DETAILED ACTION
1. This office action is in response to the Application No. 19044546 filed on 04/21/2026. Claims 1-24 are presented for examination and are currently pending.
Notice of Pre-AIA or AIA Status
2. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Continued Examination Under 37 CFR 1.114
3. A request for continued examination under 37 CFR 1.114, including the fee set
forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this
application is eligible for continued examination under 37 CFR 1.114, and the fee set
forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action
has been withdrawn pursuant to 37 CFR 1.114. Applicant’s submission filed on
04/21/2026 has been entered.
Response to Arguments
4. The Applicant’s arguments have been considered but are moot in view of the newly applied primary reference.
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.
5. Claims 1 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Dong et al. ("Spartan: A Sparsity-Adaptive Framework to Accelerate Deep Neural Network Training on GPUs, Date of Publication: 22 March 2021) in view of Yuan et al. ("STICKER: An energy-efficient multi-sparsity compatible accelerator for convolutional neural networks in 65-nm CMOS." IEEE Journal of Solid-State Circuits 55.2 (2019): 465-477)
Regarding claim 1, Dong teaches a computer system comprising a hardware memory, wherein the computer system is configured to execute software instructions stored on nontransitory machine- readable storage media that (Both systems are equipped with CPU and GPU. The first one is a desktop-grade system, equipped with an AMD Radeon RX Vega56 as the GPU platform and an Intel(R) Core(TM) i7-8700 as the CPU platform (pg. 2456, right col., first para.); main memory of the GPU, pg. 2449, left col., third bullet point):
initialize a neural network comprising a plurality of interconnected nodes arranged in layers (We select three layers (Layer A, Layer B and Layer C) as representative layers for the models, all with the same filter size. Layers A, B, and C correspond to the second convolutional layer from the fourth residual block of ResNet-10, the fifth convolutional layer of VGGNet-11 and the seventh convolutional layer of VGGNet-11, respectively. (pg. 2457, left col., last para. to right col., first para.)),
wherein the neural network includes a hierarchical supervisory structure (Fig. 3. (a) The overview of the sparsity monitor (pg. 2450, Fig. 3)) and a meta-supervisory structure (Fig. 3. (b) State transition of Dynamic Monitoring Period Management, pg. 2450, Fig. 3));
operate the hierarchical supervisory structure (Fig. 3a presents an overview of the sparsity monitor, which consists of three components: 1) a scheduler, 2) a sparsity list, and 3) a sparsity calculator. The regular workflow of the sparsity monitor is as follows. First, the model information (i.e., the structure of the DNN model and a list containing which activation maps we select to monitor) is sent to the scheduler to initiate the monitoring process. Next, the scheduler determines when to profile the activation maps and enables the sparsity calculator to compute the degree of sparsity based on the selected activation maps (Data) (pg. 2450, right col., first para.) within the neural network (we present Spartan, a lightweight hardware/software framework to accelerate DNN training...Spartan provides an efficient sparsity monitor, a tile-based sparse GEMM algorithm, abstract) by:
monitoring the neural network at various levels of granularity (In this section, we present our sparsity monitor design, providing efficient and flexible sparsity monitoring during DNN training. (pg. 2450, right col., section 3 Sparsity Monitor);
collecting activation data from a plurality of nodes within the neural network (We can summarize the sparsity patterns observed as follows... Activation maps from different layers contain different degrees of sparsity and sparsity trends over time (pg. 2449, right col., last para.);
identifying a plurality of operation patterns from the activation data (If the sparsity levels of all selected monitored activation maps become stable, the state can transit to Hibernate, a state where all activation maps share the same monitoring period. While in the Hibernate state, if any of the monitored activation maps exhibit unstable sparsity trends, the state transits to Active, resetting the monitoring period for all monitored activation maps to the initial monitoring period (pg. 2452, left col., first para.);
and implementing architectural changes to the neural network based on identified operation patterns (As per observation 2, activation maps from different layers present different levels of sparsity and the sparsity associated with these activation maps may change overtime, pg. 2451, left col., first para. The Examiner notes that different levels of sparsity are architectural changes to the neural network);
operate the meta-supervisory structure by: monitoring behavior of the hierarchical supervisory structure (The scheduler manages and monitors each individual activation map by regulating two important parameters: 1) the monitoring period, and 2) the monitoring duration (pg. 2450, right col., second to the last para.) to identify supervisory patterns (We propose multiple mechanisms to assist with periodic monitoring, dynamically adjusting the monitoring period. The mechanisms are: i) flexible monitoring, ii) fast termination, and iii) dynamic management of the monitoring period management (pg. 2450, right col., last para. to 2451, left col., first para.);
storing identified supervisory patterns in the hardware memory (storing the most recent measured sparsity for every monitored activation map managed in the sparsity list, pg. 2451, right col., first para.);
and using the stored supervisory patterns (The DPAA (Algorithm 1) can adjust the monitoring period by maintaining history and storing the most recent measured sparsity for every monitored activation map managed in the sparsity list. The algorithm first collects the measured sparsity and adds it to a history list, saving only a number of the most recent measured sparsity. Then it checks the difference between the most recent sparsity and the oldest in the history list. If the absolute difference is smaller than a threshold, we double the length of the monitoring period in the Active state, pg. 2451, right col., first para.)
