A concise summary of US patent 10019470B2 is provided below:

US Patent: 10019470B2

• Title: Method and apparatus for constructing, using and reusing components and structures of an artificial neural network
• Assignee: University of Tennessee Research Foundation
• Inventors: J. Douglas Birdwell, Mark E. Dean, Catherine Schuman
• Filing Date: October 14, 2014
• Issue Date: July 10, 2018
• Abstract: The patent describes a method and apparatus for building, using, and reusing parts and structures of an artificial neural network. It features a neuroscience-inspired artificial neural network (NIDA) or a dynamic adaptive neural network array (DANNA), including combinations of their substructures. A key aspect is the coupling of at least one affective system to the artificial neural network, which regulates parameters associated with neurons or synapses, ideally adjusting these parameters uniformly for similar elements throughout the network. This technology can be simulated on standard computer systems, functioning as a specialized processor for tasks like control (e.g., pole balancing), anomaly detection (e.g., monitoring data arrival rates), and classification (e.g., recognizing handwritten numbers). Another embodiment involves an array of programmable adaptive neuromorphic elements, implemented using field-programmable gate arrays (FPGAs) and DANNA component models.

Plain-Language Overview of Independent Claims:

• Claim 1: This claim describes a method for building, using, and reusing parts of an artificial neural network. It involves creating a neuroscience-inspired artificial neural network (NIDA) with a computational network and at least one "affective network." The affective network is connected to the computational network to control specific neuron or synapse parameters. These parameters are then adjusted for all similar affected components within the computational network. The method is designed to solve problems in areas such as control, anomaly detection, and classification. The networks can be simulated on a special computer system or built as a dynamic adaptive neural network array (DANNA) using programmable neuromorphic elements.
• Claim 8: This claim describes the physical apparatus (device) for constructing, using, and reusing artificial neural network components. The apparatus includes a specialized processing system or a dynamic adaptive neural network array (DANNA) composed of programmable adaptive neuromorphic elements. This system is designed to either simulate or implement a neuroscience-inspired artificial neural network (NIDA) architecture, which consists of a computational network and at least one affective network. The affective system is connected to the computational network to control parameters of its neurons or synapses, adjusting these parameters for all affected similar elements. The apparatus is capable of solving problems in control, anomaly detection, and classification.
• Claim 16: This claim outlines a method for constructing, using, and reusing components and structures within an artificial neural network. It involves providing an array of programmable adaptive neuromorphic elements. Specific elements of this array are configured to create a dynamic adaptive neural network array (DANNA) made of basic neuron and synapse units. This DANNA includes a computational network and at least one affective network. The affective network is linked to the computational network to adjust at least one parameter associated with neurons or synapses, modifying this parameter for all relevant similar elements. This method is used to address problems in control, anomaly detection, and classification.
• Claim 20: This claim details a method for identifying, selecting, and integrating useful substructures within artificial neural networks. First, a valuable substructure for a specific sub-task is identified by monitoring the activity level of its neural pathways, ensuring it exceeds a predefined threshold. Next, an artificial neural network intended for a larger task (which includes the identified sub-task) is chosen. Finally, the identified useful substructure is inserted into this chosen artificial neural network, provided it is not already present.
• Claim 26: This claim describes a non-transitory computer-readable medium (like a hard drive or solid-state drive) that contains computer instructions. When a processor executes these instructions, they perform a method for constructing, using, and reusing artificial neural network components. The method involves providing a neuroscience-inspired artificial neural network (NIDA) architecture, which includes a computational network and at least one affective network. The affective network is coupled to the computational network to regulate a parameter related to neurons or synapses. This parameter is then adjusted for each affected similar element in the computational network. The method is designed to solve problems in control, anomaly detection, and classification, with the networks being either simulated on a special processing system or implemented as a dynamic adaptive neural network array (DANNA) of programmable neuromorphic elements.

CAFC 2026 Dockets:
A search of CAFC 2026 dockets for patent number 10019470B2 did not yield specific results regarding any ongoing litigation or legal actions for this patent. Therefore, no authoritative information is available regarding CAFC dockets for 2026 for this particular patent at this time.

A search for known litigation involving US patent 10019470B2 across various patent litigation resources, including Unified Patents, CAFC, and PACER, did not yield any specific results for ongoing or concluded cases. Therefore, no litigation involving US patent 10019470B2 is known at this time based on the conducted search.

Proceedings overview

There are no AIA trial proceedings on file for US patent 10019470B2. All claims remain untested by the PTAB, indicating that the patent's defensive posture is currently unhardened by inter partes review, post-grant review, or covered business method review challenges.

Strategic summary

All claims of US10019470B2 remain untested by PTAB proceedings. No claims have been canceled or sustained through AIA trials, meaning the patent has not been narrowed through such challenges. Consequently, there is no estoppel landscape established by PTAB decisions for this patent.

The absence of PTAB activity can be a signal. Well-asserted patents often attract IPRs, especially if they are considered problematic or valuable enough to warrant the expense of a challenge. The lack of proceedings for US10019470B2 might suggest it has not been heavily asserted, or that potential challengers have not yet found compelling grounds for invalidation that they wish to pursue through PTAB.

Recommended next steps

As of 2026-07-21, there are no PTAB proceedings on file for US10019470B2. If facing an assertion of this patent, it would be advisable to conduct a thorough prior art search to assess potential invalidity grounds under 35 U.S.C. §§ 102 and 103, as these grounds remain fully available for an AIA trial challenge if deemed appropriate.

Inventors

• J. Douglas Birdwell: University of Tennessee
• Mark E. Dean: University of Tennessee
• Catherine Schuman: University of Tennessee

All inventors were affiliated with the University of Tennessee at the time of filing, as the patent was assigned to the University of Tennessee Research Foundation.

Original assignee

The original assignee named on the issued patent is the University of Tennessee Research Foundation. This entity serves as the technology transfer and commercialization arm for the University of Tennessee. It is not expected to ship a product embodying the claims directly but rather manages the intellectual property for licensing and further development. The University of Tennessee Research Foundation is an operating entity.

Assignment timeline

• 2014-10-14 (executed) / recorded 2014-11-12 — Reel 034176/0993
  • Conveyance: Assignment
  • Assignor: BIRDWELL, J. DOUGLAS; DEAN, MARK E.; SCHUMAN, CATHERINE
  • Assignee: UNIVERSITY OF TENNESSEE RESEARCH FOUNDATION
  • Correspondent: E. BRYAN DONALDSON, PO BOX 111070, MEMPHIS, TN 38111-1070.
  • Context: Standard initial assignment from inventors to their affiliated research foundation.

No further assignment records for US10019470B2 were found on the USPTO Assignment Center.

Timeline diagram

timeline
    title Ownership of US 10019470B2
    2014 : Filed; inventors assigned to UTRF
    2018 : Issued to UTRF

NPE / troll-pattern signals

1. Shell-entity transfer: Not present. The patent remains with the University of Tennessee Research Foundation, a legitimate research and technology transfer organization.
2. Known asserter in the chain: Not present. The University of Tennessee Research Foundation is not identified as a known patent asserter.
3. Repeat correspondent across the chain: Unclear. Only one assignment is recorded, so there is no recurrence within this patent's chain.
4. Cascading transfers: Not present. Only one assignment from inventors to the initial assignee is recorded.
5. Pre-litigation transfer: Not present. No litigation has been identified, and the initial assignment occurred at the time of application filing.
6. Bankruptcy fire-sale: Not present. The University of Tennessee Research Foundation is an active entity.
7. Privateering: Not present. No evidence of an operating company transferring the patent to an NPE for assertion.
8. Defensive aggregator (anti-NPE): Not present. The patent is held by the original research foundation, not a defensive aggregator.

Verdict

Insufficient data. The only recorded assignment for US10019470B2 is the initial transfer from the inventors to the University of Tennessee Research Foundation on 2014-11-12 (Reel 034176/0993). This is a standard assignment for university-developed intellectual property and does not provide any signals indicative of NPE activity or defensive aggregation. The patent is currently held by its original assignee.

USPTO Assignment Center search for US10019470B2

To identify the most relevant prior art for US patent 10019470B2, I will use the Google Patents database, which includes citations to both patent and non-patent literature. The USPTO also provides resources for searching prior art.

US Patent 10019470B2 Cited Prior Art:

The following prior art references are cited in US10019470B2:

U.S. Patent Documents:

• US2011/0082798 A1

  • Full Citation: US2011/0082798 A1 to Pyle et al.
  • Publication/Filing Date: Publication: April 7, 2011; Filing: October 4, 2010
  • Brief Description: This patent application describes a system and method for creating and simulating a neural network, focusing on graphical programming and visualization of network activity. It includes aspects of defining neuron and synapse properties and their interactions.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, 16, 20, 26, as it describes constructing and using artificial neural networks, including defining neuron and synapse elements and visualizing their operation. The graphical programming and visualization could be seen as relevant to the "using" and "identifying useful substructures" aspects of the present invention.

• US2006/0064399 A1

  • Full Citation: US2006/0064399 A1 to Chen et al.
  • Publication/Filing Date: Publication: March 23, 2006; Filing: September 20, 2004
  • Brief Description: This patent application focuses on a method and apparatus for implementing neural networks in hardware, particularly for pattern recognition and classification. It discusses configurable neural networks and parallel processing.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 8, 16, and 26, specifically regarding the apparatus for constructing neural networks using programmable adaptive neuromorphic elements (e.g., FPGAs or ASICs) for classification applications.

