Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Drawings
The drawings are objected to as failing to comply with 37 CFR 1.84(p)(5) because they include the following reference character(s) not mentioned in the description: 225, 230, 235, and 240 (all in Fig. 2). Corrected drawing sheets in compliance with 37 CFR 1.121(d) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either "Replacement Sheet" or "New Sheet" pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
Specification
The disclosure is objected to because of the following informalities:
On pg. 7, the heading "BREIF DESCRIPTION OF THE DRAWINGS" contains a misspelling; it should read ‘BRIEF DESCRIPTION OF THE DRAWINGS’
On pg. 10, ln. 19 "analyisis" should be ‘analysis’
On pg. 5, ln. 18, "The second time period can overlaps with the first time period" contains a grammatical error; "overlaps" should be ‘overlap’
On pg. 2, lns. 23-24, "different functional subnetworks exit at different times" contains a typographical error; "exit" should be ‘exist’
On pg. 4, ln. 12, "includes any one the first aspect" is missing a word; it should read ‘includes any one of the first aspect’
Appropriate correction is required.
Claim Objections
Claims 27 and 35 are objected to because of the following informalities:
Claim 27 recites "determining that neural network device is functioning properly" which is missing the article ‘the’ before "neural network device."
Claim 35 recites "determining that second neural network is functioning properly" which is missing the article ‘the’ before "second neural network."
Appropriate correction is required.
Claim Rejections - 35 USC § 112(a) or 35 USC § 112, first paragraph
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Written Description
Claims 28-35 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
Per claim 28, a computer-implemented method is recited that comprises four topological characterization steps (receiving input, dividing activity into time bins, recording measures, and characterizing using topological methods) followed by a fifth step of “reconstructing at least some of functioning of the first neural network in a second neural network using the characterizations provided by the topological methods”. While the specification supports steps 1-4 (via process 400 of fig. 4), the specification fails to demonstrate possession of the reconstruction step. The specification states in high-level, aspirational terms that topological characterizations "can be used in the construction and/or reconstruction of neural networks" and that they "can provide general characteristics of the functional behavior of a desirable neural network device" (pg. 8 of specification), but nowhere describes a specific algorithm, procedure, or methodology for translating topological characterizations derived from functional activity into a reconstructed second neural network.
On pg. 8, starting at ln. 4 discloses, “Structural characterizations of neural network devices can be used, e.g., in the construction and/or reconstruction of neural networks. Reconstruction of a neural network can include, e.g., copying or mimicking at least some of the structure of a first neural network in a second neural network”. This describes structural reconstruction, not functional reconstruction using topological characterizations from functional activity as claimed.
On pg. 9, starting at ln. 20 discloses, “characterizations of the patterns of communication across multiple links in a neural network device can be used in the construction and/or reconstruction of neural networks. Reconstruction of a neural network can include, e.g., copying or mimicking at least some of the function of a first neural network in a second neural network”. This states the concept but provides no implementation details for functional reconstruction that is claimed.
On pg. 10, starting at ln. 2 discloses, “the characterizations provided by topological methods can provide general characteristics of the functional behavior of a desirable neural network device. This can be beneficial to, e.g., reduce training time or even provide a partially- or fully-functional neural network device out of the box”. This is an aspirational statement of benefit, not algorithmic disclosure.
On pg. 10, starting at ln. 6 discloses, “the topological characterizations may define a desired level of “structuring” or “ordering” of the information flow within a functioning neural network device”. This describes a possible use but not how to achieve it.
On pg. 10, starting at ln. 8 discloses: 'functional sub-networks can be assembled like components, e.g., by adding structural links between different functional sub-networks to achieve desired processing results”. This describes a general concept without specific algorithm.