establish direct connections between neurons belonging to non-adjacent regions of the neural network (Both convolutional and fully-connected layers use General Matrix Multiplication (GEMM) as their primary computational kernel. The computation in the fully-connected layers can be directly represented by GEMM, pg. 2450, left col., section 2.2);The sparsity monitor detects activation sparsity before each convolutional layer in a DNN model (pg. 2450, right col., first para. The Examiner notes fully connected layers has connections between neurons that belong to non-adjacent regions of the neural network);
apply a plurality of transformations to signals propagating along respective direct connections (The computation in the fully-connected layers can be directly represented by GEMM, while the computation in the convolutional layers can be a combination of an im2col operation (transforming high-dimensional activation/feature maps into a 2D matrix, pg. 2450, left col., section 2.2 Characterization of Sparse Matrix Operations));
and coordinate timing of signal propagation through the direct connections (The scheduler manages and monitors each individual activation map by regulating two important parameters: 1) the monitoring period, and 2) the monitoring duration. The former determines the timing gap between two monitoring processes, pg. 2450, right col., second para.).
Dong does not explicitly teach to identify architectural modifications to the hierarchical supervisory structure;
Yuan teaches to identify architectural modifications to the hierarchical supervisory structure (After pruning, most convolution layers are sparse, pg. 474, left col., second to the last para.);
establish direct connections between neurons belonging to non-adjacent regions of the neural network (In the neural network in Figure 2, third neuron in the first layer is connected to the fifth neuron in the second layer, second neuron in the second layer is connected to the third neuron in the third layer, second neuron in the third layer is connected to fifth neuron in the fourth layer, pg. 466);
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong to incorporate the teachings of Yuan for the benefit of using an energy-efficient convolutional neural network (NN) processor that mainly improves energy efficiency by making full use of sparsity which can potentially lower storage and computation requirements (Yuan, abstract)
Regarding claim 13, claim 13 is similar to claim 1. It is rejected in the same manner and reasoning applying.
6. Claims 2-6, 8, 14-18 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Dong et al. ("Spartan: A Sparsity-Adaptive Framework to Accelerate Deep Neural Network Training on GPUs, Date of Publication: 22 March 2021) in view of Yuan et al. ("STICKER: An energy-efficient multi-sparsity compatible accelerator for convolutional neural networks in 65-nm CMOS." IEEE Journal of Solid-State Circuits 55.2 (2019): 465-477) in view of Chen et al. ("Multi-scale adaptive graph neural network for multivariate time series forecasting." IEEE Transactions on Knowledge and Data Engineering 35.10 (2023): 10748-10761)
Regarding claim 2, Dong and Yuan teaches the computer system of claim 1, they do not explicitly teach the limitations of claim 2.
Chen teaches wherein the transformations comprise adaptive matrices that evolve based on observed transmission effectiveness across multiple time scales (The adaptive graph learning module automatically generates adjacency matrices to represent the inter-variable dependencies among MTS (multivariate time series) (pg. 5, left col., last para.)).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yaun to incorporate the teachings of Chen for the benefit of effectively promoting the collaboration across different time scales, and automatically capture the importance of contributed temporal patterns in a multi-scale adaptive graph neural network (Chen, abstract)
Regarding claim 3, Dong, Yuan and Chen teaches the computer system of claim 2, Chen teaches wherein the transformations implement time-dependent signal modifications according to learned temporal patterns (Thus, an accurate MTS forecasting model should learn a feature representation that can comprehensively reflect all kinds of multi-scale temporal patterns (pg. 2, left col., first para.)).
The same motivation to combine dependent claim 2 applies here.
Regarding claim 4, Dong and Yuan teaches the computer system of claim 1, they do not explicitly teach the limitations of claim 4.
