DETAILED ACTION
1. This office action is in response to the Application No. 19205960 filed on 02/18/2026. Claims 1-18 are presented for examination and are currently pending.
Response to Arguments
2. The claim amendment of 02/18/2026 has overcome the 112(f) claim interpretation of 11/26/2025. As a result, the 112(f) claim interpretation has been withdrawn.
On pages 6-7 of the remarks, the Applicant argued that “First, the claims do not
recite an abstract idea under Step 2A, Prong 1. All of the claim limitations recite specific technological implementations, notably hierarchical supervisory monitoring with real-time architectural modifications, designated sleep states executing specific optimization operations, persistent state management systems storing and retrieving neural activation patterns and architectural configurations across operational sessions, and signal transmission pathways providing direct connections between non-adjacent network regions with temporal coordination during transmission, all of which improve neural network functionality. Under the August 2025 Kim Memorandum, AI inventions integrating specific architectures and technical implementations that solve technical problems are not abstract ideas at Step 2A, Prong 1. Further, claim limitations do not recite an abstract idea falling within the judicial exception of "mental processes" if they cannot be practically performed in the human mind. Id. Here, none of the claim limitations can be practically performed in the human mind”.
The improvement to the neural network functionality argued by the Applicant is not persuasive because these improvement is a conclusory statement with no details as to how the claimed invention leads to the improvement alleged by the Applicant.
The detailed analysis of the 101 rejection in this Office Action is in line with the August 2025 Kim Memorandum.
Furthermore, as highlighted in the detailed 101 rejection, the “monitoring the neural network through multiple supervisory levels”, “hierarchical supervisory system identifies operation patterns”, “implements architectural changes” and so on are claim limitations that can be practically performed in the human mind.
On page 7 of the remarks, the Applicant argued that “Second, even if some claim limitations are deemed to recite an abstract idea, the claims integrate them into practical applications at Step 2A, Prong 2 by improving how neural networks operate through hierarchical supervisory monitoring, sleep state optimization, and persistent state management across operational sessions. Under Ex Parte Desjardins, Appeal No. 2024-000567 (USPTO Sept. 26, 2025), claims integrate abstract ideas into practical applications when the specification identifies concrete improvements to computer technology, specific claim limitations reflect those improvements, and the improvements are to how the computer system itself operates, all of which are present here”.
The alleged statement that the hierarchical supervisory monitoring, sleep state optimization, and persistent state management across operational sessions improves how neural networks operate is not persuasive because these are conclusory statements that include no details how the claimed limitations achieve the improvement argued by the Applicant.
Furthermore, according to MPEP 2106(d)(1), “the specification need not explicitly set forth the improvement, but it must describe the invention such that the improvement would be apparent to one of ordinary skill in the art. Conversely, if the specification explicitly sets forth an improvement but in a conclusory manner (i.e., a bare assertion of an improvement without the detail necessary to be apparent to a person of ordinary skill in the art), the examiner should not determine the claim improves technology. Second, if the specification sets forth an improvement in technology, the claim must be evaluated to ensure that the claim itself reflects the disclosed improvement. That is, the claim includes the components or steps of the invention that provide the improvement described in the specification”. Here, the Applicant has presented the improvement in a conclusory manner. Also, the instant claim discloses no details or steps about how the hierarchical supervisory monitoring, the sleep state optimization, and the persistent state management across operational sessions improves the operation of the neural network.
On page 8 of the remarks, the Applicant argued that “A. The Claims Recite Specific Technological Implementations as in the August 2025 Kim Memorandum. The claims recite specific technological implementations that improve neural network functionality, not abstract ideas. The claim limitations do not fall under the judicial exception of mental processes because they recite "specific AI architectures, training methodologies, or technical implementations" as set forth in the August 2025 Kim Memorandum. The limitations specify: the Al architecture (hierarchical supervisory levels monitoring the neural network), the data structures (neural activation patterns and architectural configurations maintained across operational sessions), the operational mechanisms (designated sleep states executing specific optimization operations rather than continuous active processing), and the technical processing techniques (neural memory consolidation, neural insight generation, neural pruning coordination, and neural memory reorganization)”.
As stated earlier, that the neural memory consolidation, neural insight generation, neural pruning coordination, the neural memory reorganization are broadly recited limitations that lack details as to how these limitations improve neural network functionality.
Furthermore, the argued limitation data structures or the operational mechanisms (designated sleep states executing specific optimization operations rather than continuous active processing) are not reflected in the claims. It appears the Applicant is arguing what is not claimed.
In addition, the Applicant’s argument that the limitations of the claims improve the functionality of the neural network is not persuasive because the claimed limitation include no details or steps to arrive at the alleged improvements.
On page 8 of the remarks, the Applicant argued that “The specification confirms this technological focus. The Background identifies technical problems: "When a neural network is shut down or restarted, its operational state and learned patterns are typically lost unless explicitly saved as model weights, requiring complete reloading and reinitialization." [0003]. "Neural networks currently lack sophisticated mechanisms for self-optimization during periods of reduced operational demand. Unlike biological neural systems that utilize sleep states for memory consolidation and cognitive reorganization, artificial neural networks typically perform optimization only during explicit training phases." [0004]. The invention provides a specific technological solution to these problems”.
The Applicant’s argument is not persuasive because the claims lack the details that is required according to the citations of the specification above to arrive at the alleged improvements.
Furthermore, according to MPEP 2106(d)(1), “Second, if the specification sets forth an improvement in technology, the claim must be evaluated to ensure that the claim itself reflects the disclosed improvement. That is, the claim includes the components or steps of the invention that provide the improvement described in the specification”. Here, the instant claim discloses no details or steps about how the hierarchical supervisory monitoring or the sleep state optimization or the memory consolidation provides a specific solution to the technological field.
On page 8-9 of the remarks, the Applicant argued that “B. The Claims Do Not Recite the Judicial Exception Of A "Mental Process" Because The Claim Limitations Cannot Be Practically Performed In The Human Mind. The Examiner's characterization that these operations could be performed "by observing the behavior" or "making a judgement" does not accurately describe the claim limitations. Humans cannot: Monitor the neural network through multiple supervisory levels, wherein monitoring the neural network includes collecting activation data, identifying operation patterns, implementing architectural changes, detecting network sparsity, coordinating pruning decisions, and managing resource redistribution: Humans cannot simultaneously monitor thousands of neural connections across multiple hierarchical levels in real-time while collecting activation data from interconnected nodes, identifying operation patterns across these nodes, implementing architectural changes to the network structure, detecting sparsity conditions, coordinating pruning decisions across multiple levels, and managing computational resource redistribution. These operations require processing activation data from numerous interconnected nodes simultaneously, performing statistical analysis to identify patterns, and executing architectural modifications while maintaining network stability-operations that are computationally intensive and fundamentally cannot be performed through mental processes.”
The argument that the human mind cannot perform simultaneous monitoring of thousands of neural connections across multiple hierarchical levels in real-time is not persuasive. This is because the simultaneous monitoring of neural connections across multiple hierarchical levels in real-time is not claimed.
