US 10095718B2

US patent 10095718B2, titled "Method and apparatus for constructing a dynamic adaptive neural network array (DANNA)," was issued to the University of Tennessee Research Foundation on October 9, 2018. The application for this patent was filed on October 14, 2014. The inventors listed are J. Douglas Birdwell, Mark E. Dean, and Catherine Schuman.

Abstract:
The technical field relates to a method and apparatus for constructing a neuroscience-inspired artificial neural network embodied in a dynamic adaptive neural network array (DANNA). Specifically, it concerns applying a conventional von Neumann architecture to support a neuroscience-inspired artificial neural network's dynamic architecture in software, and constructing a DANNA from primary neuron and synapse elements of components like programmable logic arrays, application-specific integrated circuits, VLSI components, or other similar components. This is intended for solving problems in control, anomaly detection, and classification.

Independent Claims Overview:
(Information on claims is sourced from the full patent text available via Google Patents, as the claims were not fully detailed in the initial provided text.)

• Claim 1 (Method Claim): This claim describes a method for constructing a dynamic adaptive neural network array (DANNA). It involves defining an array of addressable electronic components, where each component can be configured as either a neuron or a synapse, but not both simultaneously. The method then configures a subset of these components to create a neuroscience-inspired artificial neural network with a dynamic architecture. This network includes a computational network and at least one affective network. The computational network has input neurons, output neurons, and hidden neurons connected by synapses, where the neurons and synapses are configured from the electronic components. The affective network is coupled to the artificial neural network to regulate at least one parameter of a neuron or a synapse. The method further involves applying at least one evolutionary optimization algorithm to dynamically modify the network structure (neurons, synapses, and their interconnections) and the parameters of the neurons and synapses.

• Claim 14 (Apparatus Claim): This claim describes an apparatus designed to construct a dynamic adaptive neural network array (DANNA). The apparatus includes an array of addressable electronic components, with each component capable of being configured as either a neuron or a synapse. It also comprises configuration circuitry, which is adapted to configure a subset of these electronic components to form a neuroscience-inspired artificial neural network with a dynamic architecture. This network includes a computational network and at least one affective network. Similar to the method claim, the computational network has input, output, and hidden neurons interconnected by synapses, all formed from the electronic components. The affective network is coupled to regulate at least one parameter of a neuron or a synapse. The apparatus further includes a control and optimizing device that applies at least one evolutionary optimization algorithm to dynamically modify the network's structure and the parameters of its neurons and synapses.

• Claim 16 (Method Claim for Dynamic Adaptive Neural Network): This claim describes a method for operating a dynamic adaptive neural network. It involves initializing a population of neural networks, each comprising an array of addressable electronic components configurable as neurons or synapses. For each network, a fitness value is calculated based on its performance in solving a problem. The method then selects at least two "parent" networks based on their fitness values and applies crossover and mutation operations to these parents to generate a "child" population of neural networks. The child population includes both modified networks and potentially identical copies of high-performing parents. Finally, the network whose fitness best meets predefined requirements is selected as the output.

CAFC 2026 Dockets:
As of April 26, 2026, no specific litigation or docket entries concerning US patent 10095718B2 were found in the CAFC 2026 dockets.

As of April 26, 2026, there is no known litigation involving US patent 10095718B2. Searches of patent litigation databases such as Unified Patents and PACER did not return any results for this specific patent number. While PACER is a national index for federal court cases, accessing detailed information typically requires a registered account and may incur fees. The CAFC dockets for 2026 also do not show any specific entries for US patent 10095718B2. Unified Patents' litigation case list and annual reports also did not yield relevant results for this patent.

Proceedings overview

There are no AIA trial proceedings on file for US patent 10095718B2.

Strategic summary

As there are no PTAB proceedings on file for US patent 10095718B2, all claims (1-16) remain unchallenged and are considered sustained as originally granted. There is no estoppel landscape to consider, as no petitions have been filed. The absence of PTAB activity suggests that the patent has not yet been asserted in a way that would provoke an AIA trial, or potential challengers have not identified strong grounds for invalidation.

