Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is responsive to the Amendment filed on 7/10/2026.
Claims 1-20 are pending in the case.
Response to Arguments
Applicant's arguments and amendments with regards to the 35 U.S.C. § 103 rejection of claim(s) 1-20 have been considered, but are not persuasive.
(A) Regarding claim 1, Applicant argues that Xie and Miret fail to disclose “identifying the structure as a removable structure based on the structure belonging to a minimal group of operators and associated parameters that are jointly removable from the model such that the model remains valid when the minimal group is removed”.
Examiner respectfully disagrees.
Miret [17, 55, 57-59, 65, 66, 71, 81, 88] discloses for compressing model-layers may be grouped together based on them being for consecutive layers, removing filter for one layer in a group may mean removing the filter for subsequent layers due to removed features based on pruning filter in previous layer (which make the filter in the subsequent layer redundant) and saves additional operations, weights for selected filters may be set to zero, compressed model may be validated to check that model is valid after compression,
and therefore sufficiently teaches identifying the structure as a removable structure based on the structure belonging to a minimal group of operators and associated parameters that are jointly removable from the model such that the model remains valid when the minimal group is removed.
(B) Applicant's arguments regards to the 35 U.S.C. § 103 rejection of claim(s) 3 and 20 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.
(C) Applicant further argues “None of the cited references, alone or in combination, teach or suggest the minimal group and group-sparsity limitations now recited in the amended claims. The Examiner's stated motivations to combine relate to user-provided compression requests (Fan), compressing untrained models (Fusi), and improving pruning efficiency (Miret). See Office Action, pages 4-5, 8. These motivations do not address the amended limitations. Moreover, Miret's dependency- based grouping where removal of one node necessitates reconfiguration of dependent nodes is inconsistent with the claimed minimal groups that are jointly removable such that the model remains valid. Accordingly, Applicant respectfully submits that claims 1, 8, and 18 are patentable over the cited references, and withdrawal of the § 103 rejections is respectfully requested”
Examiner respectfully disagrees. Applicant is directed to the responses to arguments in 4(A) and 4(B).
(D) Regarding claim(s) 2-7, 9-17, 19, and 20, applicant argues that the claims are allowable for reasons similar to those references with regards to claims 1, 8 and 18.
For reasons similar to those discussed above with regards to claims 1, 8 and 18, examiner asserts that claim(s) 2-7, 9-17, 19, and 20 are unpatentable over Xie in view of Fan, Fusi and Miret.
Claim Rejections - 35 USC § 103
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.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-20, are rejected under 35 U.S.C. 103 as being unpatentable over Xie (US 20190370658 A1), in view of Fan (US 20240232686 A1), Fusi (US 20220108168 A1) and Miret (US 20220092425 A1).
Regarding claim 1, Xie teaches a computing system comprising: a processor; and memory storing instructions that, when executed by the processor, cause the processor to perform acts comprising (Xie [6] device processor executes code stored in memory to perform operations):
obtaining ... computer-implemented model that is to be ... compressed, wherein the ... computer-implemented model includes an operator that comprises a structure (Xie [12, 14, 18] input components may include deep neural network, deep neural network may have batch normalization layers for functions such as scaling and adding bias to features (operator with structure));
obtaining training data that is to be employed to train the ... computer-implemented model (Xie [14] input can include training and validation dataset); and
based upon ... request ... compressing the ... computer-implemented model to generate a trained and compressed computer-implemented model, wherein the trained and compressed computer-implemented model fails to include the structure (Xie [14, 18, 21] end-user customized performance metric (request) may be received as input, batch normalization layer(s) may be removed (not included) from model to generate compressed model that is trained).
Xie does not specifically teach an untrained computer-implemented model that is to be trained; receiving a request from a user to train and compress the untrained computer-implemented model; identifying the structure as a removable structure based on the structure belonging to a minimal group of operators and associated parameters that are jointly removable from the model such that the model remains valid when the minimal group is removed.