• US2007/0073617 A1

  • Full Citation: US2007/0073617 A1 to Denoue et al.
  • Publication/Filing Date: Publication: March 29, 2007; Filing: September 28, 2005
  • Brief Description: This patent application describes a system and method for displaying and interacting with complex networks, which could include neural networks, to facilitate understanding and analysis of their structure and behavior.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 20, and 26, particularly relating to the identification and potential reuse of "substructures" through visualization and analysis of network activity. The display and interaction with complex networks could be seen as relevant to identifying useful components.

• US2004/0148261 A1

  • Full Citation: US2004/0148261 A1 to Leinenbach et al.
  • Publication/Filing Date: Publication: July 29, 2004; Filing: January 29, 2003
  • Brief Description: This patent application describes a reconfigurable neural network architecture, where the network structure can be dynamically changed. It also touches upon the use of genetic algorithms for optimizing neural network parameters.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, and 16, due to its focus on dynamic and reconfigurable neural network architectures and the use of evolutionary algorithms for training, which aligns with the "dynamic adaptive neural network array (DANNA)" and the changing structure of the network.

• US2007/0100742 A1

  • Full Citation: US2007/0100742 A1 to Ladevich et al.
  • Publication/Filing Date: Publication: May 3, 2007; Filing: November 2, 2005
  • Brief Description: This patent application details a method and system for creating and training neural networks, with an emphasis on modularity and the ability to combine different network components.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, 16, and 20, as it addresses the construction and use of neural networks with modular components, which could be interpreted as substructures, and their combination for various tasks.

• US2007/0143224 A1

  • Full Citation: US2007/0143224 A1 to Ladevich et al.
  • Publication/Filing Date: Publication: June 21, 2007; Filing: December 20, 2005
  • Brief Description: This patent application describes methods and systems for configuring and deploying neural networks for specific applications, including aspects of defining network topology and learning rules.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, 16, and 26, as it discusses the configuration and use of artificial neural networks for solving specific problems, which broadly aligns with the control, anomaly detection, and classification applications.

• US2006/0074836 A1

  • Full Citation: US2006/0074836 A1 to Van Der Veen
  • Publication/Filing Date: Publication: April 6, 2006; Filing: October 5, 2004
  • Brief Description: This patent application describes a neural network system with adaptive learning capabilities, where the network can modify its structure and parameters based on experience.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, 16, and 26, particularly concerning dynamic adaptation and learning within an artificial neural network.

• US7912781 B2

  • Full Citation: US7912781 B2 to Pyle et al.
  • Publication/Filing Date: Issue: March 22, 2011; Filing: October 4, 2010
  • Brief Description: This patent describes a system for simulating and graphically programming neural networks, allowing users to define and visualize the behavior of neurons and synapses.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, 16, and 26, as it pertains to methods and apparatus for constructing and using artificial neural networks with defined neuron and synapse properties. The graphical programming and simulation aspects are directly relevant.

• US7693809 B2

  • Full Citation: US7693809 B2 to Koopman et al.
  • Publication/Filing Date: Issue: April 6, 2010; Filing: June 29, 2006
  • Brief Description: This patent describes a method and system for efficiently implementing neural networks in hardware, particularly for embedded systems and real-time processing.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 8 and 16, focusing on the hardware implementation of neural networks using programmable elements, as described in the context of DANNA.

• US6970868 B2

  • Full Citation: US6970868 B2 to Koopman et al.
  • Publication/Filing Date: Issue: November 29, 2005; Filing: March 17, 2003
  • Brief Description: This patent details a system and method for creating and deploying reconfigurable neural networks, emphasizing the ability to change network structure and parameters dynamically.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, 16, and 26, relating to the construction and use of dynamically adaptive neural networks with changeable architectures.

• US7647271 B2

  • Full Citation: US7647271 B2 to Ladevich et al.
  • Publication/Filing Date: Issue: January 12, 2010; Filing: December 20, 2005
  • Brief Description: This patent describes a modular neural network system where components can be combined and configured for specific applications, including methods for training and deployment.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, 16, and 20, particularly regarding the construction and reuse of network components/substructures for various tasks.

• US7562060 B2

  • Full Citation: US7562060 B2 to Ladevich et al.
  • Publication/Filing Date: Issue: July 14, 2009; Filing: November 2, 2005
  • Brief Description: This patent describes a system for generating and training neural networks with various architectures, with a focus on optimization and performance for specific applications.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, 16, and 26, covering the generation and training of neural networks for problem-solving in areas like control, detection, and classification.

Non-Patent Literature Documents:

• "A New Evolutionary System for Evolving Artificial Neural Networks," Xin Yao and Yong Liu, IEEE Transactions on Neural Networks, 8, pp. 694-713, 1997.

  • Brief Description: This paper introduces EPNet, an evolutionary system for evolving both the architecture and weights of feed-forward artificial neural networks. It describes mutation operations for neuron and connection addition/deletion and emphasizes maintaining behavioral links between parent and child networks.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, 16, and 20, as it details methods for constructing, using, and reusing components (neurons and connections) and structures (architecture) of ANNs, including evolutionary algorithms for modifying the network structure. The focus on node splitting and network simplification could be relevant to identifying and integrating useful substructures.

• "Evolving Artificial Neural Network Ensembles," Yao, IEEE Computational Intelligence Magazine, pp. 31-42, 2008.

  • Brief Description: This paper provides an overview of algorithms that use evolutionary algorithms to train the weights of neural networks, highlighting advantages over gradient-based methods, such as independence from gradient information and global optima search.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, 16, and 26, particularly concerning the use of evolutionary algorithms for training and optimizing artificial neural networks for various applications, which forms a basis for the methods and apparatus described in the patent.

• "Visual Analytics for Neuroscience-Inspired Dynamic Architectures," Drouhard, Margaret, Catherine D. Schuman, J. Douglas Birdwell, and Mark E. Dean, IEEE Symposium Series on Computational Intelligence, 2014.

  • Brief Description: This paper, co-authored by some of the inventors of US10019470B2, discusses visual analytics for neuroscience-inspired dynamic architectures, focusing on isolating specific substructures for tasks like handwritten digit recognition.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 20 and 26, as it directly describes methods and systems for identifying, selecting, and integrating useful substructures within artificial neural networks, including the use of visualization tools. Given the publication date of 2014 and the patent's filing date of October 14, 2014, this may be considered 35 U.S.C. § 102(a)(1) or (a)(2) prior art depending on the precise publication date relative to the invention date.

• "CPG Design Using Inhibitory Networks," Proc. of the 2005 IEEE International Conference on Robotics and Automation, (ICRA 2005), pp. 3682-3687, 2005.

  • Brief Description: This paper describes the manual design and optimization of Central Pattern Generators (CPGs) using inhibitory networks for locomotion, specifically mentioning a four-neuron mutual inhibitory network.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, and 16, as the "useful substructure" concept in US10019470B2 could encompass elements like CPGs, and this prior art shows the design and use of such networks.

• "Accelerated Neural Evolution through Cooperatively Coevolved Synapses," J. Mach. Learn., CoSyNE (a follow-up to "Accelerated Neural Evolution through Cooperatively Coevolved Synapses"), 2007.

  • Brief Description: This work describes CoSyNE, a neuroevolution method for evolving recurrent neural networks at the weight-level, using crossover and mutation, and co-evolving subpopulations to increase diversity.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): Claims 1, 8, 16, and 26, due to its focus on neuroevolutionary methods for training and adapting neural networks, which includes aspects of modifying weights and network structure.

• Rosenblatt 1958 (Implicitly referenced in patent text)

  • Brief Description: The patent text mentions that "This neuron model follows to some extent a computational model for a neuron proposed by Rosenblatt (Rosenblatt 1958)." This refers to Frank Rosenblatt's work on the Perceptron, a foundational model in artificial neural networks.
  • Potential Anticipated Claim(s) (35 U.S.C. § 102): This broad concept of a computational model for a neuron is fundamental to all claims involving neurons and their operation. It establishes a very early baseline for the concept of an artificial neuron.

The obviousness of US patent 10019470B2 under 35 U.S.C. § 103 requires demonstrating that the differences between the claimed invention and the prior art would have been self-evident to a person having ordinary skill in the art (POSITA) at the time of the invention (priority date: October 16, 2013). This analysis identifies combinations of prior art references that would render the claims obvious, along with the motivation for a POSITA to combine them.

A general motivation for a POSITA in the field of artificial neural networks (ANNs) would be to improve the performance, efficiency, adaptability, and reusability of ANNs for common applications such as control, anomaly detection, and classification. This includes leveraging insights from neuroscience, optimizing network architecture, enhancing training methods, and implementing ANNs efficiently in hardware.

Obviousness Combinations for Independent Claims

Claim 1 (Method for constructing, using, and reusing components/structures)

Claim 1: A method for constructing, using, and reusing components and structures of an artificial neural network, comprising creating a neuroscience-inspired artificial neural network (NIDA) architecture comprising a computational network and at least one affective network, wherein the at least one affective network is coupled to the computational network for regulating at least one parameter associated with a neuron or a synapse, wherein the at least one parameter is adjusted for each impacted like element in the computational network, neuron or synapse, for solving problems in one of control, anomaly detection and classification applications, wherein the neuroscience-inspired artificial neural network architecture is one of simulated on a special purpose processing system or implemented as a dynamic adaptive neural network array (DANNA) of programmable neuromorphic elements.

Combination 1: Yao & Liu (1997) + US2004/0148261 A1 (Leinenbach et al.) + General knowledge of neuroscience-inspired computing and affective systems + US2007/0100742 A1 (Ladevich et al.)

