device, the deep learning model according to an acceleration method corresponding to the acceleration instruction (initial default job partition ratio can be set to some non-equal job partition ratios achieved for the same or similar sets of hybrid accelerator resources, or based on the relative performance (e.g., operating speeds) of the accelerator resources [i.e., the acceleration of the model using the chosen partition improves an inference speed] – Cui, paragraph 55; resource scheduling and provisioning module schedules and provisions computer resources for jobs pending in the request queue; service request can include user-specified conditions and demands for executing a given job; for example, a service request may specify a specific type/model of accelerator device, a desired number of accelerator devices, etc. – id. at paragraph 73 [i.e., the execution of the acceleration is based on the user preferences]; see also Fig. 1 (showing that the accelerator devices are located on the worker nodes [edge devices]));
acquiring test samples corresponding to the deep learning model (model validation module implements methods that are configured to validate the deep learning model using a validation dataset [test samples] – Cui, paragraph 43); and
testing the deep learning model by using the test samples (model validation module implements methods that are configured to validate [test] the deep learning model using a validation dataset [test samples] – Cui, paragraph 43).”
Cui appears not to disclose explicitly the further limitations of the claim. However, Verma discloses “acquiring, by the edge device, test samples (edge node uses [after acquiring] synthetic training data [test samples] that have been regenerated and reverse testing to evaluate model performance of the trained model at the edge node – Verma, paragraph 29) …; and
testing, by the edge device, the … model (reverse testing of the trained model is performed at the edge node – Verma, paragraph 50) ….”
Verma and the instant application both relate to machine learning and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Cui to test the model at the edge device, as disclosed by Verma, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the model to be tested without requiring the storage of large amounts of labeled training data at the edge node. See Verma, paragraph 50.
Claim 9 is an apparatus claim corresponding to method claim 1 and is rejected for the same reasons as given in the rejection of that claim. Similarly, claim 10 is a computer storage medium claim corresponding to method claim 1 and is rejected for the same reasons as given in the rejection of that claim.
Regarding claim 7, Cui discloses that “after testing the deep learning model by using the test samples, the method further comprises:
generating a test report according to test data obtained in the testing (model validation module provides an estimate [test report] of performance metrics [test data obtained in the testing] of the deep learning module, e.g., an unbiased evaluation of the accuracy of a model fit on the training dataset using the validation dataset – Cui, paragraph 43).”
Claims 2, 11, and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Cui in view of Verma and further in view of Samie et al., “Fast Operation Mode Selection for Highly Efficient IoT Edge Devices,” in 39.3 IEEE Transactions on Computer-Aided Design of Integrated Circuits and Sys. 572-84 (2019) (“Samie”).
Regarding claim 2, the rejection of claim 1 is incorporated. Cui further discloses “acceleration methods” and a “device used for [a] current test of the deep learning model”, as shown in the rejection of claim 1.
Cui/Verma appears not to disclose explicitly the further limitations of the claim. However, Samie discloses that “before accelerating the deep learning model, the method further comprises:
selecting, in response to a plurality of [operation] methods corresponding to the [operation] instruction, one of the plurality of [operation] methods meeting a preset performance index according to a system type and hardware performance of the edge device (operation mode (i.e., combination of computation offloading level and service quality level) [plurality of operation methods] of IoT devices [edge devices] must be determined and adapted at runtime such that the requirements of devices [hardware performance of edge device] are met, the shared limited resources of the gateway are wisely allocated, and the efficiency of the system [system type] is improved; since optimizing the overall efficiency might lead to poor efficiency in some IoT devices, the ‘max-min’ optimization goal [preset performance index] is considered – Samie, p. 2, paragraph spanning both columns and following paragraph) ….”
Samie and the instant application both relate to mode selection for edge devices and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Cui/Verma to select an operation mode of the edge device according to system type and hardware performance, as disclosed by Samie, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would increase the efficiency of the computation. See Samie, p. 2, paragraph spanning both columns and following paragraph.
Claim 11 is an apparatus claim corresponding to method claim 2 and is rejected for the same reasons as given in the rejection of that claim. Similarly, claim 16 is a computer storage medium claim corresponding to method claim 2 and is rejected for the same reasons as given in the rejection of that claim.