However Fan teaches receiving a request from a user to train and compress the ... computer-implemented model; based upon the request and the training data, and without further input from the user, training and compressing the ... computer-implemented model to generate a trained and compressed computer-implemented model (Fan [6, 41, 73, 75, 77] user may provide data descriptive of a selection of one or more candidate compression schemes (request) for compressing model, model may be compressed based on request and trained based on training data).
It would have been obvious to one of an ordinary skill in the art before the effective filing date of the claimed invention, to have incorporated the concept taught by Fan of receiving a request from a user to train and compress the ... computer-implemented model; based upon the request and the training data, and without further input from the user, training and compressing the ... computer-implemented model to generate a trained and compressed computer-implemented model, into the invention suggested by Xie; since both inventions are directed towards generating compressed computer-implemented models, and incorporating the teaching of Fan into the invention suggested by Xie would provide the added advantage of allowing data descriptive of a selection of one or more candidate compression schemes to be used in compressing a model- to be provided by a user, and the combination would perform with a reasonable expectation of success (Fan [6, 41, 73, 75, 77]).
Xie and Fan does not specifically teach an untrained computer-implemented model that is to be trained; identifying the structure as a removable structure based on the structure belonging to a minimal group of operators and associated parameters that are jointly removable from the model such that the model remains valid when the minimal group is removed.
However Fusi teaches an untrained computer-implemented model that is to be trained [Fusi [28, 29] model to be compressed may be untrained, Fusi [16, 17, 37] compressing a model before training is useful when model may be in a form that renders it difficult or impossible to train, as the computational requirements associated with the model may be greater than the computational resources that are available);
obtaining training data that is to be employed to train the untrained computer-implemented model (Fusi [32] training data may be received);
based upon ...a... request and the training data, ... training and compressing the untrained computer-implemented model to generate a trained and compressed computer-implemented model [Fusi [41, 44] based on compression indication (request) and training data, untrained model is compressed and compressed model is trained).
It would have been obvious to one of an ordinary skill in the art before the effective filing date of the claimed invention, to have incorporated the concept taught by Fusi of an untrained computer-implemented model that is to be trained; obtaining training data that is to be employed to train the untrained computer-implemented model; based upon ...a... request and the training data, ... training and compressing the untrained computer-implemented model to generate a trained and compressed computer-implemented model, into the invention suggested by Xie and Fan; since both inventions are directed towards generating compressed computer-implemented models, and incorporating the teaching of Fusi into the invention suggested by Xie and Fan would provide the added advantage of being useful when model may be in a form that renders it difficult or impossible to train, as the computational requirements associated with the model may be greater than the computational resources that are available, and the combination would perform with a reasonable expectation of success (Fusi [28, 29, 32, 41, 44]).
Xie, Fan and Fusi does not specifically teach identifying the structure as a removable structure based on the structure belonging to a minimal group of operators and associated parameters that are jointly removable from the model such that the model remains valid when the minimal group is removed.
However Miret teaches identifying the structure as a removable structure based on the structure belonging to a minimal group of operators and associated parameters that are jointly removable from the model such that the model remains valid when the minimal group is removed (Miret [17, 55, 57-59, 65, 66, 71, 81, 88] for compressing model-layers may be grouped together based on them being for consecutive layers, removing filter for one layer in a group may mean removing the filter for subsequent layers due to removed features based on pruning filter in previous layer (which make the filter in the subsequent layer redundant) and saves additional operations, weights for selected filters may be set to zero, compressed model may be validated to check that model is valid after compression).
It would have been obvious to one of an ordinary skill in the art before the effective filing date of the claimed invention, to have incorporated the concept taught by Miret of identifying the structure as a removable structure based on the structure belonging to a minimal group of operators and associated parameters that are jointly removable from the model such that the model remains valid when the minimal group is removed, into the invention suggested by Xie, Fan and Fusi; since both inventions are directed towards generating compressed computer-implemented models, and incorporating the teaching of Miret into the invention suggested by Xie, Fan and Fusi would provide the added advantage of saving additional operations and ensuring the compression techniques applied to the model continue to result in model(s) that pass validation, and the combination would perform with a reasonable expectation of success (Miret [17, 55, 57-59, 65, 66, 71, 81, 88]).