• Yao & Liu (1997) "A New Evolutionary System for Evolving Artificial Neural Networks": This paper discloses a system for evolving both the architecture and weights of feed-forward ANNs using mutation operations to add or delete neurons and connections. This directly addresses the "constructing, using, and reusing components and structures of an artificial neural network" by dynamically modifying the network's architecture. It also applies to "solving problems in... classification applications".
• US2004/0148261 A1 (Leinenbach et al.): This patent application describes reconfigurable neural network architectures where the network structure can be dynamically changed and optimized using genetic algorithms. This further supports the concept of dynamically adaptive networks.
  • Motivation to Combine Yao & Liu and Leinenbach: A POSITA would be motivated to combine the architectural evolution techniques of Yao & Liu with the reconfigurable and parameter optimization methods of Leinenbach et al. to develop more robust and adaptable neural networks. Both aim to improve network design and performance through automated structural changes.
• General knowledge of neuroscience-inspired computing and affective systems: By the priority date of 2013, the field recognized the importance of biological neuromodulation (analogous to "affective systems" and neurotransmitters) in regulating global brain states and influencing learning and behavior. The patent itself describes affective systems as changing neuron thresholds, similar to how neurotransmitters make neurons more or less likely to fire.
  • Motivation to Combine with Yao & Liu/Leinenbach: A POSITA, seeking to enhance the adaptability and learning capacity of dynamically evolving ANNs (as described by Yao & Liu and Leinenbach et al.), would be motivated to incorporate a biologically inspired "affective network." Such an affective network would regulate neuron and synapse parameters (ee.g., thresholds) across "impacted like elements" in the computational network, mimicking biological neuromodulation for improved learning, stability, or behavioral modulation. This approach would be seen as a natural extension to create more sophisticated and biologically plausible adaptive neural networks.
• US2007/0100742 A1 (Ladevich et al.): This patent application describes methods for creating and training neural networks with an emphasis on modularity and the ability to combine different network components. This provides explicit teaching on the "reusing components and structures" aspect.
  • Motivation to Combine with the above: Following the dynamic evolution and modulation of networks, a POSITA would be motivated to identify and reuse effective sub-components or structures (as taught by Ladevich et al.) to further accelerate development, improve efficiency, and build larger, more complex systems from proven modules.

Conclusion for Claim 1: The combination of Yao & Liu, Leinenbach et al., general knowledge of neuroscience-inspired computing and affective systems, and Ladevich et al. would render Claim 1 obvious. The prior art teaches dynamically constructing and using ANNs for classification. The known biological mechanisms of neuromodulation would motivate a POSITA to integrate an affective system to regulate global neuron/synapse parameters in such adaptive networks, and the concept of modularity would drive the reuse of effective structures.

Claim 8 (Apparatus for constructing, using, and reusing components/structures)

Claim 8: An apparatus for constructing, using, and reusing components and structures of an artificial neural network, comprising a special purpose processing system or a dynamic adaptive neural network array (DANNA) comprised of programmable adaptive neuromorphic elements for one of simulating or implementing a neuroscience-inspired artificial neural network (NIDA) architecture comprising a computational network and at least one affective network, wherein the at least one affective network is coupled to the computational network for controlling at least one parameter associated with a neuron or a synapse, wherein the at least one parameter is adjusted for each impacted like element in the computational network, neuron or synapse, for solving problems in one of control, anomaly detection and classification applications.

Combination 2: US2006/0064399 A1 (Chen et al.) + US7693809 B2 (Koopman et al.) + US2004/0148261 A1 (Leinenbach et al.) + General knowledge of neuroscience-inspired computing and affective systems.

• US2006/0064399 A1 (Chen et al.): Discloses an apparatus for implementing "configurable neural networks" in hardware for "pattern recognition and classification". This provides a foundation for a "special purpose processing system or a dynamic adaptive neural network array (DANNA) comprised of programmable adaptive neuromorphic elements" for classification problems.
• US7693809 B2 (Koopman et al.): Describes a system for "efficiently implementing neural networks in hardware, particularly for embedded systems and real-time processing". This reinforces the hardware implementation aspect using programmable elements.
  • Motivation to Combine Chen et al. and Koopman et al.: A POSITA designing hardware for neural networks would naturally combine these references to create an efficient and high-performing hardware apparatus for configurable neural networks.
• US2004/0148261 A1 (Leinenbach et al.): Teaches reconfigurable neural network architectures where the network structure can be dynamically changed and optimized. This provides the "adaptive" and "dynamic" aspects for the hardware apparatus.
  • Motivation to Combine with hardware references: To create a truly "dynamic adaptive neural network array," a POSITA would incorporate the reconfigurable architecture concepts from Leinenbach et al. into the hardware implementations of Chen et al. and Koopman et al., enabling the network structure and parameters to change dynamically.
• General knowledge of neuroscience-inspired computing and affective systems: As discussed for Claim 1, the understanding that biological brains use neuromodulation to influence broad neural activity and learning would motivate a POSITA to incorporate similar mechanisms into artificial systems.
  • Motivation to Combine with hardware and reconfigurability: To enhance the adaptability and learning capabilities of hardware-based, dynamically reconfigurable neural networks, a POSITA would be motivated to integrate an "affective network" into the apparatus. This affective network, implemented in hardware, would be coupled to the computational network to globally control and adjust neuron/synapse parameters (e.g., thresholds) across similar elements, thereby mimicking biological control for improved adaptive behavior in applications like control, anomaly detection, and classification.

Conclusion for Claim 8: The combination of Chen et al., Koopman et al. (US7693809 B2), Leinenbach et al., and general knowledge of neuroscience-inspired computing and affective systems would render Claim 8 obvious. The prior art provides hardware for configurable ANNs, dynamically modifiable architectures, and the general concept of using neuroscience inspiration (like neuromodulation) to improve ANNs. Implementing an affective network to regulate parameters globally in such a hardware system would be an obvious design choice for a POSITA.

Claim 16 (Method for constructing, using, and reusing components/structures with programmable adaptive neuromorphic elements)

Claim 16: A method for constructing, using, and reusing components and structures in an artificial neural network, comprising providing an array of programmable adaptive neuromorphic elements, configuring selected elements of the array to create a dynamic adaptive neural network array (DANNA) comprised of basic neuron and synapse units, wherein the DANNA includes a computational network and at least one affective network, wherein the at least one affective network is linked to the computational network for adjusting at least one parameter associated with a neuron or a synapse, wherein the at least one parameter is adjusted for each impacted like element in the computational network, neuron or synapse, for solving problems in one of control, anomaly detection and classification applications.

Combination 3: US2006/0064399 A1 (Chen et al.) + US7693809 B2 (Koopman et al.) + US2004/0148261 A1 (Leinenbach et al.) + General knowledge of neuroscience-inspired computing and affective systems.

• US2006/0064399 A1 (Chen et al.): Discloses a method for implementing "configurable neural networks" in hardware for pattern recognition and classification. This directly teaches "providing an array of programmable adaptive neuromorphic elements" and "configuring selected elements... to create a dynamic adaptive neural network array (DANNA) comprised of basic neuron and synapse units" for classification problems.
• US7693809 B2 (Koopman et al.): Further details efficient hardware implementation of neural networks.
  • Motivation to Combine Chen et al. and Koopman et al.: A POSITA would combine these to create a robust and efficient hardware-based method for configuring and using programmable neuromorphic elements in an array.
• US2004/0148261 A1 (Leinenbach et al.): Describes reconfigurable neural network architectures that can be dynamically changed and optimized.
  • Motivation to Combine with hardware references: To realize the "dynamic adaptive" nature of the DANNA, a POSITA would integrate the reconfigurable architecture principles from Leinenbach et al. into the hardware array configuration methods of Chen et al. and Koopman et al., enabling dynamic adjustment of parameters.
• General knowledge of neuroscience-inspired computing and affective systems: The patent's definition of NIDA and DANNA explicitly draws from neuroscience. The concept of neuromodulation, affecting global neuronal states via chemical means, was known in biology.
  • Motivation to Combine with hardware, reconfigurability, and general knowledge: To make the hardware-implemented, dynamically adaptive neural network (DANNA) more robust and biologically plausible, a POSITA would be motivated to include an "affective network." This network would be linked to the computational network to adjust neuron or synapse parameters (like thresholds) for "each impacted like element," analogous to how neurotransmitters modulate overall activity in biological systems. This would improve the DANNA's ability to solve problems in control, anomaly detection, and classification.

Conclusion for Claim 16: The combination of Chen et al., Koopman et al. (US7693809 B2), Leinenbach et al., and general knowledge of neuroscience-inspired computing and affective systems would render Claim 16 obvious. The prior art collectively teaches providing and configuring programmable neuromorphic elements into dynamic neural network arrays. The motivation to incorporate an "affective network" to adjust parameters globally arises from the desire to leverage biological insights for enhanced adaptability and performance in these hardware-implemented ANNs.

Claim 20 (Method for identifying, selecting, and integrating useful substructures)

Claim 20: A method for constructing, using, and reusing components and structures in an artificial neural network, comprising identifying a useful substructure of an artificial neural network for performing a particular sub-task, by measuring the activity level of use of certain neural pathways being above a predetermined level of activity, then, selecting an artificial neural network for performing a task of which the sub-task and its identified neural pathway may comprise a useful substructure, and, lastly, inserting (implanting) the identified useful substructure into the artificial neural network (if not already a substructure thereof).