Recommended next steps

Given the lack of PTAB activity, a potential defendant facing assertion of US patent 10095718B2 would have a full range of prior art and statutory grounds available for an AIA trial challenge, should they choose to pursue one. The absence of previous challenges means there is no existing record of the patent's validity being tested in the PTAB.

Inventors

• J. Douglas Birdwell (University of Tennessee Research Foundation)
• Mark E. Dean (University of Tennessee Research Foundation)
• Catherine Schuman (University of Tennessee Research Foundation)

There is no information within the provided patent text to suggest any unusual patterns regarding the inventors' employment after the filing date.

Original assignee

The entity named on the issued patent is the University of Tennessee Research Foundation. Their primary line of business is to manage intellectual property generated by researchers at the University of Tennessee, including patenting, licensing, and commercialization activities. As a university research foundation, they typically do not ship products embodying claims directly but license their technology to operating companies. The University of Tennessee Research Foundation is currently operating.

Assignment timeline

There are no assignment records found for US patent 10095718B2 in the USPTO Patent Assignment Search database beyond the initial assignment to the University of Tennessee Research Foundation as the original assignee. The patent records indicate the application was filed by the University of Tennessee Research Foundation on October 14, 2014, and subsequently assigned to them on November 12, 2014. This suggests the University of Tennessee Research Foundation still owns the patent.

Timeline diagram

timeline
    title Ownership of US 10095718B2
    2014 : Application filed by UT Research Foundation
         : Assigned to UT Research Foundation
    2018 : Patent issued
    2026 : Currently owned by UT Research Foundation

NPE / troll-pattern signals

1. Shell-entity transfer — not present. The patent remains with the original assignee, a university research foundation, which is an operating entity in the context of IP management.
2. Known asserter in the chain — not present. The University of Tennessee Research Foundation is not identified as a known NPE.
3. Repeat correspondent across the chain — not present. There is only one recorded assignment event (from inventors to the University of Tennessee Research Foundation), and thus no recurrence of a correspondent attorney across a chain of transfers.
4. Cascading transfers — not present. There are no multiple consecutive assignments.
5. Pre-litigation transfer — not present. There is no litigation information available in the CAFC 2026 dockets as previously stated, and no transfers that would precede such litigation.
6. Bankruptcy fire-sale — not present. The University of Tennessee Research Foundation is an active entity.
7. Privateering — not present. No evidence of transfer to an NPE for assertion on behalf of an operating company.
8. Defensive aggregator (anti-NPE) — not present. The patent is not currently assigned to a defensive aggregator.

Verdict

Insufficient data. Based on the USPTO Patent Assignment Search (https://assignmentcenter.uspto.gov/patent/[10095718](/patent/10095718)), there are no recorded post-issuance assignments for US10095718B2. The only assignment recorded is the initial one from the inventors to the University of Tennessee Research Foundation on November 12, 2014, which is the original assignee. Therefore, there is no evidence of the patent being transferred to an NPE or any other entity after its initial assignment.

The University of Tennessee Research Foundation (UTRF) is a non-profit organization that promotes the commercialization of UT intellectual property, encourages UT research, and supports economic development. It is the single point of contact for companies interested in working with UT on research and commercialization projects. (https://utrf.tennessee.edu/)
The University of Tennessee Research Foundation (UTRF) is an organization that facilitates the commercialization of research discoveries and intellectual property developed by faculty, staff, and students within the University of Tennessee system. (https://en.wikipedia.org/wiki/University_of_Tennessee_Research_Foundation)
2014-11-12 Assigned to UNIVERSITY OF TENNESSEE RESEARCH FOUNDATION. (https://patents.google.com/patent/US10095718B2/en)

I cannot identify the most relevant prior art for US patent 10095718B2 as the "References Cited" section, which lists the prior art documents, is not explicitly provided in the patent text I have access to, nor could it be retrieved through direct search queries for this specific information. Without this list, I cannot provide the full citation, publication/filing date, brief description, or potential anticipation under 35 U.S.C. § 102 for each reference.