Chen teaches wherein temporal coordination synchronizes signal propagation through direct pathways with traditional layer-to-layer transmission (A multi-scale pyramid network to preserve the underlying temporal hierarchy at different time scales … Multi-scale pyramid network generates multi-scale feature representations through pyramid layers. Each pyramid layer takes the outputs of a preceding pyramid layer as the input and generates the feature representations of a larger scale as the output, pg. 4, right col., second and third para.).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yaun to incorporate the teachings of Chen for the benefit of effectively promoting the collaboration across different time scales, and automatically capture the importance of contributed temporal patterns in a multi-scale adaptive graph neural network (Chen, abstract)
Regarding claim 5, Dong and Yuan teaches the computer system of claim 1, they do not explicitly teach the limitations of claim 5.
Chen teaches wherein the hierarchical supervisory structure implements multi-level decision making for architectural modifications (a refining module that consists of two full connected layers to compact the fine-grained information across different time scales (pg. 6, left col., last para.); Fig. 2 … a multi-scale pyramid network to preserve the underlying temporal hierarchy at different time scales, pg. 4, right col., first para.), with different supervisory levels coordinating through information exchange about resource availability and network capacity (Space-based methods define the graph convolution through information propagation, which aggregates the representation of a central node and the representations of its neighbors to get the updated representation for the node, pg. 4, left col., section B. Graph Neural Networks).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yaun to incorporate the teachings of Chen for the benefit of effectively promoting the collaboration across different time scales, and automatically capture the importance of contributed temporal patterns in a multi-scale adaptive graph neural network (Chen, abstract)
Regarding claim 6, Dong and Yuan teaches the computer system of claim 1, they do not explicitly teach the limitations of claim 6.
Chen teaches wherein the meta-supervisory structure implements pattern recognition algorithms that identify common elements across successful adaptation episodes while maintaining operational stability (Fig. 2 … a multi-scale temporal graph neural network to capture all kinds of scale-specific temporal patterns; d) a scale-wise fusion module to effectively promote the collaboration across different time scales, pg. 4, right col., first para.).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yaun to incorporate the teachings of Chen for the benefit of effectively promoting the collaboration across different time scales, and automatically capture the importance of contributed temporal patterns in a multi-scale adaptive graph neural network (Chen, abstract)
Regarding claim 8, Dong and Yuan teaches the computer system of claim 1, they do not explicitly teach the limitations of claim 8.
Chen teaches wherein the direct connections enable controlled signal interaction during transmission through learned interaction weights that adapt based on observed effectiveness ((d) Scale-specific representations are weighted fused to capture the contributed temporal patterns (Fig. 2, pg. 5); To pure MTS data without any prior knowledge, the weighted adjacency matrices of multiple graphs need to be learned to represent the abundant and implicit inter-variable dependencies, pg. 4, left col., second para.).
Regarding claim 14, Dong and Yuan teaches the method of claim 13, they does not explicitly teach wherein the step of applying a plurality of transformations to signals comprises adapting transformation matrices based on observed transmission effectiveness across multiple time scales.
Chen teaches wherein the step of applying a plurality of transformations to signals comprises adapting transformation matrices based on observed transmission effectiveness across multiple time scales (The adaptive graph learning module automatically generates adjacency matrices to represent the inter-variable dependencies among MTS (multivariate time series) (pg. 5, left col., last para.)).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Chen for the benefit of effectively promoting the collaboration across different time scales, and automatically capture the importance of contributed temporal patterns in a multi-scale adaptive graph neural network (Chen, abstract)
Regarding claim 15, Modified Yuan teaches the method of claim 14, Chen teaches wherein the step of modifying applying a plurality of transformations to signals further comprises implementing time-dependent signal modifications according to learned temporal patterns (Thus, an accurate MTS forecasting model should learn a feature representation that can comprehensively reflect all kinds of multi-scale temporal patterns (pg. 2, left col., first para.)).
The same motivation to combine dependent claim 14 applies here.
Regarding claim 16, Dong and Yuan teaches the method of claim 13, they do not explicitly teach wherein the step of coordinating timing of signal propagation comprises synchronizing signals through direct pathways with traditional layer-to-layer transmission.