Combination 4: US2007/0073617 A1 (Denoue et al.) + US2007/0100742 A1 (Ladevich et al.) + NPL - Drouhard et al. (2014)

• US2007/0073617 A1 (Denoue et al.): Describes a "system and method for displaying and interacting with complex networks... to facilitate understanding and analysis of their structure and behavior". This provides a basis for "measuring the activity level of use of certain neural pathways" and identifying structures within a network.
• US2007/0100742 A1 (Ladevich et al.): Discusses creating and training neural networks with an emphasis on "modularity and the ability to combine different network components". This provides the foundation for "selecting an artificial neural network for performing a task" and the mechanism for "inserting (implanting) the identified useful substructure."
  • Motivation to Combine Denoue et al. and Ladevich et al.: A POSITA, analyzing complex neural networks (Denoue et al.) to find effective sub-components, would naturally be motivated by the efficiency and reusability benefits of modular design (Ladevich et al.) to then integrate these identified useful components into other networks for related tasks.
• NPL - Drouhard et al. (2014) "Visual Analytics for Neuroscience-Inspired Dynamic Architectures": This paper, co-authored by inventors, explicitly discusses "utilization of the visualization tool to isolate specific substructures or sub-networks of networks utilized in the recognition of each of the hand-written digits 0 through 9". It also states that "a useful substructure of an artificial neural network is identified for performing a particular sub-task, for example, by measuring the activity level of use of certain neural pathways being above a predetermined level of activity". The paper further describes "forming a problem/component/substructure library or database" for reuse. (While published in 2014, after the priority date, it is listed in the patent's "Prior Art" section with potential 35 U.S.C. § 102 applicability, indicating its relevance for this analysis as provided in the instructions).
  • Motivation to Combine with Denoue et al./Ladevich et al.: Drouhard et al. provides specific examples and methodology for identifying "useful substructures" based on activity, directly supporting the first step of the claim. Combined with the general network analysis tools (Denoue et al.) and modular component integration techniques (Ladevich et al.), a POSITA would find it obvious to apply Drouhard's specific identification method within a broader modular framework to select and insert useful substructures into new or existing ANNs. This improves design efficiency and leverages successful components.

Conclusion for Claim 20: The combination of Denoue et al., Ladevich et al. (US2007/0100742 A1), and Drouhard et al. (2014) would render Claim 20 obvious. Denoue et al. provides the framework for analyzing network activity. Ladevich et al. teaches the modular construction and combination of network components. Drouhard et al. explicitly details identifying useful substructures by measuring neural pathway activity above a threshold. A POSITA, seeking to improve the efficiency and performance of ANN design, would be motivated to combine these teachings to identify effective sub-networks and integrate them into larger systems, avoiding redundant development.

Claim 26 (Non-transitory computer-readable medium)

Claim 26: A non-transitory computer-readable medium containing computer instructions that when executed by a processor cause the processor to perform a method for constructing, using, and reusing components and structures in an artificial neural network, comprising providing a neuroscience-inspired artificial neural network (NIDA) architecture comprising a computational network and at least one affective network, wherein the at least one affective network is coupled to the computational network for regulating at least one parameter associated with a neuron or a synapse, wherein the at least one parameter is adjusted for each impacted like element in the computational network, neuron or synapse, for solving problems in one of control, anomaly detection and classification applications, wherein the neuroscience-inspired artificial neural network architecture is one of simulated on a special purpose processing system or implemented as a dynamic adaptive neural network array (DANNA) of programmable neuromorphic elements.

Combination 5: Yao (2008) + US2011/0082798 A1 (Pyle et al.) + US2004/0148261 A1 (Leinenbach et al.) + General knowledge of neuroscience-inspired computing and affective systems.

• Yao (2008) "Evolving Artificial Neural Network Ensembles": This paper provides an overview of using evolutionary algorithms (EAs) to train neural network weights, highlighting their advantages and applicability to various architectures. This implies "computer instructions" to perform methods for constructing and using ANNs through evolutionary optimization.
• US2011/0082798 A1 (Pyle et al.): This patent application describes a "system and method for creating and simulating a neural network, focusing on graphical programming and visualization of network activity". This directly provides the concept of "computer instructions" on a "non-transitory computer-readable medium" that cause a processor to simulate an ANN, including defining neuron and synapse properties.
  • Motivation to Combine Yao and Pyle et al.: A POSITA implementing the evolutionary training methods described by Yao would naturally utilize a simulation environment (such as that described by Pyle et al.) to run, test, and visualize the evolving neural networks.
• US2004/0148261 A1 (Leinenbach et al.): Discloses reconfigurable neural network architectures that can be dynamically changed and optimized using genetic algorithms. These dynamic architectural changes are a key aspect of "constructing" and "reusing" structures.
  • Motivation to Combine with software/simulation: To create a truly dynamic and adaptive simulated ANN, a POSITA would incorporate the reconfigurable architecture and optimization algorithms of Leinenbach et al. into the software and simulation framework established by Yao and Pyle et al.
• General knowledge of neuroscience-inspired computing and affective systems: The patent emphasizes "neuroscience-inspired" architectures and the role of "affective networks" analogous to neuromodulators.
  • Motivation to Combine with software, simulation, and reconfigurability: To develop more sophisticated and biologically plausible ANNs in software, a POSITA would be motivated to include an "affective network" component within the computer instructions. This affective network would be designed to regulate neuron/synapse parameters (e.g., thresholds) across "impacted like elements" in the computational network, mimicking biological neuromodulation to enhance learning and adaptability for control, anomaly detection, and classification problems. Implementing these functionalities as "computer instructions" on a "non-transitory computer-readable medium" would be a straightforward engineering task.

Conclusion for Claim 26: The combination of Yao (2008), Pyle et al. (US2011/0082798 A1), Leinenbach et al., and general knowledge of neuroscience-inspired computing and affective systems would render Claim 26 obvious. Yao and Leinenbach describe the methods for evolving and reconfiguring ANNs. Pyle et al. teaches the simulation of ANNs using computer instructions on a medium. The integration of an "affective network" to regulate parameters globally stems from the known benefits of incorporating neuroscience principles, particularly neuromodulation, into ANNs for improved adaptive behavior. All these elements implemented as computer instructions on a non-transitory medium would be obvious to a POSITA.

The USPTO does not calculate expiration dates for patents; however, it provides a patent term calculator as a resource to help estimate them. The general rule for utility patents is that the term lasts 20 years from its filing date, with potential adjustments or extensions.

For US patent 10019470B2:

• Filing Date: October 14, 2014
• Issue Date: July 10, 2018
• Priority Date: October 16, 2013

Patent Term Adjustments (PTA)

Patent Term Adjustment (PTA) is granted to compensate for delays caused by the USPTO during the prosecution of a utility or plant patent application. PTA is added to the standard 20-year lifespan of the patent. These delays include, but are not limited to:

• Failure of the USPTO to issue an office action within 14 months of the application filing.
• Failure of the USPTO to respond to a reply or an appeal within four months.
• Failure of the USPTO to act on an application within four months after a decision by the Patent Trial and Appeal Board (PTAB) or a federal court.
• Failure of the USPTO to issue a patent within four months after payment of the issue fee.
• Failure of the USPTO to issue a patent within 36 months from the filing date of an application.

The Google Patents page for US10019470B2 indicates an "Adjusted expiration" date of 2035-04-26. This implies that Patent Term Adjustment (PTA) has been applied to the patent, extending its term beyond the typical 20 years from the filing date.

Patent Term Extensions (PTE)

Patent Term Extension (PTE) is available for patents covering certain products, such as drug products, medical devices, food additives, or color additives, to compensate for delays during regulatory review by agencies like the FDA. The maximum length a patent can be extended under PTE is five years.

Based on the nature of US10019470B2, which relates to artificial neural networks, it is unlikely to be eligible for Patent Term Extension (PTE) under 35 U.S.C. § 156, as it does not appear to cover a product subject to regulatory review by the FDA or other similar agencies. The provided information does not indicate any PTE for this patent.

Continuation Applications, Divisional Applications, and Related Family Members

A continuation application is a second application for the same invention claimed in a prior, co-pending parent application. It uses the same specification as the parent and claims the same priority date but can pursue additional or different claims.

A divisional application is filed when a parent application contains more than one distinct invention. It claims a different invention from the parent but is based on the same disclosure and retains the same filing date as the original application. Divisional applications often arise from a USPTO restriction requirement, where the examiner determines that the original application covers multiple inventions.

The Google Patents page for US10019470B2 lists the following "Other versions":

• US20150106311A1

This indicates that US10019470B2 is a granted patent that stemmed from the application US14/513,388, which was also published as US20150106311A1. US20150106311A1 is the patent application publication of the granted patent US10019470B2. This is a common occurrence where the "other versions" section refers to the published application from which the patent matured. Without further information, it cannot be definitively stated if US20150106311A1 is a continuation or divisional in the traditional sense of claiming new subject matter or distinct inventions from a parent. However, the listed "Application number" US14/513,388 for US10019470B2 and "US20150106311A1" under "Other versions" strongly suggests that US10019470B2 is the grant of the application that was published as US20150106311A1. This means they are directly related as the application and its resulting patent, rather than separate continuing applications.

The patent explicitly states that its application number is US14/513,388, and its publication number is US10019470B2. The "Other versions" link to US20150106311A1 likely refers to the earlier publication of the same application.

Therefore, as of the current date, based on the provided information, US20150106311A1 is a related family member as the application publication of US10019470B2. No other distinct continuation or divisional applications are explicitly identified.

Projected Expiration Date

The Google Patents page for US10019470B2 lists the "Adjusted expiration" date as 2035-04-26. This date reflects the standard 20-year patent term from the filing date (October 14, 2014) plus any Patent Term Adjustment (PTA) awarded due to USPTO delays.