Furthermore, according to MPEP 2106.04(a)(2)(III): “Nor do the courts
distinguish between claims that recite mental processes performed by humans and claims that recite mental processes performed on a computer. As the Federal Circuit has explained, “[c]ourts have examined claims that required the use of a computer and still found that the underlying, patent-ineligible invention could be performed via pen and paper or in a person’s mind.”
On page 9 of the remarks, the Applicant argued that “Humans cannot: Execute optimization operations during designated sleep states, wherein the optimization operations include at least one of neural memory consolidation, neural insight generation, neural pruning coordination, and neural memory reorganization: Humans cannot execute neural memory consolidation by evaluating neural pathways based on computational importance factors and systematically strengthening connections while maintaining mathematical stability constraints across the network. Humans cannot perform neural insight generation by discovering non-obvious connections between different network regions through systematic analysis of activation correlations. Humans cannot coordinate neural pruning across multiple supervisory levels while ensuring network stability. Humans cannot execute neural memory reorganization by optimizing network structure to improve information flow efficiency. These optimization operations require computational analysis of network-wide patterns and precise mathematical modifications that cannot be accomplished through mental effort”.
The argument above is not persuasive because the arguments are directed toward the abstract ideas. The claimed limitations like “neural memory consolidation”, “neural insight generation”, “neural pruning coordination”, and “neural memory reorganization” are so broad that it is interpreted has operations that can be performed in the human mind.
Furthermore, according to MPEP 2106.04(a)(2)(III): “Nor do the courts
distinguish between claims that recite mental processes performed by humans and claims that recite mental processes performed on a computer. As the Federal Circuit has explained, “[c]ourts have examined claims that required the use of a computer and still found that the underlying, patent-ineligible invention could be performed via pen and paper or in a person’s mind.”
In addition, according to MPEP 2106.05(a): “It is important to note, the judicial exception alone cannot provide the improvement. The improvement can be provided by one or more additional elements. See the discussion of Diamond v. Diehr, 450 U.S. 175, 187 and 191-92, 209 USPQ 1, 10 (1981)) in subsection II, below. In addition, the improvement can be provided by the additional element(s) in combination with the recited judicial exception”.
On pages 9-10 of the remarks, the Applicant argued that “Humans cannot: Maintain persistent neural network state through a state management system that stores and retrieves neural activation patterns and architectural configurations across operational sessions: Humans cannot serialize the complete state of a neural network including activation patterns across all nodes and architectural configurations including connection weights and thresholds, store this high-dimensional state representation, and then precisely restore this state to enable the network to resume operation from its previous state after system restart. This requires capturing and encoding thousands or millions of numerical values representing network state, storing them in a format enabling exact restoration, and systematically reloading and reactivating the network to its previous operational state-operations requiring computational infrastructure that fundamentally cannot be replicated through human mental processes”.
The argument above is not persuasive because the “maintain persistent neural network state ...”, the “neural state serialization system that captures and stores the state of the neural architecture” has analyzed in the 101 rejection are additional elements not abstract ideas.
Furthermore, the serialization of the state of neural network that includes connection weights and thresholds, store this high-dimensional state representation is not reflected in the claims. It appears the Applicant is arguing what is not claimed.
In addition, according to MPEP 2106.04(a)(2)(III): “Nor do the courts
distinguish between claims that recite mental processes performed by humans and claims that recite mental processes performed on a computer. As the Federal Circuit has explained, “[c]ourts have examined claims that required the use of a computer and still found that the underlying, patent-ineligible invention could be performed via pen and paper or in a person’s mind.”
On page 10 of the remarks, the Applicant argued that “Humans cannot: Manage signal transmission pathways providing direct connections between non-adjacent network regions with signal modification and temporal coordination during transmission: Humans cannot establish and manage direct communication pathways between network regions that are not adjacent in the standard network architecture, apply signal transformations during transmission through these pathways, and coordinate the timing of signal propagation to ensure temporal synchronization. This requires creating bypass connections in the network topology, implementing mathematical transformations on signals during transmission, and managing temporal coordination to ensure signals arrive at appropriate times-operations requiring real- time computational control that cannot be performed mentally”.
The argument above is not persuasive because the limitations “manage signal transmission pathways providing direct connections between non-adjacent network regions with signal modification and temporal coordination during transmission” are so broad and interpreted has a process that encompasses human mind.
Furthermore, the limitation “applying signal transformations during transmission ...” are not reflected in the claims. It appears the Applicant is arguing what is not claimed.
In addition, according to MPEP 2106.04(a)(2)(III): “Nor do the courts
distinguish between claims that recite mental processes performed by humans and claims that recite mental processes performed on a computer. As the Federal Circuit has explained, “[c]ourts have examined claims that required the use of a computer and still found that the underlying, patent-ineligible invention could be performed via pen and paper or in a person’s mind.”
On page 10 of the remarks, the Applicant argued that “Humans cannot: Manage operational states of the neural network and coordinate decision-making across the hierarchical supervisory levels: Humans cannot manage transitions between different operational states of the neural network including active processing states and designated sleep states while coordinating decision-making across multiple hierarchical levels of supervision. This requires tracking the operational status of numerous network components, making coordinated decisions about state transitions across multiple supervisory levels, and ensuring coherent operation during transitions-operations requiring computational coordination across distributed components that fundamentally cannot be accomplished through mental processes”.
The argument above is not persuasive because the limitations “manage operational states of the neural network and coordinate decision-making across the hierarchical supervisory levels”, “tracking the operational status” are so broad and interpreted has a process that can encompass the human mind.
Furthermore, the limitations “manage transitions between different operational states of the neural network including active processing states” are not reflected in the claims. It appears the Applicant is arguing what is not claimed.
In addition, according to MPEP 2106.04(a)(2)(III): “Nor do the courts
distinguish between claims that recite mental processes performed by humans and claims that recite mental processes performed on a computer. As the Federal Circuit has explained, “[c]ourts have examined claims that required the use of a computer and still found that the underlying, patent-ineligible invention could be performed via pen and paper or in a person’s mind.”
On page 10 of the remarks, the Applicant argued that “C. The Invention Addresses Computer-Specific Problems. The claimed invention addresses problems "specifically arising in the realm of" computer- implemented neural networks: loss of operational state across system restarts, inability to optimize during reduced demand periods, lack of sophisticated self-optimization mechanisms beyond explicit training, and inability to maintain dynamically evolving network architectures during operation”.
On pages 10-11 of the remarks, the Applicant argued that “II. Step 2A, Prong 2: The Claims Integrate Abstract Ideas Into Practical Applications. A. The Specification Identifies Concrete Technical Problems. The specification identifies specific technical problems: Loss of operational state: "When a neural network is shut down or restarted, its operational state and learned patterns are typically lost unless explicitly saved as model weights, requiring complete reloading and reinitialization." [0003] Lack of sleep-state optimization: "Neural networks currently lack sophisticated mechanisms for self-optimization during periods of reduced operational demand. Unlike biological neural systems that utilize sleep states for memory consolidation and cognitive reorganization, artificial neural networks typically perform optimization only during explicit training phases." [0004] Rigid architectures: "Most neural architectures operate with rigid structures that cannot dynamically adapt based on observed patterns or resource constraints. While pruning techniques exist, they typically require offline processing rather than real-time adaptation during operation." [0005] Limited hierarchical supervision: "The lack of hierarchical supervision across multiple levels further limits the sophistication of architectural modifications that can be safely implemented during runtime." [[0006]”.