Chen teaches wherein the step of coordinating timing of signal propagation comprises synchronizing signals through direct pathways with traditional layer-to-layer transmission (A multi-scale pyramid network to preserve the underlying temporal hierarchy at different time scales … Multi-scale pyramid network generates multi-scale feature representations through pyramid layers. Each pyramid layer takes the outputs of a preceding pyramid layer as the input and generates the feature representations of a larger scale as the output, pg. 4, right col., second and third para.)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Chen for the benefit of effectively promoting the collaboration across different time scales, and automatically capture the importance of contributed temporal patterns in a multi-scale adaptive graph neural network (Chen, abstract)
Regarding claim 17, Dong and Yuan teaches the method of claim 13, they does not explicitly teach wherein the step of operating the hierarchical supervisory structure further comprises coordinating decisions across supervisory levels through information exchange about resource availability and network capacity.
Chen teaches wherein the step of operating the hierarchical supervisory structure further comprises coordinating decisions across supervisory levels through information exchange about resource availability and network capacity (Space-based methods define the graph convolution through information propagation, which aggregates the representation of a central node and the representations of its neighbors to get the updated representation for the node, pg. 4, left col., section B. Graph Neural Networks).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Chen for the benefit of effectively promoting the collaboration across different time scales, and automatically capture the importance of contributed temporal patterns in a multi-scale adaptive graph neural network (Chen, abstract)
Regarding claim 18, Dong and Yuan teaches the method of claim 13, they do not explicitly teach wherein the step of operating the meta-supervisory structure further comprises identifying common elements across successful adaptation episodes while maintaining operational stability.
Chen teaches wherein the step of operating the meta-supervisory structure further comprises identifying common elements across successful adaptation episodes while maintaining operational stability (Fig. 2 … a multi-scale temporal graph neural network to capture all kinds of scale-specific temporal patterns; d) a scale-wise fusion module to effectively promote the collaboration across different time scales, pg. 4, right col., first para.).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Chen for the benefit of effectively promoting the collaboration across different time scales, and automatically capture the importance of contributed temporal patterns in a multi-scale adaptive graph neural network (Chen, abstract)
Regarding claim 20, Dong and Yuan teaches the method of claim 13, they do not explicitly teach the limitations of claim 20.
Chen teaches further comprising the step of enabling controlled signal interaction during transmission through learned interaction weights that adapt based on observed effectiveness (the learned multi-scale feature representations are flexible and comprehensive to preserve various kinds of temporal dependencies (pg. 5, left col., second para.); the weighted adjacency matrices of multiple graphs need to be learned to represent the abundant and implicit inter-variable dependencies, pg. 4, left col., second para.)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Chen for the benefit of effectively promoting the collaboration across different time scales, and automatically capture the importance of contributed temporal patterns in a multi-scale adaptive graph neural network (Chen, abstract)
7. Claim 7 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Dong et al. ("Spartan: A Sparsity-Adaptive Framework to Accelerate Deep Neural Network Training on GPUs, Date of Publication: 22 March 2021) in view of Yuan et al. ("STICKER: An energy-efficient multi-sparsity compatible accelerator for convolutional neural networks in 65-nm CMOS." IEEE Journal of Solid-State Circuits 55.2 (2019): 465-477) in view of Abraham et al. ("Back to the Future: Reversible Runtime Neural Network Pruning for Safe Autonomous Systems." 2024 Design, Automation & Test in Europe Conference & Exhibition (25 March 2024). IEEE, 2024).
Regarding claim 7, Dong and Yuan teaches the computer system of claim 1, Yuan teaches wherein the computer system is further configured to execute software instructions stored on nontransitory machine-readable storage media that (The instructions are stored in the instruction memory in the reduced instruction-set computer (RISC) controller, pg. 469, left col., second to the last para.):
The same motivation to combine independent claim 1 applies here.
Dong and Yuan does not explicitly teach stabilize monitor network performance during architectural changes while implementing temporary support structures during transitions and maintaining backup pathways that enable potential reversion of modifications.
Abraham teaches stabilize monitor network performance during architectural changes while implementing temporary support structures during transitions while implementing temporary support structures during transitions (As shown in Fig. 4, for each model we measure the time it takes to (1) load from disk, (2) perform one inference, and (3) to swap in the full model in place of the pruned model currently running (or vice versa), by reverse(), pg. 5, left col., last sentence to pg. 5, right col.)
and maintaining backup pathways that enable potential reversion of modifications (As shown in Fig. 2, Back to the Future consists of an offline portion for pruning and storing the architectures, and a runtime function, reverse(), to swap between full and pruned architectures., pg. 2, right col., last para.)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Abraham for the benefit of enhancing safety and reliability, and providing seamless reversion to the accurate version of the model, demonstrating its applicability for safe autonomous systems design (Abraham, abstract).