Regarding claim 2, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 1 above.
Xie further teaches wherein the operator is ... a batch normalization function ... (Xie[17] operator whose structure is removed for compression, may be a batch normalization function).
Regarding claim 3, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 1 above.
Xie does not specifically teach wherein training and compressing the untrained computer-implemented model are performed during a single optimization process that enforces group-level sparsity over identified groups of operators and associated parameters such that removal of the structure is determined based on sparsity of the groups
However Fusi teaches wherein training and compressing the untrained computer-implemented model are performed during a single optimization process (Fusi [29, 37, 38, 40, 41] compressing includes factorizing model, untrained model is compressed and trained using optimizer (process of the method 200 includes optimization (single optimization process)).
Xie does not specifically teach … process that enforces group-level sparsity over identified groups of operators and associated parameters such that removal of the structure is determined based on sparsity of the groups
However Miret teaches wherein training and compressing the … computer-implemented model are performed during a single optimization process that enforces group-level sparsity over identified groups of operators and associated parameters such that removal of the structure is determined based on sparsity of the groups (Miret [17, 56, 65, 66, 68, 69] model is trained, compression of model may involve a process to achieve desired sparsity level for a group, and setting parameters and removing operator(s) for a group based on sparsity).
Regarding claim 4, Fan, Fusi and Miret teach the invention as claimed in claim 1 above.
Xie does not specifically teach wherein identifying the structure as the removable structure comprises: constructing a trace graph of the computer-implemented model, where the trace graph comprises: vertices that represent operators in the computer-implemented model, where the vertices comprise a first vertex that represents the operator and a second vertex that represents a second operator; and edges that represent connections between the operators, where the edges comprise an edge between the operator and the second operator; assigning a category from amongst several potential categories to the first vertex, where the category is assigned to the first vertex based upon a parameter of the operator, and further wherein the structure is identified as the removable structure based upon the category assigned to the first vertex.
However Miret teaches wherein identifying the structure as the removable structure comprises: constructing a trace graph of the computer-implemented model, where the trace graph comprises: vertices that represent operators in the computer-implemented model, where the vertices comprise a first vertex that represents the operator and a second vertex that represents a second operator; and edges that represent connections between the operators, where the edges comprise an edge between the operator and the second operator (Miret [17, 54, 65] filters to be pruned (structures to be removed) from a neural network for compression are determined from graph generated from model (construction), graph is generated may have nodes (vertices) for activations or layers (operator) and edges between nodes may be connections or dependencies);
assigning a category from amongst several potential categories to the first vertex, where the category is assigned to the first vertex based upon a parameter of the operator, and further wherein the structure is identified as the removable structure based upon the category assigned to the first vertex (Miret [63, 65, 66, 76] nodes may be assigned to groups and or features (category), based on group characteristics- nodes may be identified for pruning (removal) for compressing models, Miret [17, 18] using a graph and group-wise pruning improves efficiency of pruning).
Regarding claim 5, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 4 above. Xie does not specifically teach wherein identifying the structure as the removable structure comprises identifying the structure as being the removable structure based upon the category assigned to the first vertex, where the removable structure belongs to a class of minimal structures that are able to be removed from the computer-implemented model without impacting output of the computer-implemented model when parameters of the minimal structures are zero
However Miret teaches wherein identifying the structure as the removable structure comprises identifying the structure as being the removable structure based upon the category assigned to the first vertex, where the removable structure belongs to a class of minimal structures that are able to be removed from the computer-implemented model without impacting output of the computer-implemented model when parameters of the minimal structures are zero (Miret [63, 65, 66, 76] nodes may be assigned to groups and or features (category), based on group characteristics- nodes may be identified for pruning (removal) for compressing models, Miret [17, 18] using a graph and group-wise pruning improves efficiency of pruning, Miret [55, 57, 58] nodes to be pruned are selected so that they do not impact model output (negligible or acceptable change in accuracy, precision and/or recall).