On pages 11-12 of the remarks, the Applicant argued that “B. The Improvements Are to Neural Network Operation Following Desjardins' guidance against excessive generality, the proper characterization is "neural networks that maintain operational continuity across sessions and optimize during designated sleep states through hierarchical supervisory monitoring"-an improvement to computer technology. The elements work together as an integrated system. The hierarchical supervisory levels monitor network operation and coordinate modifications across multiple levels. The sleep state mechanism enables the network to enter designated operational modes for executing optimization operations. The persistent state management serializes and restores neural activation patterns and architectural configurations enabling continuity across sessions. The signal transmission pathways provide direct connections between non-adjacent regions with temporal coordination. Therefore, under MPEP § 2106.04(d)(1) and the framework established in Desjardins, these improvements integrate any recited abstract ideas into a practical application”.
The above arguments are not persuasive because the limitations in the claims are so broad, the claims lack details or steps for a person of ordinary skill in the art to carry out these claimed features to arrive at the improvements argued by the Applicant.
Furthermore, according to MPEP 2106(d)(1), “Second, if the specification sets forth an improvement in technology, the claim must be evaluated to ensure that the claim itself reflects the disclosed improvement. That is, the claim includes the components or steps of the invention that provide the improvement described in the specification”. Here, the instant claim discloses no details or steps about how the persistent state management serializes and restores neural activation patterns or the signal transmission pathways provide direct connections between non-adjacent regions with temporal coordination to provide the specific improvements to the neural network alleged by the Applicant.
On page 12 of the remarks, the Applicant argued that “Conclusion. For the foregoing reasons, the independent claims satisfy §101: 1. The claims are not directed to abstract ideas at Step 2A, Prong 1 because they recite specific technological implementations that improve neural network functionality, solving technical problems specifically arising in computer-implemented neural networks. 2. The claims integrate abstract ideas into practical applications at Step 2A, Prong 2 because the specification identifies concrete technical problems, specific claim limitations reflect technological improvements, and the improvements are to how neural networks themselves operate. Since the independent claims are patentable, dependent claims are patentable by virtue of dependency. Therefore, Applicant respectfully requests that the rejection under §101 be withdrawn”.
The argument above is not persuasive because as noted above the limitations of the claimed invention do not reflect the disclosed improvement argued by the Applicant. As a result, the 101 rejection is maintained and adjust to reflect the amendments.
As regards to the prior art rejection, the Applicant argued on page 13 that “Applicant respectfully traverses the rejection of claims 1-18 under §103 as being unpatentable over Jain in view of Rostami. The combination fails to teach or suggest multiple key limitations of amended Claim 1, either alone or in combination. Examiner has improperly mapped generic computer functions and unrelated mobile device power management techniques to Applicant's specific integrated architecture for neural network optimization during designated operational states”.
The Applicant has also argued on page 13 that As detailed below, “the cited references do not teach the claimed system wherein optimization operations including neural memory consolidation, neural insight generation, neural pruning coordination, and neural memory reorganization are executed during designated sleep states that are distinct operational modes of the neural network itself. Furthermore, the references fail to teach the claimed integration of hierarchical supervisory level management with persistent state maintenance across operational sessions in the manner recited. The claimed invention represents a novel architecture that fundamentally differs from both neural network compression systems and mobile device power management by providing a neural network that maintains operational continuity across sessions while performing sophisticated optimization during designated internal operational states”.
The arguments above are not persuasive because the Applicant has not provided arguments above about how the obvious teachings of Jain in view of Rostami in the Office Action does not teach the claimed limitations.
On pages 13-14 of the remarks, the Applicant argued that “The references fail to teach at least the following limitations: A. Execute Optimization Operations During Designated Sleep States The Examiner has failed to establish that Jain in view of Rostami teaches or suggests executing optimization operations during designated sleep states as claimed. Rostami discloses wake-up scheduling for mobile devices to minimize power consumption by extending sleep periods when buffering delay approaches a maximum threshold. This is conventional device-level power management for battery conservation in wireless communication devices. The cited portions describe a mobile device entering a low-power state when network traffic is minimal, not a neural network entering a designated operational state to perform internal optimization operations. Rostami's sleep states are system-wide power-saving modes triggered by external traffic conditions, whereas the claimed sleep states are operational modes of the neural network during which specific optimization operations including neural memory consolidation, neural insight generation, neural pruning coordination, and neural memory reorganization are actively executed. The claimed invention performs optimization operations during these sleep states, which is fundamentally different from merely reducing power consumption or buffering network traffic. The Examiner has not explained how or why one skilled in the art would modify Jain's compression system to implement designated operational states for executing these specific optimization operations, nor how Rostami's traffic-based device sleep scheduling suggests such a modification. Neither reference teaches a neural network that transitions into a designated operational state specifically designed for executing optimization operations”.
The arguments above are not persuasive because as cited in the Office Action Rostami’s teaching of In duty cycling,…nodes wake up and sleep periodically (pg. 6021, left col., first para.)... A multi-step Long Short-Term Memory (LSTM) neural network is trained with data from real user applications and tailored for traffic prediction purposes, pg. 6021, right col., first para.) reads on the broadly claimed “sleep states” of Applicant.
Furthermore, since Jain as primary reference desires a neural network with lower power consumption [0476] and making predictions [0523], while Rostami as secondary reference also teaches a neural network and minimizing power consumption (pg. 6026, right col., last para. to pg. 6027, left col. first para.), then 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 Jain to incorporate the teachings of Rostami for the benefit of prediction purposes (Rostami, pg. 6021, second to the last para.).
Furthermore, Jain’s teachings meets the claimed limitation of optimization operations include at least one of neural memory consolidation, neural insight generation, neural pruning coordination, and neural memory reorganization (For example, by pruning particular neurons the corresponding weights do not need to be stored, thus enabling storage savings for the model [0476]).
On page 14 of the remarks, the Applicant argued that “The references fail to teach at least the following limitations: B. Maintain Persistent Neural Network State Across Operational Sessions The cited references do not teach or suggest maintaining persistent neural network state through a state management system that stores and retrieves neural activation patterns and architectural configurations across operational sessions as claimed. The Examiner cites generic storage capabilities in Jain and basic model retrieval functions. However, storing a compressed model is fundamentally different from maintaining persistent neural network state that includes neural activation patterns and architectural configurations across operational sessions”.
The argument above is not persuasive because the broadly claimed limitation of “maintain persistent neural network state” is taught by Jain since the Applicant has not recited specific claim limitations about how the state management system works. Jain teaches maintain persistent neural network state through a state management system that stores (To provide for persistent storage of information such as data, applications, operating systems and so forth, a storage F458 may also couple to the processor F452 via the interconnect F456 [0188]) and
retrieves neural activation patterns (In operation, an example input receiver circuitry ID6_102 is an example structure that receives, retrieves and/or otherwise obtains inputs [0480]).