To verify this, the standard 20-year term from the filing date of October 14, 2014, would be October 14, 2034. The adjusted expiration date of April 26, 2035, indicates approximately 6 months and 12 days of PTA. This adjustment is consistent with the provisions for PTA.

Defensive Disclosure for US10019470B2

This Defensive Disclosure document aims to broaden the scope of publicly available prior art related to US patent 10019470B2, titled "Method and apparatus for constructing, using and reusing components and structures of an artificial neural network." The objective is to render future incremental improvements by competitors "obvious" or "non-novel" by describing derivative variations and cross-applications of the core inventive concepts, as of the current date, April 26, 2026.

---

Derivative Variations for Core Claims of US10019470B2

This section details various technical derivatives of the independent claims, focusing on material substitutions, operational parameter expansions, cross-domain applications, integration with emerging technologies, and inverse/failure modes.

Derivatives for Claim 1: Method for constructing, using, and reusing components/structures of an artificial neural network

Claim 1: A method for constructing, using, and reusing components and structures of an artificial neural network, comprising creating a neuroscience-inspired artificial neural network (NIDA) architecture comprising a computational network and at least one affective network, wherein the at least one affective network is coupled to the computational network for regulating at least one parameter associated with a neuron or a synapse, wherein the at least one parameter is adjusted for each impacted like element in the computational network, neuron or synapse, for solving problems in one of control, anomaly detection and classification applications, wherein the neuroscience-inspired artificial neural network architecture is one of simulated on a special purpose processing system or implemented as a dynamic adaptive neural network array (DANNA) of programmable neuromorphic elements.

---

1. Material & Component Substitution

Derivative 1.1: Optoelectronic Dynamic Adaptive Neural Network Array (O-DANNA)

• Enabling Description: The programmable adaptive neuromorphic elements of the DANNA are implemented using optoelectronic components. Neurons are realized as excitable laser diodes or vertical-cavity surface-emitting lasers (VCSELs) with adjustable bias currents that serve as neuron thresholds. Synapses are implemented using electro-optic modulators (e.g., Mach-Zehnder interferometers) whose optical transmission coefficients represent synaptic weights and can be adjusted via applied electrical fields. Synaptic delays are managed by fiber optic delay lines. The "charge" accumulation is represented by optical power integration, and "firing" corresponds to laser emission beyond a threshold. The affective network regulates global parameters, such as the collective bias current thresholds of neuron clusters or the overall gain of synaptic modulators, through a central light-frequency or amplitude modulation signal distributed across the O-DANNA, influencing learning and adaptation for control and classification tasks.

  graph TD
      A[Input Optical Signals] --> B(Optoelectronic Neuron Array)
      B --> C(Optoelectronic Synapse Array)
      C --> B
      B --> D[Output Optical Signals]
      E(Affective Optical Network) --> B
      E --> C
      E -- Regulates Bias/Gain --> B
      E -- Modulates Synaptic Strength --> C
      F[Control/Feedback Loop] --> E
      subgraph O-DANNA Architecture
          B
          C
      end
  

---

2. Operational Parameter Expansion

Derivative 1.2: Deep-Cryogenic DANNA for Exoplanet Data Analysis

• Enabling Description: A DANNA system designed to operate in deep-cryogenic environments (e.g., below 4 Kelvin) for processing sensor data from exoplanetary exploration probes. The programmable neuromorphic elements are based on superconducting Josephson junctions or single-electron transistors (SETs), where qubit states or electron tunneling events represent neuron activity and charge accumulation. Synaptic weights are implemented using adjustable superconducting quantum interference devices (SQUIDs) or cryo-CMOS resistive switches. The affective network modulates the critical current thresholds of Josephson junctions or the gate voltages of SETs uniformly across selected elements, responding to environmental stability (e.g., temperature fluctuations) or data fidelity. This ensures reliable anomaly detection and classification of subtle biosignatures or geological features in extremely noisy and low-power environments, where quantum effects are pronounced.

  stateDiagram-v2
      [*] --> Initializing[Initialize Cryogenic DANNA]
      Initializing --> Operational[Operational Mode @ <4K]
      Operational --> DataAcquisition[Acquire Exoplanet Data]
      DataAcquisition --> ComputationalNetwork[Process Data (SETs/JJ)]
      ComputationalNetwork --> AffectiveNetwork[Monitor Performance/Environment]
      AffectiveNetwork --> RegulateThresholds[Adjust JJ Critical Current / SET Gate Voltage]
      RegulateThresholds --> ComputationalNetwork
      AffectiveNetwork --> DetectAnomaly[Anomaly Detection / Classification]
      DetectAnomaly --> ReportFindings[Output Results]
      Operational --> LowPower[Low Power / Standby Mode]
      LowPower --> Operational: Energy Demand/Data Arrival
      Operational --> FaultDetected[Hardware Fault / Environmental Anomaly]: Critical Failure
      FaultDetected --> SafeShutdown[Safe Shutdown / Diagnostic Mode]
  

---

3. Cross-Domain Application

Derivative 1.3.1: Personalized Medicine - Drug Response Prediction

• Enabling Description: The NIDA/DANNA is applied to personalized medicine for predicting patient response to specific drug therapies. The computational network ingests patient genomic data (e.g., SNP profiles, gene expression), clinical history, and proteomic biomarkers. Output neurons predict the likelihood and severity of response (e.g., efficacy, adverse effects). An affective network, coupled to the computational network, dynamically adjusts parameters like the sensitivity thresholds of specific neuron clusters associated with drug metabolism pathways or immune responses. This modulation is based on real-time physiological data (e.g., inflammation markers, liver enzyme levels) or patient-reported symptoms, allowing the NIDA to adapt its predictive model for individual patient variability and flag anomalous responses.

  graph TD
      A[Genomic Data Input] --> CN(Computational Network - Drug Response Model)
      B[Clinical History Input] --> CN
      C[Proteomic Biomarkers Input] --> CN
      D[Real-time Physiological Data] --> AN(Affective Network - Patient State Monitor)
      AN -- Regulates Neuron Sensitivity/Weight --> CN
      CN --> E[Predicted Drug Response (Efficacy/Side Effects)]
      CN --> F[Anomaly: Unexpected Response]
      AN --> F
      E --> G[Physician Interface]
      F --> G
  

Derivative 1.3.2: Autonomous Robotics - Swarm Coordination

• Enabling Description: A DANNA-based control system for managing autonomous robotic swarms in complex environments (e.g., warehouse automation, search and rescue). Each robot in the swarm contributes its sensor data (lidar, camera, IMU) to a distributed computational network. The NIDA's computational network processes local and global environmental states, issuing movement commands and task allocations. The affective network, operating across the swarm, monitors overall swarm coherence, energy levels, and mission progress. It adjusts parameters such as individual robot speed, collision avoidance sensitivity, or cooperative task prioritization based on observed collective performance or emergent threats, enhancing robustness and adaptability in dynamic, multi-agent control scenarios.

  flowchart TD
      A[Robot Sensor Data (Lidar, Camera, IMU)] --> CN(Computational Network - Swarm State & Path Planning)
      CN --> B{Individual Robot Action Commands}
      B --> C[Robot Actuators (Movement, Grippers)]
      D[Swarm Coherence Metrics] --> AN(Affective Network - Swarm State Modulator)
      E[Energy Levels / Mission Progress] --> AN
      AN -- Adjusts: Speed, Collision Sensitivity, Task Priority --> CN
      CN --> F[Outputs to Other Robots / Central Controller]
      subgraph Autonomous Swarm System
          A
          B
          C
          D
          E
          CN
          AN
          F
      end
  

Derivative 1.3.3: Environmental Monitoring - Predictive Hazard Mapping

• Enabling Description: The NIDA/DANNA is deployed in a network of environmental sensors (e.g., air quality, water contamination, seismic activity) for predictive hazard mapping and early warning systems. The computational network analyzes multivariate sensor streams, identifying patterns indicative of developing environmental hazards (e.g., pollution plumes, seismic precursors, wildfire risks). An affective network continuously monitors the confidence levels of the predictions, sensor network health, and the urgency of potential hazards. It dynamically tunes parameters within the computational network, such as the sensitivity thresholds of neurons processing specific pollutant signatures or the temporal window for seismic event correlation, to balance false positives/negatives based on current environmental conditions and historical data, thereby improving the accuracy and timeliness of anomaly detection and hazard classification.

  sequenceDiagram
      SensorNetwork->>ComputationalNetwork: Stream Environmental Data
      ComputationalNetwork->>AffectiveNetwork: Provide Prediction Confidence & Network Health
      AffectiveNetwork->>ComputationalNetwork: Adjust Neuron Thresholds & Synapse Weights (Sensitivity, Time Windows)
      ComputationalNetwork->>PredictionEngine: Output Hazard Probabilities
      PredictionEngine->>AlertSystem: Trigger Warnings for High Probability Hazards
      AlertSystem->>HumanOperators: Display Predictive Hazard Map
      HumanOperators->>AffectiveNetwork: Feedback on Alert Accuracy/Urgency
  

---

4. Integration with Emerging Tech

Derivative 1.4: Blockchain-Secured, AI-Optimized DANNA with IoT Edge Sensing

• Enabling Description: The DANNA operates as an IoT edge device, receiving real-time data from localized sensors. An AI-driven meta-learning agent (e.g., a Reinforcement Learning algorithm operating on a separate computational substrate) continuously observes the DANNA's performance and environment, dynamically optimizing the architecture (neuron/synapse addition/deletion, connection topology) and the affective network's parameters (e.g., learning rates, threshold adjustment magnitudes) via secure APIs. Identified useful substructures (per Claim 20) and their associated performance metrics, along with the configurations of the affective and computational networks, are cryptographically hashed and recorded on a private blockchain. This blockchain ensures immutable provenance, verifies the authenticity of reused components, and provides a transparent audit trail for model evolution and adaptation, especially critical for regulatory compliance in high-stakes applications like autonomous vehicles or critical infrastructure control.

