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 .
Response to Arguments
Regarding the previous objection to the Abstract, Applicant’s amended Abstract has overcome the objection, which is withdrawn.
Regarding the rejection of claims under 35 U.S.C. 102 and 35 U.S.C. 103, Applicant’s arguments are directed towards amended claims which have not been previously examined, and for which new grounds of rejection are given below.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1–5, 7–10, and 14–22 rejected under 35 U.S.C. 103 over Weimin Zhou et al., US Pre-Grant Publication No. 2022/0414426 (hereafter Weimin) in view of Benyahia et al., US Pre-Grant Publication No. 2020/0104688 (hereafter Benyahia) and Zhou et al., US Pre-Grant Publication No. 2019/0354837 (hereafter Zhou).
Regarding claim 1 and analogous claims 14 and 17:
Weimin teaches:
“A method for generating a neural network, comprising”: Weimin, paragraph 0006, “This application provides a neural architecture search method [A method for generating a neural networ]. In the method, different initial child model training processes are decoupled, and a neural architecture search process is decoupled from an initial child model training process, to resolve a problem in a related technology that search efficiency is low and is difficult to meet service requirements. This application further provides an apparatus, a device, a computer-readable storage medium, and a computer program product that correspond to the method.”
“training a plurality of neural networks for a plurality of performance parameters to obtain a plurality of parameter values for each of the plurality of performance parameters”: Weimin, paragraphs 0008–0009, “The generator can generate a plurality of neural network architectures based on a search space, and may obtain a plurality of initial child models by performing weight initialization on the plurality of neural network architectures. A model training platform may train the plurality of initial child models in parallel to obtain a plurality of child models [training a plurality of neural networks ]. […] A model inference platform may perform inference on each child model on first hardware, to obtain an evaluation indicator value of the trained child model on the first hardware. The searcher may obtain evaluation indicator values of the plurality of child models on the first hardware [obtain a plurality of parameter values for each of the plurality of performance parameters], and determine, based on the evaluation indicator values and the neural network architectures corresponding to the child models, a first target neural network architecture that meets a preset condition.”
(bold only) “training a plurality of neural network predictors based on the parameter values and the neural networks”: Weimin, paragraph 0027, “In some possible implementations, the searcher includes an evaluator and a controller, and the search module is specifically configured to: train the evaluator based on the neural network architectures corresponding to the plurality of child models and the evaluation indicator values of the plurality of child models on the first hardware; and train the controller by using the trained evaluator [training a … neural network predictors based on the parameter values and the neural networks], and determine based on the trained controller, the first target neural network architecture that meets the preset condition.”
“determining a target neural network using the trained neural network predictors”: Weimin, paragraph 0012, “The searcher trains the evaluator based on the neural network architectures corresponding to the plurality of child models and the evaluation indicator values of the plurality of child models on the first hardware. The searcher trains the controller by using the trained evaluator, and determines, based on the trained controller, the first target neural network architecture that meets the preset condition [determining a target neural network using the trained neural network predictors].”
(bold only) “obtaining a network structure for the iteration based on a network structure obtained in the previous iteration and the gradient structures, wherein the gradient structures respectively corresponding to different neural network predictors are assigned different weights”: Weimin, paragraph 0127, “The training of the evaluator 2042 belongs to supervised learning. Based on this, the searcher 204 may perform iterative training by using a gradient descent method such as a stochastic gradient descent method, so that training efficiency is improved and training convergence time is reduced [the weights and other parameters of the child models are interpreted as gradient structures, hence training the child models interpreted as obtaining a network structure for the iteration based on … the gradient structures]”; Weimin, paragraph 0111, “Each child model may be represented by using a code of a neural network architecture corresponding to the child model and a parameter set (a set of model parameters, usually a set of weights) [wherein the gradient structures respectively corresponding to different neural network predictors are assigned different weights].”