The Applicant has also argued “The claimed limitation requires that the state management system stores and retrieves both neural activation patterns and architectural configurations such that the neural network's operational state persists across sessions. The claimed invention enables the neural network to maintain continuity of its operational state when transitioning between distinct operational sessions, including system shutdowns and restarts. Jain's disclosure of saving compressed models and retrieving them does not teach maintaining the neural network's activation patterns and architectural configurations in a manner that enables the network to resume operation from a previous state after an operational session ends and a new session begins. The cited backup and restoration of weights during a restoration operation relates to error recovery within a single session, not persistence of neural network state across distinct operational sessions as claimed. Neither reference teaches or suggests a state management system that maintains the operational continuity claimed by storing and retrieving both neural activation patterns and architectural configurations across operational sessions”.
The argument above is not persuasive that because Jain teaches initial data for the neural network [0574], and Lessons learned from one deployment can be fed forward so that future neural network searches can learn from prior data collection and analysis [0418]. This indicates Jain teaches a state management system a state management system operational continuity by storing and retrieving data through the interconnect, as Jain also teaches (To provide for persistent storage of information such as data, applications, operating systems and so forth, a storage F458 may also couple to the processor F452 via the interconnect F456 [0188]).
On pages 14-15 of the remarks, the Applicant argued that “The references fail to teach at least the following limitations: C. Integration of Monitoring Through Multiple Supervisory Levels With Sleep State Optimization. The Examiner has not established that the combination teaches the claimed integration wherein monitoring the neural network through multiple supervisory levels that includes collecting activation data, identifying operation patterns, implementing architectural changes, detecting network sparsity, coordinating pruning decisions, and managing resource redistribution operates in conjunction with executing optimization operations during designated sleep states. Jain discloses hierarchical agents for compression decisions and Rostami discloses device sleep scheduling, but neither reference alone or in combination teaches how monitoring through multiple supervisory levels that performs the recited functions integrates with executing optimization operations during designated sleep states. The claimed invention requires a unified architecture where the monitoring functions and the sleep state optimization functions are integrated components of a single system that manages the neural network across different operational modes. The claimed system requires that the same architecture that monitors the neural network through multiple supervisory levels also executes optimization operations during designated sleep states, creating a unified system that transitions between operational modes. The Examiner has mapped Jain's compression agents to the supervisory levels and Rostami's device sleep to the sleep states, but has not explained how these disparate teachings from different technical domains would be combined to create the claimed integrated system where the supervisory monitoring functions operate in coordination with sleep state optimization operations. The claimed invention is not simply the combination of a compression system and a power management system, but rather represents a novel integrated architecture where monitoring and optimization functions are coordinated across operational states”.
The argument above is not persuasive because Jain as primary reference also teaches sleep states because Jain teaches of CNNs may identify redundancies in the network during training and statically remove the redundancies to obtain a final network configuration. [0642]. It is noted that instant specification of Applicant recites “Neural pruning coordinator 3630 works during sleep states to identify underutilized neural components (US20250363367 [0531])... distinguishing between truly redundant components and those with occasional but critical functions (US20250363367 [0532]).
Furthermore, the argument that neither reference alone or in combination teaches how monitoring through multiple supervisory levels is not persuasive because as detailed in the Office Action,
Jain teaches implement a hierarchical supervisory system (agent A ID6_602 → agent B ID6_604 → agent C ID6_606 (FIG. ID6_6); FIG. ID6_6 is the same diagram as FIG.ID6_2 [0525]) monitoring the neural network through multiple supervisory levels (Furthermore, the example compression environment performer circuitry ID6_116 provides feedback on post-compression accuracy and network dynamic observations to the agent ID6_110 [0487]; An example agent A ID6_202 is an example of the agent ID6_110 [0499]; An example agent B ID6_206 is an example of the agent ID6_110 [0500]; An example agent C ID6_210 is an example of the agent ID6_110 [0501]),
implement a hierarchical supervisory system (agent A ID6_602 → agent B ID6_604 → agent C ID6_606 (FIG. ID6_6); FIG. ID6_6 is the same diagram as FIG.ID6_2 [0525]) monitoring the neural network through multiple supervisory levels (Furthermore, the example compression environment performer circuitry ID6_116 provides feedback on post-compression accuracy and network dynamic observations to the agent ID6_110 [0487]; An example agent A ID6_202 is an example of the agent ID6_110 [0499]; An example agent B ID6_206 is an example of the agent ID6_110 [0500]; An example agent C ID6_210 is an example of the agent ID6_110 [0501]).
On pages 15-16 of the remarks, the Applicant argued that “The references fail to teach at least the following limitations: D. Managing Operational States and Coordinating Decision-Making Across Hierarchical Supervisory Levels. The cited references do not teach managing operational states of the neural network and coordinating decision-making across the hierarchical supervisory levels as claimed. The Examiner cites Jain's orchestrator circuit that receives workloads with quality of service requirements and manages resources. However, Jain's orchestrator manages workload distribution and resource allocation for inference tasks, not operational states of the neural network itself. The claimed limitation requires managing operational states of the neural network, which necessarily includes the transition between different modes of operation such as active processing and the designated sleep states during which optimization operations are executed. The claimed invention provides management of operational states that defines how the neural network functions in different modes, not merely how workloads are distributed. Furthermore, the claimed limitation requires coordinating decision-making across the hierarchical supervisory levels, creating an integrated control structure. The Examiner has not identified where Jain's orchestrator coordinates decision-making across Jain's hierarchical agents in the manner claimed, particularly with respect to managing transitions between operational states including the designated sleep states. Neither reference teaches the claimed coordination of decision-making across hierarchical supervisory levels in the context of managing operational states of the neural network itself”.
The argument above is not persuasive because the broadly claimed limitations of the Applicant are taught by Jain. Since the Applicant has not recited claim limitations that provide specific computational details of how the broadly claimed “manages operational states of the neural network and coordinates decision-making” occurs, then Jain’s teachings reads on the claimed invention.
Jain teaches implement a cognitive neural orchestrator (the example orchestrator circuit ID3_104 of FIGS. ID3_1 and ID3_3 [0317]) that manages operational states of the neural network and coordinates decision-making across the hierarchical supervisory system (the means for orchestrating ID3_104 includes means for data interfacing, means for node availability determining, means for request validating, means for model generating, means for resource managing [0316]; the example data interface circuit ID3_308 of the example orchestrator circuit ID3_104 receives a workload (e.g., instance, app, artificial intelligence model) with a user-defined quality of service from a tenant ID3_304. For example, the quality of service may be a function of frequency, cache, memory bandwidth, power, DL precision (e.g., INT8, BF16), DL model characteristics, and migration-tolerance [0325]);
On page 16 of the remarks, the Applicant argued that “The references fail to teach at least the following limitations: E. Signal Transmission Pathways With Temporal Coordination During Transmission. The Examiner has failed to establish that the references teach managing signal transmission pathways providing direct connections between non-adjacent network regions with signal modification and temporal coordination during transmission. Examiner cites Jain's disclosure of data propagating along a data path such as a bus and agent reassignment for alternate paths. Generic computer bus architecture and workload reassignment do not teach signal transmission pathways that provide direct connections between non-adjacent network regions with signal modification and temporal coordination during transmission as claimed. The claimed limitation requires managing pathways that connect non-adjacent regions, meaning regions that are not directly connected in the standard network architecture, and further requires that these pathways include both signal modification and temporal coordination during transmission. The claimed invention provides direct connections between regions that would otherwise require signal propagation through intermediate network layers, and manages these connections with specific signal modification and temporal coordination during transmission. Jain's references to data buses and alternate execution paths do not disclose or suggest this specific structural arrangement and functional coordination. Examiner has not explained how Jain's compression framework or Rostami's wake scheduling would teach or suggest creating direct pathways between non-adjacent network regions with the claimed signal modification and temporal coordination capabilities. Neither reference teaches managing signal transmission pathways that bypass standard network architecture to connect non-adjacent regions while providing signal modification and temporal coordination”.