  graph TD
      IoT_Sensors[IoT Sensor Array] --> Data_Stream[Real-time Data Stream]
      Data_Stream --> DANNA_Edge[DANNA Edge Device (Computational Network)]
      DANNA_Edge --> Affective_Net[Affective Network]
      Affective_Net --> DANNA_Edge
      DANNA_Edge -- Performance Metrics --> AI_Optimizer[AI Meta-Learning Optimizer]
      AI_Optimizer -- Architectural & Parameter Updates --> DANNA_Edge
      DANNA_Edge -- Substructure Hashes & Config Logs --> Blockchain_Ledger[Blockchain Ledger]
      Blockchain_Ledger -- Verifiable Records --> Trusted_Actors[Regulators/Developers]
      subgraph Core DANNA Operations
          DANNA_Edge
          Affective_Net
      end
      subgraph External Systems
          IoT_Sensors
          AI_Optimizer
          Blockchain_Ledger
          Trusted_Actors
      end
  

---

5. The "Inverse" or Failure Mode

Derivative 1.5: Resilient DANNA with Self-Diagnostic Affective Shutdown

• Enabling Description: A DANNA system incorporating a specialized "resilience affective network" designed for safe failure and graceful degradation. This network constantly monitors internal computational network health metrics, including neuron firing rates, synapse weight distributions, and internal error propagation, as well as external parameters like power supply stability and external input integrity. Upon detecting anomalies exceeding predefined thresholds (e.g., sustained hyper-activity, zero firing, critical power drop), the resilience affective network initiates a cascade of parameter adjustments. This includes increasing neuron refractory periods, imposing maximum/minimum bounds on synaptic weights to prevent runaway potentiation/depression, and logically isolating compromised substructures. In severe cases, it can trigger a "low-power, limited-functionality mode" (e.g., only processing vital safety inputs with a pre-trained, minimal network) or a "diagnostic-only mode" where all learning is paused, and network state is logged for external analysis, ensuring system integrity and preventing catastrophic failure in critical control or anomaly detection applications.

  stateDiagram-v2
      [*] --> Operational
      Operational --> Monitoring[Monitoring Internal & External Health]
      Monitoring --> Operational: All OK
      Monitoring --> AnomalyDetected[Anomaly Detected (e.g., High Error Rate, Power Drop)]
      AnomalyDetected --> AdjustParameters[Adjust Neuron Refractory Periods, Weight Bounds]
      AdjustParameters --> IsolateSubstructures[Isolate Compromised Substructures]
      IsolateSubstructures --> DegradedMode[Degraded Performance Mode]
      DegradedMode --> Operational: Anomaly Resolved
      DegradedMode --> LowPowerMode[Low-Power, Limited-Functionality Mode]
      LowPowerMode --> Operational: Anomaly Resolved
      LowPowerMode --> DiagnosticMode[Diagnostic-Only Mode]
      DiagnosticMode --> SafeShutdown[Safe System Shutdown]: Unrecoverable Anomaly
      Operational --> SafeShutdown: Manual Override / Catastrophic Failure
  

---

Derivatives for Claim 8: Apparatus for constructing, using, and reusing components/structures of an artificial neural network

Claim 8: An apparatus for constructing, using, and reusing components and structures of an artificial neural network, comprising a special purpose processing system or a dynamic adaptive neural network array (DANNA) comprised of programmable adaptive neuromorphic elements for one of simulating or implementing a neuroscience-inspired artificial neural network (NIDA) architecture comprising a computational network and at least one affective network, wherein the at least one affective network is coupled to the computational network for controlling at least one parameter associated with a neuron or a synapse, wherein the at least one parameter is adjusted for each impacted like element in the computational network, neuron or synapse, for solving problems in one of control, anomaly detection and classification applications.

---

1. Material & Component Substitution

Derivative 8.1: Spiking Neural Network (SNN) Processor with Memristor-based DANNA

• Enabling Description: The apparatus is a specialized SNN processor where programmable adaptive neuromorphic elements are implemented using hybrid CMOS-memristor technology. Neurons are integrated CMOS circuits simulating Izhikevich or Leaky Integrate-and-Fire (LIF) models, and synapses are non-volatile memristors whose conductance directly represents synaptic weight and can be analogically tuned (LTP/LTD). Synaptic delays are implemented via digitally controlled delay lines within the CMOS substrate. The affective network is a dedicated analog neuromodulation circuit that generates global control signals (e.g., variable bias currents, global reset signals, or neuromodulator-mimicking analog voltage levels) which are distributed to specific memristor arrays or CMOS neuron circuits. This allows for dynamic, network-wide adjustment of parameters like neuron excitability thresholds or memristor learning rates, facilitating highly energy-efficient processing for real-time control applications.

  classDiagram
      class SNNProcessor {
          +CMOS_Neurons
          +Memristor_Synapses
          +Digital_DelayLines
          +Neuromodulation_Circuit
      }
      class CMOS_Neurons {
          +LIF_Model()
          +Izhikevich_Model()
          +Threshold_Control_Input
      }
      class Memristor_Synapses {
          +Conductance_Weight
          +LTP_LTD_Mechanism
          +LearningRate_Control_Input
      }
      class Neuromodulation_Circuit {
          +Generate_GlobalControlSignals()
          +AdjustNeuronExcitability()
          +AdjustMemristorLearningRate()
      }
      SNNProcessor *-- CMOS_Neurons
      SNNProcessor *-- Memristor_Synapses
      SNNProcessor *-- Digital_DelayLines
      SNNProcessor *-- Neuromodulation_Circuit : Affective Network
      Neuromodulation_Circuit --> CMOS_Neurons : Controls Threshold
      Neuromodulation_Circuit --> Memristor_Synapses : Controls Learning Rate
  

---

2. Operational Parameter Expansion

Derivative 8.2: Terahertz (THz) Communication Enabled DANNA for High-Throughput Processing

• Enabling Description: The DANNA apparatus is designed for extremely high-frequency operation, leveraging terahertz (THz) band communication for inter-element signaling. Programmable neuromorphic elements are realized using graphene-based field-effect transistors (FETs) or plasmonic modulators for neurons and optically-controlled THz waveguides for synapses. Neuron firing events are encoded as short THz pulses, and synaptic weights are determined by the attenuation or phase shift induced in the THz waveguides. The affective network, implemented with a dedicated THz frequency comb generator and array of programmable filters/attenuators, modulates global parameters by shifting the central frequency, bandwidth, or amplitude of the THz carrier waves propagating through the DANNA. This provides ultra-fast, distributed control over neuron excitability or synaptic plasticity, enabling real-time anomaly detection in massive, high-bandwidth data streams (e.g., 6G wireless network traffic, advanced radar systems).

  graph TD
      A[High-Bandwidth Data Input] --> B(THz Transceiver Array)
      B --> C(Graphene Neuron Array)
      C --> D(Plasmonic Synapse Array)
      D --> C
      C --> E(THz Output Transceiver Array)
      F(THz Frequency Comb Generator) --> G(Affective Control Module)
      G -- Modulates THz Carrier --> C
      G -- Modulates THz Carrier --> D
      E --> H[Processed High-Throughput Output]
      subgraph THz-DANNA Apparatus
          B
          C
          D
          E
      end
  

---

3. Cross-Domain Application

Derivative 8.3.1: Autonomous Maritime Navigation & Anomaly Detection

• Enabling Description: An apparatus comprising a ruggedized DANNA system onboard autonomous underwater vehicles (AUVs) or surface vessels for real-time navigation and anomaly detection in marine environments. The programmable neuromorphic elements process sonar, lidar, GPS, and environmental sensor data (current, salinity, temperature). The computational network performs simultaneous localization and mapping (SLAM), obstacle avoidance, and mission planning. An affective network, coupled to the computational network, monitors the AUV's operational state (e.g., power consumption, sensor reliability, system integrity) and environmental conditions (e.g., unexpected currents, marine life density). It dynamically adjusts the confidence thresholds for SLAM features, the sensitivity of obstacle detection neurons, or the reactive speed of evasion maneuvers, to adapt to changing mission parameters or detect unpredicted environmental events, improving autonomous operation and safety.

  flowchart LR
      SensorInputs[Sonar, Lidar, GPS, Environmental Data] --> DANNA_Apparatus
      DANNA_Apparatus[DANNA Apparatus] --> NavigationOutputs[SLAM, Obstacle Avoidance, Mission Plan]
      DANNA_Apparatus --> AnomalyDetectionOutputs[Environmental Anomaly, System Fault]
      OperationalMetrics[Power, Sensor Health, System Integrity] --> AffectiveNetwork_Module
      EnvironmentalConditions[Currents, Marine Life, Seabed Changes] --> AffectiveNetwork_Module
      AffectiveNetwork_Module[Affective Network] -- Adjusts: Confidence Thresholds, Detection Sensitivity, Maneuver Speed --> DANNA_Apparatus
  

Derivative 8.3.2: Industrial Process Control - Adaptive Manufacturing

• Enabling Description: An apparatus featuring a DANNA integrated into a manufacturing plant's control system for adaptive process optimization and predictive maintenance. The programmable neuromorphic elements receive real-time data from various sensors (e.g., temperature, pressure, flow rates, vibration, optical inspection) throughout the production line. The computational network performs closed-loop control of machinery, quality inspection, and identifies deviations from optimal operating parameters. An affective network dynamically adjusts control loop gains, tolerance thresholds for quality defects, or predictive maintenance scheduling parameters. This adjustment is based on overall production yield, energy consumption, and detected subtle shifts in machine health, enabling the system to adapt to material variations, wear and tear, and environmental changes, classifying and preventing manufacturing anomalies before they lead to defects or downtime.