Weimin does not explicitly teach:
(bold only) “training a plurality of neural network predictors based on the parameter values and the neural networks”
“wherein determining the target neural network using the trained neural network predictors comprises: performing multiple iterations using the trained neural network predictors to obtain a group of network structures, comprising, for each iteration”
“determining a plurality of gradient structures based on the network structure obtained in a previous iteration using the trained neural network predictors”
(bold only) “obtaining a network structure for the iteration based on a network structure obtained in the previous iteration and the gradient structures, wherein the gradient structures respectively corresponding to different neural network predictors are assigned different weights”
“wherein obtaining the network structure for the iteration based on the network structure obtained in the previous iteration and the gradient structures comprises: modifying the network structure obtained in the previous iteration using the gradient structures”
“determining whether the modified network structure belongs to a set of network structures which refers to a search space of the network structures; determining, from the set of network structures, a network structure closest to the modified network structure as the network structure for the iteration in response to the modified network structure not belonging to the set of network structures”
Benyahia teaches:
“wherein determining the target neural network using the trained neural network predictors comprises: performing multiple iterations using the trained neural network predictors to obtain a group of network structures, comprising, for each iteration”: Benyahia, paragraph 0123, “The subgraphs suggested by the controller may be examined when making the determination, and different criteria may be used in determining whether or not to keep iterating. The suggested subgraphs may be used in step 420 as the obtained set of subgraphs, and steps 420 to 460 may be repeated using these newly-suggested candidate models. By iterating this process, the method may get closer to obtaining a preferred model for performing the selected task [performing multiple iterations using the trained neural network predictors to obtain a group of network structures].”
“determining a plurality of gradient structures based on the network structure obtained in a previous iteration using the trained neural network predictors”: Benyahia, paragraph 0123, “The subgraphs suggested by the controller [trained neural network predictors] may be examined when making the determination, and different criteria may be used in determining whether or not to keep iterating. The suggested subgraphs may be used in step 420 as the obtained set of subgraphs, and steps 420 to 460 may be repeated using these newly-suggested candidate models. By iterating this process, the method may get closer to obtaining a preferred model for performing the selected task [determining a plurality of gradient structures based on the network structure obtained in a previous iteration].“
(bold only) “obtaining a network structure for the iteration based on a network structure obtained in the previous iteration and the gradient structures, wherein the gradient structures respectively corresponding to different neural network predictors are assigned different weights”: Benyahia, paragraph 0123, “The subgraphs suggested by the controller may be examined when making the determination, and different criteria may be used in determining whether or not to keep iterating. The suggested subgraphs may be used in step 420 as the obtained set of subgraphs, and steps 420 to 460 may be repeated using these newly-suggested candidate models. By iterating this process, the method may get closer to obtaining a preferred model for performing the selected task [based on a network structure obtained in the previous iteration].”
“wherein obtaining the network structure for the iteration based on the network structure obtained in the previous iteration and the gradient structures comprises: modifying the network structure obtained in the previous iteration using the gradient structures”: Benyahia, paragraph 0050, “The weightings may comprise a plurality of matrices that provide numerical weightings for scaling a value of information as it flows through the model. The method may comprise identifying weightings that have a low rate of change in response to training. For example, a second derivative may indicate that a weighting is not changing much ( e.g., changes below a threshold level) in response to training, which may indicate that it is close to an optimum value. A gradient associated with a cost function may be monitored for each given weighting. This may indicate a degree of sensitivity of the model on that weighting. Weightings are applied to scale information passing between nodes. Weightings may be applied at one or both of: each node and each edge. Applying weightings at each edge may enable dimensionality to be managed more easily. The weightings may be stored in the edges.”
“determining whether the modified network structure belongs to a set of network structures which refers to a search space of the network structures; determining, from the set of network structures, a network structure closest to the modified network structure as the network structure for the iteration in response to the modified network structure not belonging to the set of network structures”: Benyahia, paragraph 0124, “Criteria may be applied to determine whether a suggested subgraph [the modified network structure], or the best subgraph from the previous set of candidate models [a set of network structures which refers to a search space of the network structures], should be used as the preferred model that the neural network will use for performing the selected task. For example, a comparison may be made between a suggested candidate model and the previous best candidate model. If the two are functionally very similar, e.g., have an architectural degree of similarity above a threshold level/contain a threshold number of the same neurons [determining, from the set of network structures, a network structure closest to the modified network structure as the network structure for the iteration ] ( e.g., in the same location in the network), it may be determined that the suggested model/the previous best model should be taken as the preferred model [determining whether the modified network structure belongs to a set of network structures which refers to a search space of the network structures]. If the controller suggests a plurality of different subgraphs, a comparison may be performed to see if any of these have been previously suggested by the controller.”