The Applicant’s arguments above are not persuasive because the Office Action clearly indicated that Jain teaches the claimed “managing signal transmission pathways providing direct connections between non-adjacent network regions with signal modification and temporal coordination during transmission”. As indicated in the Office Action, Jain’s graph represented as a neural architecture is a transmission pathway which transmits signals across layers of the neural network.
Jain teaches In some examples, an optimized graph is represented as a neural architecture having a particular number of layers, connections (nodes), weights and/or hyperparameters … Such data may also propagate from a first data structure to a second or any number of subsequent data structures along a data path, such as a bus [0219]).
Furthermore, Jain’s teachings of selecting an alternate path reads on the claimed “direct connections between non-adjacent network regions”. Jain teaches (Considering that one or more conditions have changed, the example agent managing circuitry ID5_C310 assigns another agent to re-assess performance of the selected alternate path [0229]; The per-layer compression policy is a way to gain higher performance while preserving the accuracy of a neural network. Compression of neural networks reduces the amount of data transfer and computing during the execution of the neural network [0484]; there is a distinct advantage with Example B ID6_412 where the scale of delivery of compressed models for a given turn around time/lead shows a temporal improvement for finding optimal compression policies [0506]);
On page 17 of the remarks, the Applicant argued that “For at least the reasons set forth above, the § 103 rejection should be withdrawn. The claimed invention provides a fundamentally novel architecture that integrates monitoring of a neural network through multiple supervisory levels with execution of optimization operations during designated sleep states while maintaining persistent neural network state across operational sessions. This integration is not taught or suggested by Jain's neural network compression system or Rostami's mobile device power management, either alone or in combination. The Examiner has failed to establish a prima facie case of obviousness because the combination does not teach key claim limitations including but not limited to the execution of specific optimization operations during designated sleep states of the neural network, the maintenance of persistent neural network state including both neural activation patterns and architectural configurations across operational sessions, and the integration of hierarchical supervisory monitoring with sleep state optimization in a unified system that manages operational states of the neural network itself. The claimed invention represents a significant advancement over conventional neural network architectures by providing a system that can optimize itself during designated operational states while maintaining operational continuity across sessions, which is neither taught nor suggested by the cited combination of a compression framework and a device power management scheme. Applicant respectfully requests reconsideration and withdrawal of the rejection”.
The arguments above are not persuasive since Jain as primary reference desires a neural network with lower power consumption [0476] and making predictions [0523], while Rostami as secondary reference also teaches a neural network and minimizing power consumption (pg. 6026, right col., last para. to pg. 6027, left col. first para.),
Furthermore, Jain as primary reference also teaches sleep states because Jain teaches of CNNs may identify redundancies in the network during training and statically remove the redundancies to obtain a final network configuration. [0642], while Rostami’s teaching of nodes wake up and sleep periodically (pg. 6021, left col., first para.)... A multi-step Long Short-Term Memory (LSTM) neural network is trained with data from real user applications and tailored for traffic prediction purposes, pg. 6021, right col., first para.), and as a result, then 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 Jain to incorporate the teachings of Rostami for the benefit of prediction purposes (Rostami, pg. 6021, second to the last para.).
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
3. Claims 1-18 are rejected under 35 U.S.C 101 because the claimed invention is directed towards an abstract idea without significantly more.
Step 1
Independent claim 1 is directed to a system, and falls into one of the four statutory categories.
Step 2A, Prong 1
Claim 1 recites the following abstract ideas:
monitor the neural network through multiple hierarchical supervisory levels (Mental process directed to monitoring neural network which can be performed by observing the behavior of the hierarchical supervisory system),
monitoring the neural network includes identifying operation patterns (Mental process directed to identifying operation patterns which can be performed by observing and making a judgement to identify patterns of the operations),
implementing architectural changes (Mental process directed to implementing architectural changes which can be performed by observing the architectural changes and making a judgement on when to modify the changes),
detecting network sparsity (Mental process directed to detecting network sparsity which can be performed by observing the sparsity of the network),
coordinating pruning decisions (Mental process directed to pruning decisions by observing the network and making a judgement on when to prune the network), and
managing resource redistribution (Mental process directed to managing resource distribution which can be performed by observing the network and making a judgement for the distribution of resources);
manage signal transmission pathways providing direct connections between non-adjacent network regions with signal modification and temporal coordination during transmission (Mental process directed to managing signal transmission pathways which can be performed by observing and making a judgement on the transmission of signals);
manage the hierarchical supervisory levels by tracking supervisory behavior patterns, and pruning patterns (Mental process directed to tracking behavior patterns and pruning which can be performed by observing the behavioral and pruning patterns), and
manage operational states of the neural network (Mental process directed to managing the state of operation which can be performed by observing the states of operation of the neural network) and
coordinates decision-making across the hierarchical supervisory levels (Mental process directed to coordination of decision-making which can be performed by observing the different levels of the hierarchical system and making a judgement on the decision);
wherein the optimization operations include at least one of neural memory consolidation, neural insight generation, neural pruning coordination, and neural memory reorganization (Mental process directed to pruning coordination of the network which can be performed by observation of the operations of the network and making a judgement on the coordination of the pruning).
Step 2A, Prong 2
Claim 1 recites the following additional elements:
operate a neural network comprising interconnected nodes arranged in layers
(this limitation is directed to linking the use of a judicial exception to a particular technological environment or field of use. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(h));
collecting activation data (this limitation is directed to insignificant extra-solution activity of data transmission. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(g))
storing successful modification and extracting generalizable principles (this
limitation is directed to insignificant extra-solution activity of data gathering/storage. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(g));
maintain persistent neural network state through a state management system that stores and retrieves neural activation patterns and architectural configurations across operational sessions (this limitation is directed to insignificant extra-solution activity of data gathering/storage. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(g)); and
execute optimization operations during designated sleep states (this limitation is directed to mere instruction to apply to a generic computing component. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(f)),
Step 2B
Claim 1 recites the following additional elements:
operate a neural network comprising interconnected nodes arranged in layers
(this limitation is directed to linking the use of a judicial exception to a particular technological environment or field of use. This does not amount to significantly more than judicial exception. See MPEP 2106.05(h));
collecting activation data (this limitation is directed to insignificant extra-solution activity of data transmission that is well understood, routine and conventional. This does not amount to significantly more than judicial exception. See MPEP 2106.05(d)(II), example i)
storing successful modification and extracts generalizable principles (this
limitation is directed to insignificant extra-solution activity of data gathering/storage that is well understood, routine and conventional. This does not amount to significantly more than judicial exception. See MPEP 2106.05(d)(II), example i);
maintain persistent neural network state through a state management system that stores and retrieves neural activation patterns and architectural configurations across operational sessions (this limitation is directed to insignificant extra-solution activity of data gathering/storage that is well understood, routine and conventional. This does not amount to significantly more than judicial exception. See MPEP 2106.05(d)(II), example i); and
execute optimization operations during designated sleep states (this limitation is directed to mere instruction to apply to a generic computing component. This does not amount to significantly more than judicial exception. See MPEP 2106.05(f)),