The obviousness of US patent 10095718B2 under 35 U.S.C. § 103 can be analyzed by considering combinations of prior art references explicitly mentioned or described within the patent's "Definitions" section. A person having ordinary skill in the art (POSITA) would have been motivated to combine these references to achieve predictable results, particularly given the recognized advantages of such combinations in the field of artificial neural networks. The patent itself identifies many components of its invention as building upon existing knowledge.

Obviousness Analysis of Independent Claims

Claim 1 (Method Claim)

Claim 1 describes a method for constructing a dynamic adaptive neural network array (DANNA) by defining an array of electronic components configurable as either neurons or synapses. It further specifies configuring a subset of these components into a neuroscience-inspired artificial neural network with a dynamic architecture, comprising a computational network and at least one affective network. An evolutionary optimization algorithm is then applied to dynamically modify the network structure and parameters.

Several combinations of prior art would render Claim 1 obvious:

1. Combination of Hardware Implementation of Neural Networks with Variable-Structure/Neuroevolutionary Algorithms:

   • Prior Art Elements: The patent explicitly mentions that the DANNA can be constructed from "primary neuron and synapse elements of one of a programmable logic array, application specific integrated circuit, VLSI component or other component" [cite: Definitions section]. The general field of "Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means" (G06N3/063) is a recognized classification. Furthermore, the patent describes "Networks with training algorithms that can change the architecture may be considered variable-structure" [cite: Definitions section]. Crucially, "Yao presents a thorough overview of algorithms that use evolutionary algorithms to train the weights of neural networks" [cite: Definitions section], and "Yao and Liu introduce an evolutionary system called EpNet for evolving the architecture and weights of feed-forward artificial neural networks" [cite: Definitions section]. Other neuroevolution methods like NEAT and CoSyNE are also acknowledged. [cite: Definitions section]
   • Motivation for Combination: A POSITA would be motivated to combine the known concept of hardware implementation of neural networks (e.g., using FPGAs or ASICs for neurons and synapses) with established neuroevolutionary algorithms for variable-structure networks. The patent itself highlights the advantages of evolutionary algorithms, stating they "do not depend on gradient information," "can be applied to any neural network architecture," and "always search for global optima, rather than local optima" [cite: Definitions section]. These advantages provide a clear motivation for using EAs to dynamically modify the structure and parameters of neural networks, including those implemented in hardware, to achieve optimal performance and adaptability. The goal of the machine learning community to produce networks with capabilities similar to biological systems further motivates exploring dynamic architectures. [cite: Definitions section]

2. Combination including Affective Systems for Parameter Regulation:

   • Prior Art Elements: The patent notes that "special-purpose emotion-related substructures and neurotransmitters can be incorporated into artificial neural networks," and "emotions such as fear or anger have been artificially simulated in the prior art individually but not collectively as to the collection of many emotion-related substructures." [cite: Definitions section] The patent also describes biological processes like "LTP" and "LTD" that play a role in learning. [cite: Definitions section]
   • Motivation for Combination: Given that individual "emotion-related substructures" or simulations of emotions were known in artificial neural networks, a POSITA would be motivated to integrate these into a larger, dynamically evolving neural network. The biological inspiration, which the patent frequently references, provides a strong rationale: "Biological neural networks are known to have many desirable characteristics... For these reasons and many others, it has been a goal of the machine learning community to produce networks with similar capabilities to biological central nervous systems, brains and, in particular to the human brain." [cite: Definitions section] If individual emotional responses can be simulated, it would be a logical next step to have a system (an "affective network") that regulates the parameters of the main computational network in a biologically inspired way (e.g., by adjusting neuron thresholds, analogous to neurotransmitters [cite: Definitions section]) to improve learning or adaptation, especially when the network structure itself is evolving.