Therefore, there is no algorithm, flowchart, step-by-step process, mathematical formulation, or working example is provided for the reconstruction step of claim 28. The specification provides only high-level statements that topological characterizations 'can be used' for reconstruction, without disclosing the specific algorithm or steps for performing the reconstruction. The specification describes WHAT to do (use topological characterizations for reconstruction) but not HOW to do it. Under Vasudevan Software, Inc. v. MicroStrategy, Inc., 782 F.3d 671 (Fed. Cir. 2015), the specification for a computer-implemented method must disclose the algorithm for performing each claimed function. The specification fails to disclose any algorithm for translating topological characterizations of functional activity into a reconstructed second neural network. The disclosure describes a desired result without demonstrating possession of a specific means to achieve it. The claim covers ANY method of reconstructing ANY functioning of a first neural network in ANY second neural network using ANY topological characterizations. The specification describes the concept of reconstruction at a high level but discloses no specific embodiment of functional reconstruction using topological methods. The gap between the described concept and the full claim scope is not bridged by the disclosure.
Claims 29–35 depend from claim 28 and inherit the written description deficiency of the 'reconstructing' limitation. Each adds a further limitation, but none cures the underlying deficiency.
Enablement
Claims 28-35 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the enablement requirement.
Claim 28 recites the step of, “reconstructing at least some of functioning of the first neural network in a second neural network using the characterizations provided by the topological methods”. This step covers any method of reconstructing any functioning of a first neural network in any second neural network using any topological characterization. This encompasses an extremely wide range of potential implementations, none of which are specifically described.
To satisfy the enable requirement of 35 U.S.C §112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, the specification must teach those skilled in the art how to make and use the full scope of the claimed invention without "undue experimentation" (see MPEP 2161.01(III)).
A showing of undue experimentation is given here for the recited claim limitations involving generating one or a plurality affinity vectors by applying one or more neural networks, based on some of the factors cited in MPEP §2164.01(a) as pertaining to In re Wands.
Breadth of Claims: Very broad - Claim 28 covers any method of reconstructing any functioning of a first neural network in any second neural network using any topological characterization. This encompasses an extremely wide range of potential implementations, none of which are specifically described.
Nature of Invention: Complex and interdisciplinary - The reconstruction step requires translating abstract topological parameters (simplex counts, Betti numbers, homological dimensions) into concrete neural network structure and weights, which is a fundamentally complex inverse problem spanning algebraic topology and machine learning.
State of Prior Art: Nascent for reconstruction application - While neural networks and algebraic topology are individually well-established, the application of topological characterizations from functional activity to neural network reconstruction was nascent at the effective filing data of the patent application.
Level of Ordinary Skill: arguably PhD-level interdisciplinary expertise - A PHOSITA would arguably require PhD-level expertise in both algebraic topology and neural network architectures. Even at this level, the reconstruction step presents a novel research problem without established methodology.
Predictability: Low for reconstruction step - While topological characterization (forward direction) is mathematically deterministic, the inverse mapping from topological parameters to neural network functioning is inherently underdetermined - many different networks can produce the same topological characteristics, making the reconstruction problem ill-posed and unpredictable.
Direction/Guidance Provided: Minimal for reconstruction - The specification provides detailed guidance for topological characterization (steps 1-4, process 400 of fig. 4), but only high-level aspirational statements for reconstruction: ‘can be used’, ‘may define’, ‘can provide general characteristics’. No specific algorithm, procedure, optimization objective, or methodology is provided for the reconstruction step.
Working Examples: None for reconstruction - The specification contains no working examples of functional reconstruction using topological methods. Process 100 (fig. 1) describes structural characterization and process 400 (fig. 4) describes functional characterization and distinguishing, but no process describes reconstruction. No experimental data, simulation results, or concrete reconstruction examples are provided.
Quantity of Experimentation: Undue - A PHOSITA would need to: (a) determine which topological parameters are meaningful targets for reconstruction, (b) develop a novel methodology for the inverse mapping from topological parameters to network structure/weights/connectivity, (c) address the fundamental ill-posedness of the inverse problem (many-to-one mapping), (d) validate that the reconstructed network actually reproduces the desired functioning. This constitutes research-level experimentation that goes well beyond routine work in the art. Therefore, there is undue experimentation in the independent claims and they fail to comply with the enablement requirement.