Claims 3, 12, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Cui in view of Verma and further in view of Martinez Canedo et al. (WO 2020027852) (“Martinez”).
Regarding claim 3, Cui/Verma appears not to disclose explicitly the further limitations of the claim. However, Martinez discloses that “before testing the deep learning model by using the test samples, the method further comprises:
determining a compiler according to a system type of the edge device used for current test of the deep learning model (neural code is compiled into a target framework by selecting the appropriate neural compiler; the training or testing [current test] process for neural networks [deep learning model] may be performed in the host system (e.g., laptop [edge device] or the cloud) [i.e., the compilation takes into consideration the fact that the code will be executed on an edge device] – Martinez, paragraphs 32-33); and
compiling and packaging, by the compiler, algorithmic code corresponding to the deep learning model into a library (for neural code, compilation results in compiled neural code [algorithmic code corresponding to the deep learning model]; compiled neural code is deployed to neural co-processor [i.e., packaged into a library for use by said co-processor] – Martinez, paragraphs 32-33).”
Martinez and the instant application both relate to compilation of deep learning code and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Cui/Verma to determine a compiler according to the system type of the edge device, as disclosed by Martinez, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the deep learning code to be read properly by the processor that executes it. See Martinez, paragraphs 32-33.
Claim 12 is an apparatus claim corresponding to method claim 3 and is rejected for the same reasons as given in the rejection of that claim. Similarly, claim 17 is a computer storage medium claim corresponding to method claim 3 and is rejected for the same reasons as given in the rejection of that claim.
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Cui in view of Martinez and Verma and further in view of Liu et al. (US 20200250585) (“Liu”). Claims 13 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Cui in view of Martinez and Verma and further in view of Liu and Skaljak (US 20220051093) (“Skaljak”).
Regarding claim 4, neither Cui, Verma, nor Martinez appears to disclose explicitly the further limitations of the claim. However, Liu discloses that “a type of the packaged library is determined by:
determining, in response to the compiler being one of a GNU Compiler Collection compiler, a G++ compiler and a cross compiler, the type of the [packaged] library as a Shared Object (SO) library (runtime library is a special computer program library that is used by the compiler to implement a programming language built-in function to provide runtime support for the language program, for example, a .so library [SO library] compiled by GCC/LLVM – Liu, paragraph 27) ….”
Liu and the instant application both relate to machine learning and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Cui, Verma, and Martinez to determine that the library is an SO library in response to the compiler being a GCC compiler, as disclosed by Liu, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the system to provide runtime support for the language program. See Liu, paragraph 27.
Neither Cui, Verma, Martinez, nor Liu appears to disclose explicitly the further limitations of the claim. However, Skaljak discloses “determining, in response to the compiler being another compiler different from the GNU Compiler Collection compiler, the G++ compiler or the cross compiler, the type of the [packaged] library as a Dynamic-Link Library (DLL) (software may be compiled with a plugin (e.g., as a DLL) [note that, since it is not stated that the compiler is a GCC, G++, or cross compiler, it may reasonably be inferred that in general the compiler may be another one of these options] – Skaljak, paragraph 73)1.”
Skaljak and the instant application both relate to machine learning and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Cui, Verma, Martinez, and Liu to determine that the code should be compiled as a DLL, as disclosed by Skaljak, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would ensure that the compiled code is compatible with WINDOWS environments . See Skaljak, paragraph 73.
Claim 13 is an apparatus claim corresponding to method claim 4 and is rejected for the same reasons as given in the rejection of that claim, except as noted above. Similarly, claim 18 is a computer storage medium claim corresponding to method claim 4 and is rejected for the same reasons as given in the rejection of that claim, except as noted above.
Claims 5, 14, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Cui in view of Martinez and Verma and further in view of Drego et al. (US 20200257467) (“Drego”).