4. Dependent claim 2 is directed to a system, and falls into one of the four statutory categories.
Claim 2 recites the following abstract ideas:
wherein the hierarchical supervisory system detects network sparsity using thresholds that adapt based on neural network state (Mental process directed to the detection of network sparsity which can be done by observing the network sparsity and making a judgement on the detection).
Claim 2 do not recite any additional elements.
5. Dependent claim 3 is directed to a system, and falls into one of the four statutory categories.
Claim 3 do not recite any abstract ideas.
Claim 3 recite the following additional elements:
wherein the hierarchical supervisory system exchanges information about resource availability and network sparsity across the multiple supervisory levels (this limitation is directed to insignificant extra-solution activity of data transmission. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(g)).
Claim 3 recite the following additional elements:
wherein the hierarchical supervisory system exchanges information about resource availability and network sparsity across the multiple supervisory levels (this limitation is directed to insignificant extra-solution activity of data transmission and it is well understood routine and conventional. This does not amount to significantly more than judicial exception. See MPEP 2106.05(d)(II), example i).
6. Dependent claim 4 is directed to a system, and falls into one of the four statutory categories.
Claim 4 do recite the following abstract ideas:
wherein the meta-supervisory system maintains network stability while identifying patterns across implemented pruning decisions (Mental process directed to maintaining network stability of the system which can be performed by observation of the system and making a judgement on the stability of the network).
Claim 4 do not recite any additional elements.
7. Dependent claim 5 is directed to a system, and falls into one of the four statutory categories.
Claim 5 recite the following abstract ideas:
wherein the cognitive neural orchestrator comprises at least a state management controller that tracks operational states across the neural architecture and a decision coordination framework that makes real-time decisions about resource allocation and process scheduling (Mental process directed to tracking operational states across the neural architecture and making decisions which can be done by observing the operational states of the neural architecture and making a judgement about resource allocation and process scheduling)
Claim 5 do not recite any additional elements.
8. Dependent claim 6 is directed to a system, and falls into one of the four statutory categories.
Claim 6 recites the following abstract ideas:
that manages restoration of neural network state after system restarts (Mental process directed to managing restoration of neural network state which can be done by observing the neural network state).
Claim 6 recite the following additional elements ideas:
wherein the persistent neural network state is maintained by at least a neural state serialization system that captures and stores the state of the neural architecture (this limitation is directed to insignificant extra-solution activity of data transmission. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(g))and
a neural recovery controller (this limitation is directed to linking the use of a judicial exception to a particular technological environment or field of use. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(h))
Claim 6 recite the following additional elements ideas:
wherein the persistent neural network state is maintained by at least a neural state serialization system that captures and stores the state of the neural architecture (this limitation is directed to insignificant extra-solution activity of data transmission and it is well understood routine and conventional. This does not amount to significantly more than judicial exception. See MPEP 2106.05(d)(II), example i) and
a neural recovery controller (this limitation is directed to linking the use of a judicial exception to a particular technological environment or field of use. This does not amount to significantly more than judicial exception. See MPEP 2106.05(h))
9. Dependent claim 7 is directed to a system, and falls into one of the four statutory categories.
Claim 7 do not recite any abstract ideas.
Claim 7 recite the following additional elements ideas:
further comprising a hierarchical sleep management system that comprises at least a sleep scheduler hierarchy implementing sleep scheduling at multiple levels of the supervisory hierarchy (this limitation is directed to linking the use of a judicial exception to a particular technological environment or field of use. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(h) and
a multi-level wake trigger system establishing wake trigger mechanisms with sensitivity thresholds for different types of stimuli (this limitation is directed to linking the use of a judicial exception to a particular technological environment or field of use. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 7 recite the following additional elements ideas:
further comprising a hierarchical sleep management system that comprises at least a sleep scheduler hierarchy implementing sleep scheduling at multiple levels of the supervisory hierarchy (this limitation is directed to linking the use of a judicial exception to a particular technological environment or field of use. This does not amount to significantly more than judicial exception. See MPEP 2106.05(h) and
a multi-level wake trigger system establishing wake trigger mechanisms with sensitivity thresholds for different types of stimuli (this limitation is directed to linking the use of a judicial exception to a particular technological environment or field of use. This does not amount to significantly more than judicial exception. See MPEP 2106.05(h)).
10. Dependent claim 8 is directed to a system, and falls into one of the four statutory categories.
Claim 8 do not recite any abstract ideas.
Claim 8 recite the following additional elements ideas:
wherein the optimization operations include neural memory consolidation, and wherein the neural memory consolidation comprises at least evaluating neural pathways based on importance factors and strengthening connections identified as important within the neural network (this limitation is directed to linking the use of a judicial exception to a particular technological environment or field of use. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 8 recite the following additional elements ideas:
wherein the optimization operations include neural memory consolidation, and wherein the neural memory consolidation comprises at least evaluating neural pathways based on importance factors and strengthening connections identified as important within the neural network (this limitation is directed to linking the use of a judicial exception to a particular technological environment or field of use. This does not amount to significantly more than judicial exception. See MPEP 2106.05(h)).
11. Dependent claim 9 is directed to a system, and falls into one of the four statutory categories.
Claim 9 recite the following abstract ideas:
wherein the optimization operations include neural insight generation, and wherein the neural insight generation comprises at least discovering non-obvious connections between different network regions and generating potential bundle connections between functionally related regions (Mental process directed to neural insight generation which discovering non-obvious connections, this can be performed by observing the connections of different network regions and making a judgement on the connections that are non-obvious).
Claim 9 do not recite any additional elements.
12. Independent claim 10 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 10, it is substantially similar to claim 1, and is rejected in the same manner and reasoning applying.
13. Dependent claim 11 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 11, it is substantially similar to claim 2, and is rejected in the same manner and reasoning applying.
14. Dependent claim 12 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 12, it is substantially similar to claim 3, and is rejected in the same manner and reasoning applying.
15. Dependent claim 13 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 13, it is substantially similar to claim 4, and is rejected in the same manner and reasoning applying.
16. Dependent claim 14 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 14, it is substantially similar to claim 5, and is rejected in the same manner and reasoning applying.
17. Dependent claim 15 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 15, it is substantially similar to claim 6, and is rejected in the same manner and reasoning applying.
18. Dependent claim 16 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 16, it is substantially similar to claim 7, and is rejected in the same manner and reasoning applying.