Benyahia and Weimin are analogous arts as they are both related to neural architecture search. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the iterative search techniques of Benyahia with the teachings of Weimin to arrive at the present invention, in order to provide improved neural architecture search results, as stated in Benyahia, paragraph 0123, “By iterating this process, the method may get closer to obtaining a preferred model for performing the selected task.”
Zhou teaches (bold only) “training a plurality of neural network predictors based on the parameter values and the neural networks”: Zhou, paragraph 0046, “In one or more embodiments, a value network 140 takes in network embedding of the generated target network 145 and data distributions to approximate the reward by ascertain metrics, such as network accuracy 150 and training time 155--although other metrics may also be determined. In one or more embodiments, the value network may predict target network accuracy and training time without actually running the target network till convergence. In one or more embodiments, both the accuracy network 150 and the training time network 155 are trainable neural networks that may be pre-trained or trained jointly with the policy network [training a plurality of neural network predictors based on the parameter values and the neural networks].”
Zhou and Weimin are analogous arts as they are both related to neural architecture search. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the multiple neural predictors of Zhou with the teachings of Weimin to arrive at the present invention, in order to perform neural architecture searches on multiple criteria, as stated in Zhou, paragraph 0046, “In one or more embodiments, a value network 140 takes in network embedding of the generated target network 145 and data distributions to approximate the reward by ascertain metrics, such as network accuracy 150 and training time 155--although other metrics may also be determined. In one or more embodiments, the value network may predict target network accuracy and training time without actually running the target network till convergence.”
Regarding claim 2 and analogous claims 15 and 18:
Weimin as modified by Benyahia and Zhou teaches “the method according to claim 1.”
Zhou further teaches “wherein the neural network predictors comprise a main predictor and an auxiliary predictor for predicting different ones of the plurality of performance parameters for the neural networks, respectively”: Zhou, paragraph 0046, “In one or more embodiments, a value network 140 takes in network embedding of the generated target network and data distributions to approximate the reward by ascertain metrics, such as network accuracy 150 and training time 155--although other metrics may also be determined. In one or more embodiments, the value network may predict target network accuracy and training time without actually running the target network till convergence. In one or more embodiments, both the accuracy network 150 and the training time network 155 [comprise a main predictor and an auxiliary predictor for predicting different ones of the plurality of performance parameters for the neural networks, respectively] are trainable neural networks that may be pre-trained or trained jointly with the policy network.”
Zhou and Weimin are combinable for the rationale given under claim 1.
Regarding claim 3 and analogous claims 16 and 19:
Weimin as modified by Benyahia and Zhou teaches “the method according to claim 1.”
Weimin further teaches “determining the set of network structures, each network structure in the set of network structures characterizing a neural network, wherein training the plurality of neural networks comprises: selecting a plurality of network structures characterizing the plurality of neural networks from the set of network structures”: Weimin, paragraph 0008, “The generator can generate a plurality of neural network architectures based on a search space, and may obtain a plurality of initial child models by performing weight initialization on the plurality of neural network architectures [each network structure in the set of network structures characterizing a neural network]. A model training platform may train the plurality of initial child models in parallel to obtain a plurality of child models”; Weimin, paragraph 0018, “In some possible implementations, the search space is represented by using an attribute value space of each attribute of a neuron. The generator may randomly select an attribute value for each attribute from the attribute value space of the attribute of the neuron, to obtain the plurality of neural network architectures [selecting a plurality of network structures characterizing the plurality of neural networks from the set of network structures]. In this way, balance of samples for training the evaluator can be ensured.”
Regarding claim 4 and analogous claim 20:
Weimin as modified by Benyahia and Zhou teaches “the method according to claim 3.”