Regarding claim 6, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 4 above. Xie does not specifically teach where training and compressing the computer-implemented model to generate the trained and compressed computer-implemented model further comprises: identifying the removable structure as being redundant with another removable structure in the computer-implemented model.
However Miret teaches where training and compressing the computer-implemented model to generate the trained and compressed computer-implemented model further comprises: identifying the removable structure as being redundant with another removable structure in the computer-implemented model (Miret [66, 81] based on determining a node is dependent on second node (and therefore redundant if the second node is removed)- they may be grouped together so that removing the second node will also remove the first node too, for compressing model-layers may be grouped together based on them being for consecutive layers, removing filter for one layer in a group may mean removing the filter for subsequent layers due to removed features based on pruning filter in previous layer (which make the filter in the subsequent layer redundant).
Regarding claim 7, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 4 above. Xie does not specifically teach removing the removable structure from the computer-implemented model based upon the removable structure being identified as being redundant with the another removable structure
However Miret teaches removing the removable structure from the computer-implemented model based upon the removable structure being identified as being redundant with the another removable structure (Miret [66] based on determining a node is dependent on second node (and therefore redundant if the second node is removed)- they may be grouped together so that removing the second node will also remove the first node too).
Claim 8 is directed towards a method performing instructions similar in scope to the instructions executed by the system of claim 4, and is rejected under the same rationale.
Xie further teaches a method performed by a computing system, ... wherein the computer-implemented model is a ... computer-implemented deep neural network (DNN) (Xie [6, 3, 12] device processor executes code stored in memory to perform operations for a model which can be a deep neural network).
Regarding claim 9, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 8 above. Xie does not specifically teach using the DNN to generate output without fine-tuning the DNN.
However Miret teaches using the DNN to generate output without fine-tuning the DNN (Miret [56, 57] DNN fine-tuning does not necessarily need to be performed, compressed DNN output may be generated).
Regarding claim 10, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 8 above.
Xie further teaches wherein the structure corresponds to at least one of an activation operator, a convolution operator, a batch normalization operator (Xie[17] operator whose structure is removed for compression, may be a batch normalization function).
Regarding claim 11, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 8 above. Xie does not specifically teach wherein the graph comprises: vertices that represent operators in the computer-implemented model, where the vertices comprise a first vertex that represents a first operator and a second vertex that represents a second operator, and further where the first operator comprises the structure; and edges that represent connections between the operators, where the edges comprise an edge between the first operator and the second operator; wherein training and compressing the untrained computer-implemented DNN to generate a trained and compressed DNN comprises assigning a category from amongst several potential categories to the first vertex, where the category is assigned to the first vertex based upon a parameter of the first operator, and further where the structure is identified as being removable from the computer-implemented DNN based upon the category assigned to the first vertex
However Miret teaches wherein the graph comprises: vertices that represent operators in the computer-implemented model, where the vertices comprise a first vertex that represents a first operator and a second vertex that represents a second operator, and further where the first operator comprises the structure; and edges that represent connections between the operators, where the edges comprise an edge between the first operator and the second operator(Miret [17, 54, 65] filters to be pruned (structures to be removed) from a neural network for compression are determined from graph generated from model (construction), graph is generated may have nodes (vertices) for activations or layers (operator) and edges between nodes may be connections or dependencies);
wherein training and compressing the untrained computer-implemented DNN to generate a trained and compressed DNN comprises assigning a category from amongst several potential categories to the first vertex, where the category is assigned to the first vertex based upon a parameter of the first operator, and further where the structure is identified as being removable from the computer-implemented DNN based upon the category assigned to the first vertex (Miret [56, 63, 65, 66, 76] nodes may be assigned to groups and or features (category), based on group characteristics- nodes may be identified for pruning (removal) for compressing models which may be trained, Miret [17, 18] using a graph and group-wise pruning improves efficiency of pruning).