19. Dependent claim 17 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 17, it is substantially similar to claim 8, and is rejected in the same manner and reasoning applying.
20. Dependent claim 18 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 18, it is substantially similar to claim 9, and is rejected in the same manner and reasoning applying.
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.
21. Claims 1-18 are rejected under 35 U.S.C. 103 as being unpatentable over Jain et al. (US20240007414 PCT filed 06/25/2021) in view of Rostami et al. ("Wake-up scheduling for energy-efficient mobile devices." IEEE Transactions on Wireless Communications 19.9 (2020): 6020-6036, Date of publication June 9, 2020; date of current version September 10, 2020)
Regarding claim 1, Jain teaches a computer system comprising a hardware memory (The one or more illustrative data storage devices/disks D110 may be embodied as one or more of any type(s) of physical device(s) configured for short-term or long-term storage of data such as, for example, memory devices, memory, circuitry, memory cards, flash memory, hard disk drives, solid-state drives (SSDs), and/or other data storage devices/disks [0114]),
wherein the computer system is configured to execute software instructions stored on nontransitory machine-readable storage media that: operate a neural network comprising interconnected nodes arranged in layers (FIG. ID6_1B … A fully trained NN ID6_126 is illustrated on the left side with a fully connected network of nodes. In the illustrated example of FIG. ID6_1B, nodes are represented as circles and weights are represented as lines between respective circles. The fully trained NN ID6_126 undergoes a fine-grained/layer-wise/per-layer compression policy [0493]);
monitor the neural network (Layer-wise mixed-precision configuration is a technique used to find the optimal configuration for every layer of a trained neural network so inference is accelerated, and accuracy is maintained [0486]) through multiple hierarchical supervisory levels (Example layer-wise mixed-precision sparsity policy predictor circuitry ID6_113 is an example structure that takes the prediction from the agent ID6_110 [0486]; The example compression environment performer circuitry ID6_116 receives a neural network fine-grained compression framework (e.g., realized as a policy and/or a particular configuration) from the layer-wise mixed-precision sparsity policy predictor circuitry ID6_113 [0487]; The Examiner notes ID6_110 → ID6_113 →ID6_116 is the multiple hierarchical supervisory levels),
wherein monitoring the neural network includes collecting activation data (The experience buffer ID6_134 stores data that is independently distributed to break any temporal correlation of behavior and distributes/averages data over many of its previous states to avoid oscillations or divergence in the model [0697]), identifying operation patterns (The agent controller ID6_132 samples the experiences from the experience replay buffer ID6_134 and uses the samples to train the NN [0497]), implementing architectural changes (The agent ID6_131 uses samples of the quantized data ID6_147 from the buffer and uses them to alter its network (e.g., alter, change, improve) [0497]), detecting network sparsity (An example of pruning weights can include setting individual parameters to zero and making the network sparse. This would lower the number of parameters in the model while keeping the architecture the same [0494]), coordinating pruning decisions (Agent ID6_131 predicts pruning/quantization for one, multiple, or all layers at once. The agent ID6_131 may be implemented in a manner consistent with the example agent ID6_110 of FIG. ID6_1A [0497]), and managing resource redistribution (Example layer-wise pruning ID6_128 includes varying a sparsity level (e.g., number of zeros) at different layers [0493]);
manage the hierarchical supervisory levels by tracking supervisory behavior patterns (During each iteration, the layer-wise mixed-precision sparsity policy predictor circuitry ID6_113 explores a potential solution and by the end of the iteration [0486]), storing successful modification and pruning patterns (During each iteration, the layer-wise mixed-precision sparsity policy predictor circuitry ID6_113 explores a potential solution and by the end of the iteration [0486]), and extracting generalizable principles (ID6_116 creates a hardware specific execution graph for hardware performance evaluation. The hardware specific execution graph provides feedback on latency, throughput, and/or power [0487]); manage signal transmission pathways providing direct connections (In some examples, an optimized graph is represented as a neural architecture having a particular number of layers, connections (nodes), weights and/or hyperparameters … Such data may also propagate from a first data structure to a second or any number of subsequent data structures along a data path, such as a bus [0219]) between non-adjacent network regions with signal modification and temporal coordination during transmission (Considering that one or more conditions have changed, the example agent managing circuitry ID5_C310 assigns another agent to re-assess performance of the selected alternate path [0229]; The per-layer compression policy is a way to gain higher performance while preserving the accuracy of a neural network. Compression of neural networks reduces the amount of data transfer and computing during the execution of the neural network [0484]; there is a distinct advantage with Example B ID6_412 where the scale of delivery of compressed models for a given turnaround time/lead shows a temporal improvement for finding optimal compression policies [0506]);
manage operational states of the neural network and coordinate decision-making across the hierarchical supervisory levels (ID6_113...checks for changes in delta (e.g., monitoring for diminishing or increasing returns) to decide when to stop iterations when performance is no longer improving... The example agent ID6_110 is responsible for, in part, predicting/inferencing a layer-wise mixed-precision configuration that is consumed by an example compression environment performer circuitry ID6_116. [0486]);
maintain persistent neural network state through a state management system that stores (To provide for persistent storage of information such as data, applications, operating systems and so forth, a storage F458 may also couple to the processor F452 via the interconnect F456 [0188]) (To provide for persistent storage of information such as data, applications, operating systems and so forth, a storage F458 may also couple to the processor F452 via the interconnect F456 [0188]) and retrieves neural activation patterns (In operation, an example input receiver circuitry ID6_102 is an example structure that receives, retrieves and/or otherwise obtains inputs [0480]) and architectural configurations across operational sessions (The parse model to identify operations (block ID7_308) is an example of a process that automatically receives its input when it retrieves a model (block ID7_302) in addition to receiving all the operation configurations [0598]); and
wherein the optimization operations include at least one of neural memory consolidation, neural insight generation, neural pruning coordination, and neural memory reorganization (For example, by pruning particular neurons the corresponding weights do not need to be stored, thus enabling storage savings for the model [0476]).
Jain does not explicitly teach execute optimization operations during designated sleep states.
Rostami teaches execute optimization operations during designated sleep states (Based on these assumptions, we focus on optimizing the buffer size threshold (γ) in order to minimize the UE’s power consumption while satisfying a specific delay requirement (i.e., average buffering delay should be less than or equal to a maximum tolerable delay, Dmax), under Poisson traffic model assumption, for given values of tw, ti, ton, tpd and tsu (pg. 6026, right col., last para. to pg. 6027, left col. first para.); In this regard, the proposed wake-up scheduler increases the sleep period of the UE as much as possible in a greedy manner by not sending WI = 1 until the average buffering delay approaches Dmax (pg. 6027, right 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 Jain to incorporate the teachings of Rostami for the benefit of neural network trained with data from real user applications and tailored for traffic prediction purposes (Rostami, pg. 6021, second to the last para.)
Regarding claim 2, Modified Jain teaches the computer system of claim 1, Jain teaches wherein the hierarchical supervisory system detects network sparsity using thresholds that adapt based on neural network state (n example of pruning weights can include setting individual parameters to zero and making the network sparse [0494]; Additionally, a threshold layer ID11_B112, implemented by the example second branch implementation controller ID11_B101, may set (e.g., truncate) all negative values in the output of the adaptive bias component ID11_B110 to zero and set all positive values in the output of the adaptive bias component ID11_B110 to one [0649]).