Zhou further teaches “wherein the set of network structures comprises a network structure represented by a directed acyclic graph, wherein each node of the directed acyclic graph represents an operation, wherein each edge of the directed acyclic graph represents a connection relationship between two corresponding nodes of the directed acyclic graph”: Zhou, Fig. 8,
PNG
media_image1.png
744
678
media_image1.png
Greyscale
[showing a directed acyclic graph representing a network structure, where each node, e.g., “Convolution + RELU,” represents an operation and each of the edges represents a connection relationship between two corresponding nodes].
Zhou and Weimin are analogous arts as they are both related to neural architecture search. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the graph representation of Zhou with the teachings of Weimin to arrive at the present invention, in order to perform neural architecture search layer-by-layer, as stated in Zhou, paragraph 0068, “In one or more embodiments, layer-by-layer search aims to find the optimal architecture with a search granularity of predefined layers. FIG. 8 graphically depicts an example insert operation for layer-by-layer search, according to embodiments of the present disclosure.”
Regarding claim 5:
Weimin as modified by Benyahia and Zhou teaches “the method according to claim 4.”
Zhou further teaches “wherein the set of network structures further comprises a network structure represented by a one-dimensional vector”: Zhou, paragraph 0060, “FIG. 5 graphically shows an alternative depiction of an embedding network 500, where a layer embedding network 505 takes a layer description and maps layer features into multiple lookup tables, according to embodiments of the present disclosure. In one or more embodiments, lookup tables (e.g., lookup tables 510-x) transform the discrete feature space into trainable feature vectors. An LSTM network takes layer feature vectors (e.g., 515-x) and generates a layer embedding 520 [a network structure represented by a one-dimensional vector]. After multiple layer embedding have been produced, a network embedding LSTM network 525 processes the sequential information in these layer embeddings and generates a network embedding 535. In one or more embodiments, this network embedding 535 is used as by the policy network and by a value network.”
Zhou and Weimin are analogous arts as they are both related to neural architecture search. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the vector architecture representation of Zhou with the teachings of Weimin to arrive at the present invention, in order to enable direct processing of the architecture by machine learning models, as stated in Zhou, paragraph 0060, “In one or more embodiments, this network embedding 535 is used as by the policy network and by a value network.”
Regarding claim 7:
Weimin as modified by Benyahia and Zhou teaches “the method according to claim 1.”
Weimin further teaches “wherein determining the target neural network using the trained neural network predictors further comprises: selecting a network structure characterizing the target neural network from the group of network structures according to a predetermined rule”: Weimin, paragraph 0009, “The searcher may obtain evaluation indicator values of the plurality of child models on the first hardware, and determine, based on the evaluation indicator values and the neural network architectures corresponding to the child models, a first target neural network architecture that meets
a preset condition [selecting a network structure characterizing the target neural network from the group of network structures according to a predetermined rule].”
Regarding claim 8:
Weimin as modified by Benyahia and Zhou teaches “the method according to claim 3.”
Zhou further teaches:
“wherein determining the target neural network using the trained neural network predictors comprises: performing multiple iterations to iteratively train the neural network predictors, comprising, for each iteration”: Zhou, paragraph 0077, “In one or more embodiments, the updated policy network is used for the next episode. The number of episodes may be user-selected or may be based upon one or more stop conditions ( e.g., runtime of RENA embodiment, number of iterations, convergence ( or difference between iteration is not changing more than a threshold, divergence, and/or performance of the neural network meets criteria) [performing multiple iterations to iteratively train the neural network predictors].”
“obtaining a group of network structures using the trained neural network predictors”: Zhou, paragraph 0048, “In one or more embodiments, an initial neural network architecture configuration is mapped (205) into a representation, such as using a lookup table. In one or more embodiments, a policy network converts (210) that initial neural network architecture configuration representation into a network embedding. Then, in embodiments, the policy network uses (215) that network embedding to automatically generate adaptations to the neural network architecture configuration [obtaining a group of network structures]. In one or more embodiments, the adapted neural network is trained (220) to convergence, and the trained adapted neural network architecture may be evaluated (225) based upon one or more metrics (e.g., accuracy, memory footprint, power consumption, inference latency, etc.). In one or more embodiments, a policy gradient method may be used (230) to compute a multi-objective reward that is feed back to the policy network to improve the policy network's ability to automatically generate a set of one or more best architectures. In one or more embodiments, a number of adapted neural network architectures may be processed in parallel per episode as part of the reinforcement step”; Zhou, Fig. 1,
PNG
media_image2.png
709
900
media_image2.png
Greyscale
[showing how trained neural network predictors, such as Accuracy Network 150 and Training Time Network 155, are used in the Value Network 140 to provide policy gradient data to influence policy in the Policy Network 110 to produce further network structures].