Regarding claim 12, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 11 above. Xie does not specifically teach identifying the structure as removable from the computer-implemented DNN based upon the category assigned to the first vertex, where the structure belongs to a class of minimal structures that are removable from the computer-implemented DNN without impacting output of the computer-implemented DNN when parameters of the minimal structures are zero
However Miret teaches identifying the structure as removable from the computer-implemented DNN based upon the category assigned to the first vertex, where the structure belongs to a class of minimal structures that are removable from the computer-implemented DNN without impacting output of the computer-implemented DNN when parameters of the minimal structures are zero (Miret [63, 65, 66, 76] nodes may be assigned to groups and or features (category), based on group characteristics- nodes may be identified for pruning (removal) for compressing models, Miret [17, 18] using a graph and group-wise pruning improves efficiency of pruning, Miret [55, 57, 58] nodes to be pruned are selected so that they do not impact model output (negligible or acceptable change in accuracy, precision and/or recall).
Regarding claim 13, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 12 above. Xie does not specifically teach subsequent to identifying the structure as being removable from the computer-implemented DNN, identifying the structure as being redundant with another structure in the computer-implemented DNN
However Miret teaches subsequent to identifying the structure as being removable from the computer-implemented DNN, identifying the structure as being redundant with another structure in the computer-implemented DNN (Miret [66] based on determining a node is dependent on second node (and therefore redundant if the second node is removed)- they may be grouped together so that removing the second node will also remove the first node too).
Regarding claim 14, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 13 above. Xie does not specifically teach removing the structure from the computer-implemented DNN based upon the structure being identified as being redundant with the another structure in the computer-implemented DNN
However Miret teaches removing the structure from the computer-implemented DNN based upon the structure being identified as being redundant with the another structure in the computer-implemented DNN (Miret [66] based on determining a node is dependent on second node (and therefore redundant if the second node is removed)- they may be grouped together so that removing the second node will also remove the first node too).
Regarding claim 15, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 8 above. Xie does not specifically teach where the DNN is a convolutional neural network
However Miret teaches where the DNN is a convolutional neural network (Miret [28] DNN can be a convolutional neural network).
Regarding claim 16, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 8 above. Xie does not specifically teach wherein the untrained computer-implemented DNN, when trained, consumes a first amount of computer-readable memory, the trained and compressed computer-implemented DNN consumes a second amount of computer-readable memory, and further where the second amount of computer-readable memory is less than the first amount of computer-readable memory
However Miret teaches wherein the untrained computer-implemented DNN, when trained, consumes a first amount of computer-readable memory, the trained and compressed computer-implemented DNN consumes a second amount of computer-readable memory, and further where the second amount of computer-readable memory is less than the first amount of computer-readable memory (Merit [40] compression reduces resources (including memory) needed for training and storing trained model, reduction may be based on pruning ratio, so trained model after compression uses less memory than trained model before compression).
Regarding claim 17, Xie, Fan, Fusi and Miret teach the invention as claimed in claim 16 above. Xie does not specifically teach where the second amount of computer-readable memory is between 20% and 50% less than the first amount of computer-readable memory
However Miret teaches where the second amount of computer-readable memory is between 20% and 50% less than the first amount of computer-readable memory (Merit [40] compression reduces resources (including memory) needed for training and storing trained model, reduction may be based on pruning ratio, pruning ratio can be 20%).
Claim 18 is directed towards a medium storing instructions similar in scope to the instructions executed by the system of claim 1, and is rejected under the same rationale. Xie further teaches a computer-readable storage medium comprising instructions that, when executed by a processor, cause the processor to perform acts (Xie [6] device processor executes code stored in memory to perform operations for a model).
Claim(s) 19, 20, is/are dependent on claim 18 above, is/are directed towards a medium storing instructions similar in scope to the instructions executed by the system of claim(s) 2, 3 respectively, and is/are rejected under the same rationale.
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 SANCHITA ROY whose telephone number is (571)272-5310. The examiner can normally be reached Monday-Friday 12-8.
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, Usmaan Saeed can be reached at (571) 272-4046. 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.
SANCHITA ROY
Primary Examiner
Art Unit 2146
/SANCHITA ROY/Primary Examiner, Art Unit 2146
Prosecution Insights