Regarding claim 3, Modified Jain teaches the computer system of claim 1, Jain teaches wherein the hierarchical supervisory system exchanges information about resource availability (Examples disclosed herein use a model-based approach to dynamically adapt resource availability across multiple AI models in a computing environment [0276]) and network sparsity across the multiple supervisory levels (The example compression environment performer circuitry ID6_116 receives a neural network fine-grained compression framework (e.g., realized as a policy and/or a particular configuration) from the layer-wise mixed-precision sparsity policy predictor circuitry ID6_113 and from the agent ID6_110 [0487]).
Regarding claim 4, Modified Jain teaches the computer system of claim 1, Jain teaches wherein the meta-supervisory system maintains network stability while identifying patterns across implemented pruning decisions (quantifying the need for more or a smaller number of neurons in a fully connected layer in order to classify optimally the patterns in the training database [0740]; An example of pruning nodes can include removing entire nodes from the network. This would make the NN architecture itself smaller, while aiming to keep the accuracy of the initial larger network [0494]).
Regarding claim 5, Modified Jain teaches the computer system of claim 1, Jain teaches wherein the cognitive neural orchestrator comprises at least a state management controller that tracks operational states across the neural architecture (An example accuracy/network states checker circuitry ID6_114 is an example structure that checks for the accuracy and state of the neural network. The accuracy/network states checker circuitry ID6_114 compares a value (e.g., input by user, a threshold value and/or a predetermined value) to the accuracy and state of the neural network to determine if it has reached a predefined threshold (e.g., determined by user input and/or a predetermined value) [0488]) and
a decision coordination framework that makes real-time decisions about resource allocation and process scheduling (benchmark managing circuitry ID5_C302 attaches the union of graphs to the workload (block ID5_D114) so that dynamic decisions may occur in real time during an inference/runtime phase of the workload [0241]; Example graph configurations may include different layer structure arrangements (e.g., in the event a convolutional neural network (CNN) is used) [0222]).
Regarding claim 6, Modified teaches the computer system of claim 1, Jain teaches wherein the persistent neural network state is maintained by at least a neural state serialization system that captures and stores the state of the neural architecture (The illustrated example of FIG. ID6_4 includes a first example A ID6_402 to illustrate the total time it would take to train three agents using a regular neural network method … In other words, it can take the same amount of time or more time to train one or more models due to the serial nature of the learning agent training process [0504]) and
a neural recovery controller that manages restoration of neural network state after system restarts (A backup of the original weights is generated as W′ and replaces W with W′ during a restoration operation [0837]).
Regarding claim 7, Modified Jain teaches the computer system of claim 1, Jain does not explicitly teach further comprising a hierarchical sleep management system that comprises at least a sleep scheduler hierarchy implementing sleep scheduling at multiple levels of the supervisory hierarchy and a multi-level wake trigger system establishing wake trigger mechanisms with sensitivity thresholds for different types of stimuli.
Rostami teaches further comprising a hierarchical sleep management system (Fig. 5. Overall block diagram of the WuSched-Online, pg. 6028, left col.,) that comprises at least a sleep scheduler hierarchy implementing sleep scheduling at multiple levels of the supervisory hierarchy (In duty cycling, … nodes wake up and sleep periodically (pg. 6021, left col., first para.); an online optimization is proposed through the WuSched-Online. It uses a proactive scheduler that takes decisions every wake-up cycle based on traffic predictions over a forecast horizon. A multi-step Long Short-Term Memory (LSTM) neural network is trained with data from real user applications and tailored for traffic prediction purposes, pg. 6021, right col., first para.) and
a multi-level wake trigger system establishing wake trigger mechanisms with sensitivity thresholds for different types of stimuli (As it can be observed in video results of the WuSched-Online, a large number of packets are served with near to zero delay, and the reason is due to the consecutive packet arrivals that are served while the inactivity timer is triggered, pg. 6031, right col., last para.); From the system-level point of view, the tunable parameter of the WuSched-Offline is the buffer size threshold (γ ≥ 1),… parameters of the wakeup scheduler (ton, tpd, tsu) depend on physical constraints and signal design, pg. 6026, right col., last para.; The Examiner notes threshold (γ ≥ 1) are the sensitivity thresholds for different parameters)
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 Jain to incorporate the teachings of Rostami for the benefit of introducing a novel concept called wake-up scheduling (WuSched) to further improve the energy efficiency of mobile devices (Rostami, pg. 6021, right col., first para.)
Regarding claim 8, Modified Jain teaches the computer system of claim 1, Jain teaches wherein the optimization operations include neural memory consolidation (Additional or consolidated instances of the Edge aggregation nodes A340 and the aggregation points A342, A344, including those deployed on a single server framework, may also be present within the Edge cloud A110 [0103]; The Edge cloud A110 thus may be embodied as any type of network that provides Edge computing and/or storage resources [0101]), and
wherein the neural memory consolidation comprises at least evaluating neural pathways based on importance factors and strengthening connections identified as important within the neural network (In a differential approach, the controller ID3_420 may contain either a supergraph and/or a supernetwork to create a path in the neural network architecture [0291]; Additionally, execution of the instructions ID11_F120 may cause the system ID11_F100 to conduct an importance classification of the aggregated context information and selectively exclude one or more channels in the first network layer from consideration by the second network layer based on the importance classification [0663]).
Regarding claim 9, Modified Jain teaches the computer system of claim 1, Jain teaches wherein the optimization operations include neural insight generation (Based on telemetry and insights from mile markers, neural architecture search strategies can be improved based on past learning, incorrect predictions, etc., as evaluated using the telemetry data, mile marker insights, etc [0440]), and
wherein the neural insight generation comprises at least discovering non-obvious connections between different network regions (neural network layers are quantized in a non-uniform way as some layers are more sensitive to distortion [0495]) and
generating potential bundle connections between functionally related regions (Examples disclosed herein discover and accelerate inference of dictionary-based weighting with non-uniform quantized NNs [0824]).
Regarding claim 10, claim 10 is similar to claim 1. It is rejected in the same manner and reasoning applying.
Regarding claim 11, claim 11 is similar to claim 2. It is rejected in the same manner and reasoning applying.
Regarding claim 12, claim 12 is similar to claim 3. It is rejected in the same manner and reasoning applying.
Regarding claim 13, claim 13 is similar to claim 4. It is rejected in the same manner and reasoning applying.
Regarding claim 14, claim 14 is similar to claim 5. It is rejected in the same manner and reasoning applying.
Regarding claim 15, claim 15 is similar to claim 6. It is rejected in the same manner and reasoning applying.
Regarding claim 16, claim 16 is similar to claim 7. It is rejected in the same manner and reasoning applying.
Regarding claim 17, claim 17 is similar to claim 8. It is rejected in the same manner and reasoning applying.
Regarding claim 18, claim 18 is similar to claim 9. It is rejected in the same manner and reasoning applying.
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 extension fee 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 date of this final action.
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 8am-5pm 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