  graph TD
      Sensors[Process Sensors (Temp, Pressure, Flow, Vibration)] --> DANNA_Apparatus
      DANNA_Apparatus[DANNA Apparatus] --> ControlOutputs[Machine Actuator Commands]
      DANNA_Apparatus --> QualityAnalysis[Defect Detection, Process Deviation]
      ManufacturingMetrics[Production Yield, Energy Consumption, Machine Health] --> AffectiveNetwork_Module
      AffectiveNetwork_Module[Affective Network] -- Adjusts: Control Gains, Tolerance Thresholds, Maintenance Schedule --> DANNA_Apparatus
  

Derivative 8.3.3: Space Debris Tracking and Collision Avoidance

• Enabling Description: An apparatus for space debris tracking and collision avoidance, comprising a DANNA system deployed in ground-based or orbital observatories. The programmable neuromorphic elements process data from radar, optical telescopes, and lidar systems, constructing and updating a real-time catalog of orbital objects. The computational network performs trajectory prediction, identifies close approaches, and classifies objects based on size and velocity. An affective network, coupled to the computational network, monitors the fidelity of tracking data, the density of debris in specific orbital regions, and the risk level of potential collisions. It dynamically adjusts parameters such as the sensitivity thresholds for detecting faint radar returns, the weight given to recent observation data in trajectory calculations, or the urgency thresholds for issuing collision warnings. This ensures an adaptive and robust system for anomaly detection (new debris, unexpected trajectories) and classification of collision risks in a highly dynamic and sparse data environment.

  sequenceDiagram
      ObservationSystems->>DANNA_Apparatus: Radar, Optical, Lidar Data
      DANNA_Apparatus->>ComputationalNetwork: Raw Observation Data
      ComputationalNetwork->>AffectiveNetwork: Tracking Fidelity, Debris Density, Collision Risk
      AffectiveNetwork->>ComputationalNetwork: Adjust: Detection Thresholds, Data Weighting, Warning Urgency
      ComputationalNetwork->>SpaceTrafficControl: Output Debris Catalog & Collision Warnings
      SpaceTrafficControl->>SatelliteOperators: Alert on Collision Risks
  

---

4. Integration with Emerging Tech

Derivative 8.4: Federated Learning Enabled DANNA Fabric with Homomorphic Encryption

• Enabling Description: The DANNA apparatus is architected as a distributed neuromorphic fabric where multiple individual DANNA nodes (special purpose processing systems) collaboratively train a global model using federated learning principles. Each local DANNA contains a computational and affective network. Training data remains local to each node. Updates to model parameters (neuron thresholds, synapse weights) and affective network configurations are exchanged between nodes. To ensure data privacy, these updates are performed using homomorphic encryption before aggregation by a central server or blockchain. The affective network at each node, and potentially a federated affective network, dynamically adjusts local learning rates, privacy-preserving noise addition parameters, or communication frequencies based on local data characteristics, node performance, and global model convergence, providing secure and adaptive anomaly detection or classification across diverse datasets without compromising privacy.

  graph LR
      Node1[DANNA Node 1] -- Encrypted Updates --> Aggregator
      Node2[DANNA Node 2] -- Encrypted Updates --> Aggregator
      NodeN[DANNA Node N] -- Encrypted Updates --> Aggregator
      Aggregator[Federated Aggregator (Homomorphic)] -- Global Model Update --> Node1
      Aggregator -- Global Model Update --> Node2
      Aggregator -- Global Model Update --> NodeN
      Node1 -- Local Affective Control --> Node1
      Node2 -- Local Affective Control --> Node2
      NodeN -- Local Affective Control --> NodeN
      subgraph Each DANNA Node
          Node1_CN[Computational Net]
          Node1_AN[Affective Net]
          Node1_CN <--> Node1_AN
          Node1_LocalData[Local Private Data] --> Node1_CN
      end
  

---

5. The "Inverse" or Failure Mode

Derivative 8.5: Fault-Tolerant DANNA with Redundant Affective Sub-Networks

• Enabling Description: An apparatus where the DANNA is implemented with inherent fault tolerance through hardware redundancy and distributed affective sub-networks. The programmable neuromorphic elements are arranged in redundant blocks (e.g., triple modular redundancy for critical processing paths). The overall affective system is decomposed into multiple, geographically or logically distributed affective sub-networks, each monitoring a subset of the computational network and its environmental context. If a fault is detected (e.g., via voter disagreement in redundant blocks, or anomalous behavior reported by a computational network segment), the affected local affective sub-network initiates localized parameter adjustments (e.g., re-routing signals, increasing refractory periods in faulty areas, activating backup elements). In catastrophic failure of a sub-network, a higher-level "supervisory affective network" takes over, adjusting global parameters to isolate the faulty region, reduce overall processing load, and maintain essential functionality in a degraded, but stable, operational mode for anomaly detection or control.

  flowchart TD
      CN_A[Computational Network Block A] -- Voting Logic --> Output_A
      CN_B[Computational Network Block B (Redundant)] -- Voting Logic --> Output_A
      CN_C[Computational Network Block C (Redundant)] -- Voting Logic --> Output_A
      AS_1[Affective Sub-Network 1] -- Monitor & Regulate --> CN_A
      AS_2[Affective Sub-Network 2] -- Monitor & Regulate --> CN_B
      AS_3[Affective Sub-Network 3] -- Monitor & Regulate --> CN_C
      AS_1 -- Fault Report --> SAN
      AS_2 -- Fault Report --> SAN
      AS_3 -- Fault Report --> SAN
      SAN[Supervisory Affective Network] -- Global Regulation & Reconfiguration --> CN_A
      SAN -- Global Regulation & Reconfiguration --> CN_B
      SAN -- Global Regulation & Reconfiguration --> CN_C
      SAN -- Fault Isolation --> Output_A
      subgraph Fault-Tolerant DANNA
          CN_A
          CN_B
          CN_C
          AS_1
          AS_2
          AS_3
          SAN
      end
  

---

Derivatives for Claim 20: Method for identifying, selecting, and integrating useful substructures

Claim 20: A method for constructing, using, and reusing components and structures in an artificial neural network, comprising identifying a useful substructure of an artificial neural network for performing a particular sub-task, by measuring the activity level of use of certain neural pathways being above a predetermined level of activity, then, selecting an artificial neural network for performing a task of which the sub-task and its identified neural pathway may comprise a useful substructure, and, lastly, inserting (implanting) the identified useful substructure into the artificial neural network (if not already a substructure thereof).

---

1. Material & Component Substitution

Derivative 20.1: Biologically-Inspired Substructure Identification via Neurotransmitter Activity Mapping

• Enabling Description: The method is applied to in-vitro biological neural networks (e.g., neuronal cultures on multi-electrode arrays or brain-on-a-chip platforms). A "useful substructure" for a specific bio-computational sub-task (e.g., pattern completion, basic learning) is identified by mapping regions of high neurotransmitter release or receptor activation, instead of electrical activity, above a predetermined chemical concentration threshold. This mapping is achieved using advanced microscopy and biosensors. The selected "artificial neural network" is another in-vitro or in-silico network being developed for a broader task. The "inserting" involves either microfluidic delivery of neurotrophic factors to promote specific axonal sprouting and synaptogenesis in the in-vitro network, or direct replication of the identified network topology and synaptic weight profile (derived from the chemical activity map) into a simulated ANN (e.g., with neuromodulated LIF neurons).

  flowchart TD
      A[In-Vitro Neural Network (MEA/Organoid)] --> B(Apply Sub-task Stimulus)
      B --> C{Monitor Neurotransmitter Release / Receptor Activation}
      C --> D{Map Regions of Activity > Threshold}
      D --> E[Identify Useful Substructure (Chemical Activity Map)]
      E --> F[Select Target ANN (In-Vitro / In-Silico)]
      F --> G{Microfluidic Synaptogenesis / In-Silico Replication}
      G --> H[Implant Substructure into Target ANN]
  

---

2. Operational Parameter Expansion

Derivative 20.2: Ultra-Large Scale (ULS) Network Substructure Extraction with Distributed Activity Tracing

• Enabling Description: A method for identifying useful substructures within an ultra-large-scale (ULS) artificial neural network comprising billions of neuromorphic elements, distributed across geographically dispersed data centers. "Activity level of use" of neural pathways is measured by distributed telemetry and asynchronous logging of activation events, aggregated and analyzed using graph database technologies. Instead of a simple "predetermined level," the threshold for activity is dynamically set based on the network's overall sparse activity levels and the statistical significance of a pathway's contribution to successful sub-task completion. The "selecting" and "inserting" steps involve automated, resource-aware deployment mechanisms to instantiate the identified substructure (potentially with dynamic scaling) within existing or newly provisioned ULS ANN segments, ensuring optimal resource utilization for the overall task. This applies to large-scale data processing for scientific simulations or global internet traffic management.

  graph TD
      ULS_NN[Ultra-Large Scale Neural Network (Distributed)] --> DT[Distributed Telemetry & Logging]
      DT --> GDB[Graph Database Analytics]
      GDB --> DSA[Dynamic Statistical Analysis of Activity]
      DSA --> ID_Sub[Identify Useful Substructure (Statistical Significance)]
      ID_Sub --> RS[Resource-Aware Substructure Selection]
      RS --> ADP[Automated Deployment & Provisioning]
      ADP --> ULS_NN
  