“training a neural network characterized by at least one network structure of the group of network structures for the plurality of performance parameters to obtain a group of parameter values for each of the plurality of performance parameters”: Zhou, paragraph 0048, “In one or more embodiments, an initial neural network architecture configuration is mapped (205) into a representation, such as using a lookup table. In one or more embodiments, a policy network converts (210) that initial neural network architecture configuration representation into a network embedding. Then, in embodiments, the policy network uses (215) that network embedding to automatically generate adaptations to the neural network architecture configuration. In one or more embodiments, the adapted neural network is trained (220) to convergence, and the trained adapted neural network architecture may be evaluated (225) based upon one or more metrics (e.g., accuracy, memory footprint, power consumption, inference latency, etc.) [training a neural network characterized by at least one network structure of the group of network structures for the plurality of performance parameters to obtain a group of parameter values for each of the plurality of performance parameters].”
“and training the neural network predictors based on at least the group of parameter values and the group of network structures”: Zhou, paragraph 0046, “In one or more embodiments, a value network 140 takes in network embedding of the generated target network 145 and data distributions to approximate the reward by ascertain metrics, such as network accuracy 150 and training time 155--although other metrics may also be determined. In one or more embodiments, the value network may predict target network accuracy and training time without actually running the target network till convergence. In one or more embodiments, both the accuracy network 150 and the training time network 155 are trainable neural networks that may be pre-trained or trained jointly with the policy network [training a plurality of neural network predictors based on the parameter values and the neural networks].”
Zhou and Weimin are analogous arts as they are both related to neural architecture search. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the iterative performance-based training of Zhou with the teachings of Weimin to arrive at the present invention, in order to improve neural architecture search towards particular goals, as stated in Zhou, paragraph 0047, “In one or more embodiments, the final reward engine 160 sets weights to various metrics, such as network accuracy, model size, and training time, which may be set according to user specification. The configurable reward engine 160 enables finding neural architectures with various resource constraints, such as memory size and GPU time.”
Regarding claim 9:
Weimin as modified by Benyahia and Zhou teaches “the method according to claim 8.”
Zhou further teaches:
“wherein obtaining the group of network structures using the trained neural network predictors comprises: performing multiple iterations using the trained neural network predictors to obtain the group of network structures, comprising, for each iteration”: Zhou, paragraph 0077, “In one or more embodiments, the updated policy network is used for the next episode. The number of episodes may be user-selected or may be based upon one or more stop conditions ( e.g., runtime of RENA embodiment, number of iterations, convergence ( or difference between iteration is not changing more than a threshold, divergence, and/or performance of the neural network meets criteria) [performing multiple iterations]”; Zhou, Fig. 1,
PNG
media_image2.png
709
900
media_image2.png
Greyscale
[showing how trained neural network predictors, such as Accuracy Network 150 and Training Time Network 155, are used in the Value Network 140 to provide policy gradient data to influence policy in the Policy Network 110 to produce further network structures].
“determining a plurality of gradient structures based on the network structure obtained in a previous iteration using the trained neural network predictors, and”: Zhou, Fig. 1,
PNG
media_image2.png
709
900
media_image2.png
Greyscale
[showing trained neural network predictors, such as Accuracy Network 150 and Training Time Network 155, used in the Value Network 140 to provide policy gradient data to influence policy in the Policy Network 110 to produce further network structures, used from one iteration to the next]; Zhou, paragraph 0046, “In one or more embodiments, a value network 140 takes in network embedding of the generated target network 145 and data distributions to approximate the reward by ascertain metrics, such as network accuracy 150 and training time 155-although other metrics may also be determined. In one or more embodiments, the value network may predict target network accuracy and training time without actually running the target network till convergence. In one or more embodiments, both the accuracy network 150 and the training time network 155 are trainable neural networks that may be pre-trained or trained jointly with the policy network”; Zhou, paragraph 0089, “The parameters Θv if [sic] the value network is updated via gradient descent [determining a plurality of gradient structures] using
PNG
media_image3.png
39
283
media_image3.png
Greyscale
.”