Regarding claim 19, Dong and Yuan teaches the method of claim 13, they do not teach explicitly teach the limitations of claim 19.
Abraham teaches further comprising the step of managing stability by monitoring network performance during architectural changes while implementing temporary support structures during transitions (As shown in Fig. 4, for each model we measure the time it takes to (1) load from disk, (2) perform one inference, and (3) to swap in the full model in place of the pruned model currently running (or vice versa), by reverse(), pg. 5, left col., last sentence to pg. 5, right col.)
and maintaining backup pathways that enable potential reversion of modifications (As shown in Fig. 2, Back to the Future consists of an offline portion for pruning and storing the architectures, and a runtime function, reverse(), to swap between full and pruned architectures., pg. 2, right col., last para.).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Abraham for the benefit of enhancing safety and reliability, and providing seamless reversion to the accurate version of the model, demonstrating its applicability for safe autonomous systems design (Abraham, abstract).
8. Claims 9, 10, 12, 21, 22 and 24 are rejected under 35 U.S.C. 103 as being unpatentable over Dong et al. ("Spartan: A Sparsity-Adaptive Framework to Accelerate Deep Neural Network Training on GPUs, Date of Publication: 22 March 2021) in view of Yuan et al. ("STICKER: An energy-efficient multi-sparsity compatible accelerator for convolutional neural networks in 65-nm CMOS." IEEE Journal of Solid-State Circuits 55.2 (2019): 465-477) in view of Kim et al. ("LSTM-based fault direction estimation and protection coordination for networked distribution system." IEEE Access 10 (2022): 40348-40357, date of publication April 12, 2022, date of current version April 20, 2022).
Regarding claim 9, Dong and Yuan teaches the computer system of claim 1, they do not explicitly teach the limitations of claim 9.
Kim teaches wherein the pattern database maintains contextual signatures for stored patterns, enabling relevant pattern retrieval for similar operational scenarios (Because a separate structure determines whether to store historical time series data, LSTM (long short-term memory) shows excellence in the retention of relevant information, pg. 40350, left col., first para.; The Examiner notes in the LSTM, memory of past input is retrieved for solving sequence learning tasks)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Kim for the benefit of a high-speed communication systems in a network distribution system comprising a long short-term memory (LSTM) neural network (Kim abstract)
Regarding claim 10, Dong and Yuan teaches the computer system of claim 1, Yuan wherein the computer system is further configured to execute software instructions stored on nontransitory machine-readable storage media that (The instructions are stored in the instruction memory in the reduced instruction-set computer (RISC) controller, pg. 469, left col., second to the last para.)
The same motivation to combine independent claim 1 applies here.
Dong and Yuan does not explicitly teach implement adaptive thresholds for resource allocation based on current network state and performance requirements.
Kim teaches implement adaptive thresholds for resource allocation based on current network state and performance requirements (The parameters of the LSTM network are adjusted and optimized using the adaptive moment estimation (ADAM) method. 128 mini-batch-sized feature data are randomly extracted for training. In addition, we set the initial learning rate to 0.001, the gradient decay factor to 0.9, and the squared gradient decay factor to 0.999, pg. 40353, left col., second to the last para.).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Kim for the benefit of a high-speed communication systems in a network distribution system comprising a long short-term memory (LSTM) neural network (Kim abstract)
Regarding claim 12, Dong and Yuan teaches the computer system of claim 1, Yuan teaches wherein the computer system is further configured to execute software instructions stored on nontransitory machine-readable storage media that (The instructions are stored in the instruction memory in the reduced instruction-set computer (RISC) controller, pg. 469, left col., second to the last para.):
The same motivation to combine independent claim 1 applies here.
Dong and Yuan does not explicitly teach implement hierarchical circuit breakers coordinating across supervisory levels to isolate and address potential instabilities.
Kim teaches implement hierarchical circuit breakers (circuit breaker A [Wingdings font/0xE0] circuit breaker B [Wingdings font/0xE0] circuit breaker c, Fig. 2, pg. 40351) coordinating across supervisory levels to isolate and address potential instabilities (Utilization of the LSTM algorithm is anticipated to yield significant reductions in error when determining fault direction, which is the major cause of circuit breaker malfunction issues in the existing CLS protection coordination algorithm, pg. 40350, left col., first para.).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Kim for the benefit of a high-speed communication systems in a network distribution system comprising a long short-term memory (LSTM) neural network (Kim abstract)
Regarding claim 21, Dong and Yuan teaches the method of claim 13, they do not explicitly teach the limitations of claim 21.