Regarding claim 5, neither Cui, Verma, nor Martinez appears to disclose explicitly the further limitations of the claim. However, Drego discloses that “determining the compiler comprises one of:
determining that the compiler is one of a GNU Compiler Collection compiler, a G++ compiler or a cross compiler in response to a first operating system being used for [the] current test (instructions generator may be implement3ed using one or more general purpose computers (e.g., a LINUX computer); compiler module may be implemented using any suitable compiler software (e.g., a GNU Compiler Collection (GCC)) – Drego, paragraph 56);
determining that the compiler is one of a GNU Compiler Collection compiler, a G++ compiler or a cross compiler in response to that a second operating system being used for [the] current test;
determining that the compiler is one of a GNU Compiler Collection compiler, a G++ compiler or a cross compiler in response to a third operating system being used for [the] current test; and
determining that the compiler is another compiler different from the GNU Compiler Collection compiler, the G++ compiler or the cross compiler in response to a fourth operating system being used for [the] current test.”
Drego and the instant application both relate to machine learning and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Cui, Verma, and Martinez to use a GCC compiler in response to determining that a LINUX system is being used, as disclosed by Drego, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would enable the system to generate computation instructions, execution instructions, and/or data movement instructions. See Drego, paragraph 56.
Claim 14 is an apparatus claim corresponding to method claim 5 and is rejected for the same reasons as given in the rejection of that claim. Similarly, claim 19 is a computer storage medium claim corresponding to method claim 5 and is rejected for the same reasons as given in the rejection of that claim.
Claims 6, 15, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Cui in view of Martinez and Verma and further in view of Barrett et al. (US 20070299768) (“Barrett”).
Regarding claim 6, neither Cui, Verma, nor Martinez appears to disclose explicitly the further limitations of the claim. However, Barrett discloses that “after compiling and packaging, by the compiler, algorithmic code corresponding to the deep learning model into a library, the method further comprises:
encapsulating at least one preset function library into the library, wherein the preset function library is configured to realize one or more of an authentication function, an encryption function and a network function (within a programming environment such as MICROSOFT’s Visual Studio.NET, a new iexplore.exe process can be created and a connection opened using shdocvw.dll, where shdocvw.dll is a library[preset function library] used by WINDOWS applications to add basic file and networking operations [network functions], and is encapsulated in a .NET InternetExplorer object [library] – Barrett, paragraph 107).”
Barrett and the instant application both relate to computer networking libraries and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Cui, Verma, and Martinez to encapsulate a library performing networking functions into the system, as disclosed by Barrett, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would enhance the ability of the system to perform networking operations. See Barrett, paragraph 107.
Claim 15 is an apparatus claim corresponding to method claim 6 and is rejected for the same reasons as given in the rejection of that claim. Similarly, claim 20 is a computer storage medium claim corresponding to method claim 6 and is rejected for the same reasons as given in the rejection of that claim.
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Cui in view of Verma and further in view of Wang et al. (US 20210256384) (“Wang”).
Regarding claim 8, Cui/Verma appears not to disclose explicitly the further limitations of the claim. However, Wang discloses that “the acceleration method comprises one or more of:
a mobile neural network (MNN);
an inference framework (TNN); and
a neural network inference engine (Tengine-Lite) (in recent years, there have been intensive efforts in DNN inference acceleration frameworks targeting mobile devices, including Mobile Neural Network (MNN) – Wang, paragraph 42).”
Wang and the instant application both relate to machine learning acceleration on edge devices and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Cui/Verma to have used MNN as the acceleration framework, as disclosed by Wang, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the networks to be run on lightweight edge devices, thereby reducing the need for communication with a central server. See Wang, paragraph 42.
Response to Arguments
Applicant’s arguments with respect to the claims have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. Specifically, Applicant’s argument that Cui does not teach every operation occurring at the edge device is rendered moot by the use of newly cited reference Verma to teach certain operations occurring at the edge device.
Conclusion
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 RYAN C VAUGHN whose telephone number is (571)272-4849. The examiner can normally be reached M-R 7:00a-5:00p ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Kamran Afshar, can be reached at 571-272-7796. 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.
/RYAN C VAUGHN/ Primary Examiner, Art Unit 2125
1 Note that claim 4 is a method claim and that the two options are mutually exclusive contingent limitations (i.e., determining that the compiler is one of a GCC, G++, or cross compiler and determining that the compiler is another compiler). Therefore, the broadest reasonable interpretation of claim 4 in light of the specification requires only one of these two options. MPEP § 2111.04(II). However, claims 13 and 18, being apparatus and storage medium claims respectively, have narrower broadest reasonable interpretations because they require hardware capable of performing all of the claimed functions. Id.
Prosecution Insights