“obtaining a network structure for the iteration based on a network structure obtained in the previous iteration and the gradient structures”: Zhou, paragraph 0089, “The parameters Θv if [sic] the value network is updated via gradient descent [gradient structures] using
PNG
media_image3.png
39
283
media_image3.png
Greyscale
”; Zhou, Fig. 1,
PNG
media_image2.png
709
900
media_image2.png
Greyscale
[showing trained neural network predictors, such as Accuracy Network 150 and Training Time Network 155, used in the Value Network 140, to provide policy gradient data to influence policy in the Policy Network 110 to produce further network structures, used from one iteration to the next].
Zhou and Weimin are combinable for the rationale given under claim 8.
Regarding claim 10:
Weimin as modified by Benyahia and Zhou teaches “the method according to claim 9.”
Weimin further teaches (bold only) “wherein determining the target neural network using the trained neural network predictors further comprises: selecting a network structure characterizing the target neural network from the group of network structures obtained in a last iteration for iteratively training the neural network predictors according to a predetermined rule”: Weimin, paragraph 0009, “The searcher may obtain evaluation indicator values of the plurality of child models on the first hardware, and determine, based on the evaluation indicator values and the neural network architectures corresponding to the child models, a first target neural network architecture that meets
a preset condition [selecting a network structure characterizing the target neural network from the group of network structures according to a predetermined rule].”
Benyahia further teaches (bold only) “wherein determining the target neural network using
the trained neural network predictors further comprises: selecting a network structure characterizing the target neural network from the group of network structures obtained in a last iteration for iteratively training the neural network predictors according to a predetermined rule”: Benyahia, paragraph 0123, “The subgraphs suggested by the controller may be examined when making the determination, and different criteria may be used in determining whether or not to keep iterating. The suggested subgraphs may be used in step 420 as the obtained set of subgraphs, and steps 420 to 460 may be repeated using these newly-suggested candidate models. By iterating this process, the method may get closer to obtaining a preferred model for performing the selected task [network structures obtained in a last iteration for iteratively training the neural network predictors].”
Benyahia and Weimin are combinable for the rationale given under claim 1.
Regarding claim 21:
Weimin as modified by Benyahia and Zhou teaches “the method according to claim 1.”
Zhou further teaches “wherein a number of neural network predictors trained corresponds to a number of performance parameters, and each neural network predictor is used for predicting one performance parameter for the neural networks”: Zhou, paragraph 0046, “In one or more embodiments, a value network 140 takes in network embedding of the generated target network and data distributions to approximate the reward by ascertain metrics, such as network accuracy 150 and training time 155--although other metrics may also be determined. In one or more embodiments, the value network may predict target network accuracy and training time without actually running the target network till convergence. In one or more embodiments, both the accuracy network 150 and the training time network 155 [wherein a number of neural network predictors trained corresponds to a number of performance parameters, and each neural network predictor is used for predicting one performance parameter for the neural networks] are trainable neural networks that may be pre-trained or trained jointly with the policy network.”
Zhou and Weimin are combinable for the rationale given under claim 1.
Regarding claim 22:
Weimin as modified by Benyahia and Zhou teaches “the method according to claim 1.”
Zhou further teaches “wherein the plurality of performance parameters comprises an accuracy, a number of parameters, an amount of delay at run-time, and an amount of computation needed at run-time”: Zhou, paragraph 0048, “In one or more embodiments, an initial neural network architecture configuration is mapped (205) into a representation, such as using a lookup table. In one or more embodiments, a policy network converts (210) that initial neural network architecture configuration representation into a network embedding. Then, in embodiments, the policy network (215) uses that network embedding to automatically generate adaptations to the neural network architecture configuration. In one or more embodiments, the adapted neural network is trained (220) to convergence, and the trained adapted neural network architecture may be evaluated (225) based upon one or more metrics (e.g., accuracy [accuracy], memory footprint [a number of parameters], power consumption [an amount of computation needed at run-time], inference latency [an amount of delay at run-time], etc.).”