Kim teaches further comprising the step of maintaining contextual signatures for stored patterns, enabling relevant pattern retrieval for similar operational scenarios (Because a separate structure determines whether to store historical time series data, LSTM (long short-term memory) shows excellence in the retention of relevant information, pg. 40350, left col., first para.; The Examiner notes in the LSTM, memory of past input is retrieved for solving sequence learning tasks)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Kim for the benefit of a high-speed communication systems in a network distribution system comprising a long short-term memory (LSTM) neural network (Kim abstract)
Regarding claim 22, Dong and Yuan teaches the method of claim 13, they do not explicitly teach the limitations of claim 22.
Kim teaches further comprising the step of managing resources by implementing adaptive thresholds for resource allocation based on current network state and performance requirements (The parameters of the LSTM network are adjusted and optimized using the adaptive moment estimation (ADAM) method. 128 mini-batch-sized feature data are randomly extracted for training. In addition, we set the initial learning rate to 0.001, the gradient decay factor to 0.9, and the squared gradient decay factor to 0.999, pg. 40353, left col., second to the last para.).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Kim for the benefit of a high-speed communication systems in a network distribution system comprising a long short-term memory (LSTM) neural network (Kim abstract)
Regarding claim 24, Dong and Yuan teaches the method of claim 13, they do not explicitly teach the limitations of claim 24.
Kim teaches further comprising the step of implementing hierarchical circuit breakers (circuit breaker A [Wingdings font/0xE0] circuit breaker B [Wingdings font/0xE0] circuit breaker c, Fig. 2, pg. 40351) coordinating across supervisory levels to isolate and address potential instabilities (Utilization of the LSTM algorithm is anticipated to yield significant reductions in error when determining fault direction, which is the major cause of circuit breaker malfunction issues in the existing CLS protection coordination algorithm, pg. 40350, left col., first para.).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Kim for the benefit of a high-speed communication systems in a network distribution system comprising a long short-term memory (LSTM) neural network (Kim abstract)
9. Claims 11 and 23 are rejected under 35 U.S.C. 103 as being unpatentable over Dong et al. ("Spartan: A Sparsity-Adaptive Framework to Accelerate Deep Neural Network Training on GPUs, Date of Publication: 22 March 2021) in view of Yuan et al. ("STICKER: An energy-efficient multi-sparsity compatible accelerator for convolutional neural networks in 65-nm CMOS." IEEE Journal of Solid-State Circuits 55.2 (2019): 465-477) in view of Hamm et al. ("Global optimization of neural network weights." Proceedings of the 2002 International Joint Conference on Neural Networks. IJCNN'02 (Cat. No. 02CH37290). Vol. 2. IEEE, 2002).
Regarding claim 11, Dong and Yuan teaches the computer system of claim 1, Yuan teaches wherein the computer system is further configured to execute software instructions stored on nontransitory machine-readable storage media that (The instructions are stored in the instruction memory in the reduced instruction-set computer (RISC) controller, pg. 469, left col., second to the last para.):
The same motivation to combine independent claim 1 applies here.
Dong and Yuan does not explicitly teach implement both local and global optimization strategies ensuring that adaptations beneficial in one region maintain overall network performance.
Hamm teaches implement both local and global optimization strategies ensuring that adaptations beneficial in one region maintain overall network performance (Therefore, it is common to combine a local algorithm with a global algorithm by using the weights obtained from the global algorithm as starting values for the local routine, pg. 1230, right col., last para.).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Hamm for the benefit of more efficient use of computational resources (Hamm abstract)
Regarding claim 23, Dong and Yuan teaches the method of claim 13, they do not explicitly teach the limitations of claim 23.
Hamm teaches further comprising the step of implementing both local and global optimization strategies ensuring that adaptations beneficial in one region maintain overall network performance (Therefore, it is common to combine a local algorithm with a global algorithm by using the weights obtained from the global algorithm as starting values for the local routine, pg. 1230, right col., last para.).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Dong and Yuan to incorporate the teachings of Hamm for the benefit of more efficient use of computational resources (Hamm abstract)
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MORIAM MOSUNMOLA GODO whose telephone number is (571)272-8670. The examiner can normally be reached Monday-Friday 8:00am-5:00pm EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Michelle T. Bechtold can be reached on (571) 431-0762. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/M.G./Examiner, Art Unit 2148
/MICHELLE T BECHTOLD/Supervisory Patent Examiner, Art Unit 2148
Prosecution Insights