Claim 14 (Apparatus Claim)

Claim 14 describes an apparatus for constructing a DANNA, comprising an array of addressable electronic components configurable as either neurons or synapses, configuration circuitry, and a control and optimizing device that applies evolutionary optimization algorithms to modify the network's structure and parameters, including a computational network and at least one affective network.

This claim is the apparatus counterpart to Claim 1, and the same arguments regarding prior art and motivation for combination apply:

• Prior Art Elements: The patent acknowledges that its elements can be "constructed from field programmable gate arrays" and mentions VLSI components and ASICs [cite: Definitions section]. This demonstrates the prior existence of hardware capable of implementing configurable neural elements. The functionality of configuration circuitry is inherent in FPGAs and ASICs, which are by definition "programmable" or "application-specific integrated circuits." The "control and optimizing device" embodies the computational engine required to run evolutionary algorithms, which, as discussed for Claim 1, were well-known for evolving neural network architectures and weights (Yao, Yao & Liu). [cite: Definitions section]
• Motivation for Combination: A POSITA, understanding the benefits of dynamically evolving neural network architectures and parameters through evolutionary algorithms (as taught by Yao and Liu), would find it obvious to implement such a system using readily available and suitable hardware components like FPGAs or custom ASICs. The motivation is to achieve the practical, high-performance realization of these adaptive networks. The explicit mention of FPGAs and ASICs as suitable for DANNA construction within the patent description suggests that their use for configurable neuromorphic elements was within the purview of a skilled artisan.

Claim 16 (Method Claim for Dynamic Adaptive Neural Network)

Claim 16 describes a method for operating a dynamic adaptive neural network, involving initializing a population of neural networks, calculating a fitness value for each, selecting parent networks based on fitness, applying crossover and mutation to generate a child population, and selecting the best-performing network.

This claim essentially outlines a standard neuroevolutionary algorithm as applied to a dynamic adaptive neural network:

• Prior Art Elements: The patent explicitly defines and describes all steps of this claim as known prior art within the context of evolutionary algorithms and neuroevolution:
  • "One or more relatively fit members of the population may be selected to reproduce." [cite: Definitions section]
  • "Evolutionary algorithms typically rely on a fitness function" [cite: Definitions section], and "Performance is a key metric and is a problem-specific issue." [cite: Definitions section]
  • "Roulette selection" and "tournament selection" are described as known selection algorithms. [cite: Definitions section]
  • "Crossover" and "mutation" are defined as established operations in evolutionary algorithms. [cite: Definitions section] "Yao and Liu introduce five mutation operations." [cite: Definitions section]
  • The overall process of "the combination of a genetic algorithm at a higher level and another algorithm... at a lower level" is discussed. [cite: Definitions section]
• Motivation for Combination: Claim 16 describes the fundamental steps of neuroevolution, which, as established by Yao (2008) and Yao and Liu (1997), were known methods for evolving both the weights and architecture of artificial neural networks. A POSITA would find it entirely obvious to apply these well-established evolutionary optimization techniques to any form of dynamic adaptive neural network, including the DANNA described in the patent. The motivation is to leverage the known advantages of EAs for neural network design, such as their ability to explore global optima and adapt network structures without requiring gradient information. [cite: Definitions section] The patent explicitly states that "a neuroevolution approach was selected for an embodiment of the present invention," further underscoring the prior art nature of these techniques. [cite: Definitions section]

In summary, the core components of US patent 10095718B2, including the use of reconfigurable hardware for neural elements, dynamic network architectures, evolutionary optimization for structure and parameter modification, and even the incorporation of emotion-related substructures (albeit individually), are discussed within the patent as existing or inspired by prior art and biological systems. A POSITA would have been motivated to combine these known elements to create more robust, adaptable, and biologically plausible artificial neural networks, particularly given the stated goals of the machine learning community and the recognized benefits of evolutionary algorithms for architectural exploration.