---

3. Cross-Domain Application

Derivative 20.3.1: Supply Chain Optimization - Anomaly-Resilient Routing Substructure

• Enabling Description: The method is applied in supply chain management to identify and reuse neural network substructures that are highly effective at handling routing anomalies (e.g., unexpected delays, port closures, inventory shortages). The "sub-task" is resilient route planning under perturbation. The "activity level" is measured as the activation frequency and magnitude of neural pathways involved in re-routing decisions that successfully maintain delivery schedules despite disruptions. Once a highly effective "anomaly-resilient routing substructure" is identified within a global logistics planning ANN, it is selected and inserted into regional or local supply chain networks to improve their robustness and adaptability to unforeseen events.

  flowchart TD
      GlobalLogisticsANN[Global Logistics Planning ANN] --> MonitorRouting[Monitor Routing Decisions Under Disruption]
      MonitorRouting --> PathwayActivity[Measure Neural Pathway Activity (Re-routing)]
      PathwayActivity --> IdentifyResilient[Identify Resilient Routing Substructure (High Activity, Successful Resolution)]
      IdentifyResilient --> SelectTarget[Select Target Regional/Local SCM ANN]
      SelectTarget --> InsertSubstructure[Insert Substructure into Target SCM ANN]
  

Derivative 20.3.2: Cybersecurity - Adaptive Threat Detection Module

• Enabling Description: The method is used in cybersecurity to identify efficient neural network substructures for detecting specific types of cyber threats (e.g., particular malware families, phishing patterns). The "sub-task" is the accurate classification of a known threat signature. The "activity level" is measured by the neuronal firing rates and synaptic current flows within specific pathways that correlate strongly with correct threat identification. Once an effective "adaptive threat detection module" is identified (e.g., a specific convolutional or recurrent sub-network), it is extracted and inserted into new or evolving intrusion detection systems (IDS) or security information and event management (SIEM) platforms to enhance their ability to recognize emerging or polymorphic threats without extensive retraining of the entire system.

  sequenceDiagram
      CyberThreatANN->>ThreatDetectionEngine: Process Network Traffic
      ThreatDetectionEngine->>PathwayMonitor: Monitor Neuronal Activity for Threat Signatures
      PathwayMonitor->>SubstructureIdentifier: Identify High Activity Pathways for Correct Detections
      SubstructureIdentifier->>ModuleSelector: Select Effective Threat Detection Substructure
      ModuleSelector->>IDS/SIEMPlatform: Insert Substructure as Adaptive Module
  

Derivative 20.3.3: Game AI - Behavior Generation for NPCs

• Enabling Description: The method is applied in game development for identifying reusable neural network substructures that generate specific complex behaviors for Non-Player Characters (NPCs) (e.g., realistic pathfinding, nuanced emotional responses, sophisticated combat tactics). The "sub-task" is the generation of a particular desired NPC behavior. The "activity level" is measured by analyzing the activation patterns within the NPC's behavior-generating ANN during successful execution of the sub-task (e.g., shortest path found, emotionally appropriate dialogue generated). Once a "useful behavior generation substructure" is identified, it is extracted and implanted into other NPCs or other game levels, enabling rapid prototyping and consistent, high-quality AI behavior across different game contexts, reducing development time and computational overhead.

  graph LR
      NPCAI_ANN[NPC AI ANN] --> BehaviorSim[Simulate Desired Behavior (Sub-task)]
      BehaviorSim --> ActivityMonitor[Monitor Neural Activity Patterns]
      ActivityMonitor --> IdentifyBehaviorSub[Identify Useful Behavior Substructure (High Activity for Success)]
      IdentifyBehaviorSub --> OtherNPCs[Target NPCs/Game Levels]
      OtherNPCs --> ImplantSubstructure[Implant Substructure]
  

---

4. Integration with Emerging Tech

Derivative 20.4: Decentralized Substructure Marketplace with Smart Contract Verified Performance

• Enabling Description: The method is augmented by integrating with blockchain and smart contract technologies to facilitate a decentralized marketplace for verified ANN substructures. When a "useful substructure" is identified by measuring activity levels for a sub-task, its architecture, functional description, and validated performance metrics are hashed and uploaded to a blockchain. A smart contract then tokenizes this substructure. When an "artificial neural network" requires this sub-task capability, it can query the blockchain marketplace. The "inserting" process is mediated by a smart contract: upon purchase/license, the contract automatically verifies the substructure's integrity and performance (against recorded hashes and metrics) before releasing it for implantation. This system enables secure, transparent, and economically incentivized reuse of high-performing ANN components across different organizations and ensures the provenance and quality of acquired substructures.

  sequenceDiagram
      ANN_Developer->>Substructure_Identifier: Identify Useful Substructure
      Substructure_Identifier->>Blockchain_Registry: Register Substructure (Hash, Performance, Metadata)
      Blockchain_Registry->>Smart_Contract: Create Tokenized Substructure
      Another_ANN_Developer->>Blockchain_Registry: Search for Substructure
      Blockchain_Registry->>Smart_Contract: Initiate License/Purchase
      Smart_Contract->>Another_ANN_Developer: Verify & Release Substructure
      Another_ANN_Developer->>Target_ANN: Insert Substructure
  

---

5. The "Inverse" or Failure Mode

Derivative 20.5: Inactive Pathway Pruning for Computational Efficiency in Limited-Resource Environments

• Enabling Description: A method focused on identifying unnecessary or detrimental substructures to improve efficiency in resource-constrained environments (e.g., edge AI devices, embedded systems). The "sub-task" is identified as any operation performed by a neural pathway whose "activity level of use" falls below a predetermined minimal threshold for an extended period, or whose activity is statistically correlated with decreased overall network performance or error rates. Instead of selecting an ANN for insertion, the method identifies segments of the current ANN that are either redundant or actively harmful. The "inserting" step is replaced by a "pruning" or "deactivation" process, where these identified inactive or detrimental neural pathways/substructures are logically disconnected, their weights set to zero, or their corresponding physical neuromorphic elements are reconfigured for other tasks or put into a low-power dormant state, thereby optimizing resource usage and improving overall network efficiency.

  graph TD
      ANN_ResourceConstrained[ANN in Resource-Constrained Environment] --> MonitorPathwayActivity[Monitor Neural Pathway Activity & Performance Correlation]
      MonitorPathwayActivity --> IdentifyInactiveDetrimental[Identify Inactive or Detrimental Substructures (Activity < Threshold / Negative Correlation)]
      IdentifyInactiveDetrimental --> Resource_Optimizer[Resource Optimizer]
      Resource_Optimizer --> Prune_Deactivate[Prune / Deactivate Pathways]
      Prune_Deactivate --> ReconfigureElements[Reconfigure Neuromorphic Elements]
      ReconfigureElements --> ANN_ResourceConstrained
  

---

Combination Prior Art Scenarios with Open-Source Standards

This section outlines how the concepts within US10019470B2, and its derivatives, could be combined with existing open-source standards to establish obviousness for further incremental improvements.

1. US10019470B2 (Claim 1/8/16 - NIDA/DANNA with Affective Networks) + TensorFlow/Keras (Open-Source Machine Learning Framework):

   • Scenario: A method or apparatus for implementing a neuroscience-inspired artificial neural network (NIDA) or dynamic adaptive neural network array (DANNA) (as described in claims 1, 8, or 16), which includes a computational network and an affective network coupled to regulate neuron/synapse parameters, is straightforwardly implemented using the TensorFlow/Keras framework. The computational network can be constructed using Keras layers (Dense, Conv2D, LSTM), while the affective network's logic for parameter regulation (e.g., adjusting learning rates, dropout rates, activation function thresholds, or custom layer weights in a Keras Callback or custom Layer) can be coded within TensorFlow's graph operations. The core concept of adaptive parameter adjustment by an affective component would be an obvious design choice for a practitioner using these flexible frameworks to achieve desired network behaviors or optimize performance, leveraging TensorFlow's extensive API for dynamic computation graphs and custom operations.

2. US10019470B2 (Claim 1/8 - Control Applications with Affective Networks) + OpenAI Gym (Open-Source Reinforcement Learning Platform):

   • Scenario: The application of a NIDA/DANNA (as described in claims 1 or 8) for solving "control problems" (e.g., pole balancing, robotics control) using an affective network to regulate computational network parameters. This would be an obvious combination with OpenAI Gym environments. An agent implementing the NIDA/DANNA (e.g., via a Python interface to a simulated or hardware DANNA) would interact with the Gym environment's observation and action spaces. The affective network's input could derive from the environment's reward signals or specific state variables (e.g., cumulative error, stability metrics), and its output (parameter adjustments) would directly influence the learning and behavior of the computational network within the Gym's simulation loop. This integration would be a natural extension for research and development in adaptive control using biologically inspired neural networks.

3. US10019470B2 (Claim 20 - Identifying and Reusing Substructures) + ONNX (Open Neural Network Exchange Format):

   • Scenario: The method of identifying "useful substructures" within an artificial neural network (as described in claim 20) by measuring activity levels and subsequently "inserting" these substructures into other ANNs. This becomes an obvious practice when combined with the ONNX format. Identified substructures, once isolated (e.g., as sub-graphs of a larger network), can be exported into the ONNX standard format, enabling their seamless "insertion" and reuse across different deep learning frameworks (TensorFlow, PyTorch) or hardware platforms (DANNA implementations that support ONNX import). The ONNX format explicitly supports modular network definitions, making the identification, serialization, and deserialization of "substructures" a technically straightforward and desirable operation for promoting interoperability and reusability of learned components.