Zhou and Weimin are analogous arts as they are both related to neural architecture search. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the performance metrics of Zhou with the teachings of Weimin to arrive at the present invention, in order to direct neural architecture search towards particular goals, as stated in Zhou, paragraph 0048, “the trained adapted neural ne work architecture may be evaluated (225) based upon one or more metrics (e.g., accuracy, memory footprint, power consumption, inference latency, etc.).”
Claim 23 rejected under 35 U.S.C. 103 over Weimin as modified by Benyahia and Zhou in view of Kehler, US Pre-Grant Publication No. 2022/0108195 (hereafter Kehler).
Weimin as modified by Benyahia and Zhou teaches “the method according to claim 1.”
Weimin as modified by Benyahia and Zhou does not explicitly teach “wherein the network structure of the neural network comprises an operation matrix and a connection matrix, the method further comprises: encoding nodes of a directed acyclic graph to form the operation matrix; and encoding connection relationships between the nodes in the directed acyclic graph to form the connection matrix representing edges of the directed acyclic graph.”
Kehler teaches “wherein the network structure of the neural network comprises an operation matrix and a connection matrix, the method further comprises: encoding nodes of a directed acyclic graph to form the operation matrix; and encoding connection relationships between the nodes in the directed acyclic graph to form the connection matrix representing edges of the directed acyclic graph”: Kehler, paragraph 0196, “In some examples, one or more nodes of a PGN may correspond to responses, and edges between the different nodes may be indicative of associations between the responses. In some examples, one or more nodes may correspond to an identified theme ( e.g., for one or more responses), an evaluation question or stimulus for which a response is received, or other information component described herein. In some examples, the edges may be directed, such as a pointer in a directed acyclic graph [directed acyclic graph], and indicative of a direction of the association ( e.g., a plurality of pointers may point from a stimulus to evaluation questions for the stimulus, another plurality of pointers may point from an evaluation question to responses submitted in association with the evaluation question, and another one or more pointers may point from a response to other related responses (e.g., based on determined relevance) or to an identified theme ( e.g., based on determined relevance) or vice versa)”; Kehler, paragraph 0201, “In some embodiments, the edges associated with a node may be represented as vectors to determine a matrix (e.g., like a win/loss matrix), from which an adjacency matrix A may be determined [encoding connection relationships between the nodes in the directed acyclic graph to form the connection matrix representing edges of the directed acyclic graph]”; Kehler, paragraph 0206, “Examples of probabilistic graphical networks may map, such as in a graph, which in some embodiments may be processed for display by a visualization system, information about an evaluation like that described above based on the encoding of nodes [encoding nodes of a directed acyclic graph to form the operation matrix] and relationships, or edges, between nodes. The graph may display results determined based on the responses provided for different features (e.g., evaluation requests, stimuli, etc.) of an evaluation, or other information about the evaluation process ( e.g., how rating entities participated in the evaluation and upon information submitted by other rating entities).”
Kehler and Weimin as modified by Benyahia and Zhou are analogous arts as they are both related to graph representations. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the embeddings and matrix graph representation of Kehler with the teachings of Weimin as modified by Benyahia and Zhou to arrive at the present invention, in order to convert the graph to a purely numeric representation for further processing, as shown in the example given in Kehler, paragraph 0203, “Here, the adjacency matrix A may be processed, such as by application of the matrix power low, to determine an eigenvalue/eigenvector with respect to the different ranking entities, and thus a ranked order and influence or engagement metric thereof ( e.g., similar to that of relevance for responses).”
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Dai et al., US Pre-Grant Publication No. 2022/0188599, discloses a method of neural architecture search which iteratively processes a search space, including selection of models using performance criteria.
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 VINCENT SPRAUL whose telephone number is (703) 756-1511. The examiner can normally be reached M-F 9:00 am - 5:00 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, MICHAEL HUNTLEY can be reached at (303) 297-4307. 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.
/VAS/Examiner, Art Unit 2129
/MICHAEL J HUNTLEY/Supervisory Patent Examiner, Art Unit 2129
Prosecution Insights