Claims 29–35 inherit the enablement deficiency of claim 28. No dependent claim limitation provides additional guidance that would cure the lack of enablement for the reconstruction step.
Claim Rejections - 35 USC § 112(b) or 35 USC § 112, second paragraph
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 21-35 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 21 recites “the neural network device” in the recording steps (e.g., “recording, for each of the first time bins, a first measure of the functional activity in the neural network device during that first time bin”). However, claim 21 introduces only “a neural network” in the receiving step. The terms “neural network” and “neural network device” are not identical. There is insufficient antecedent basis for the limitation “the neural network device” in the claim. For purposes of examination, “the neural network device” is interpreted under BRI to refer to the same “neural network” introduced in the receiving step of claim 21.
Claim 21 recites “the recorded second measure of the functional activity” in the second characterizing step. However, the second recording step introduces only “a measure of the functional activity”, not “a second measure”. There is insufficient antecedent basis for the limitation “the recorded second measure” in the claim. For purposes of examination, “the recorded second measure” is interpreted under BRI to refer to the measure of functional activity recorded during the second time bins (i.e., the “measure” introduced in the second recording step).
Claim 21 recites a step of “recording, for each of the second time bins, a measure of the functional activity in the neural network device during that second time bin that is responsive to the first input”. However, the immediately preceding step recites “dividing functional activity in the neural network that is responsive to the second input into second time bins”. The dividing step establishes that the second time bins pertain to activity responsive to the second input, but the recording step states the activity is “responsive to the first input”. This internal inconsistency renders the claim indefinite because a person of ordinary skill in the art cannot determine with reasonable certainty whether the recording step is intended to record activity responsive to the first input or the second input. For purposes of examination, this limitation is interpreted under BRI as recording a measure of the functional activity responsive to the second input (not the first input).
Claim 21 recites “the functional response of the neural network to the first input” and “the functional response of the neural network to the second input” in the distinguishing step. However, the claim body does not previously introduce “a functional response”. The claim recites “functional activity” and “a measure of the functional activity”, but “functional response” is a different term. There is insufficient antecedent basis for the limitation “the functional response” in the claim. For purposes of examination, “the functional response” is interpreted under BRI to refer to the overall functional behavior of the neural network in response to a given input, as characterized by the topological parameters determined in the preceding steps.
Claims 22-27 are rejected as being dependent upon a rejected base claim.
Claim 24 recites “the time bins”. However, claim 21 (from which claim 24 depends) introduces “first time bins” and “second time bins”, not “time bins” generically. It is unclear whether “the time bins” in claim 24 refers to the first time bins, the second time bins, or both. There is insufficient antecedent basis for the limitation “the time bins“ in the claim. For purposes of examination, “the time bins” is interpreted under BRI to refer to both the first time bins and the second time bins introduced in claim 21.
Claim 25 recites “the measure of the functional activity”. However, claim 21 (from which claim 25 depends) introduces both “a first measure of the functional activity” and “a measure of the functional activity” (in the second recording step). The use of “the measure” (singular, without “first” or “second”) is ambiguous as to which measure is being further limited. There is insufficient antecedent basis for the limitation “the measure of the functional activity” as a singular definite reference when two measures have been introduced. For purposes of examination, “the measure of the functional activity” is interpreted under BRI to refer to both the first measure and the second measure of functional activity recorded in claim 21.
Claim 27 recites “determining that neural network device is functioning properly or trained”, the phrase “that neural network device” lacks the definite article “the” and additionally, the term “neural network device” lacks antecedent basis in claim 21, which introduces only “a neural network”. The missing article creates grammatical ambiguity and the mismatch between “neural network device” and “neural network” creates further uncertainty. There is insufficient antecedent basis for this limitation in the claim. For purposes of examination, “that neural network device”" is interpreted under BRI as “the neural network” introduced in claim 21.
Claim 28 recites "the neural network" in the dividing step, “dividing functional activity in the neural network that is responsive to the input into time bins”. However, claim 28 introduces “a first neural network” in the receiving step and subsequently introduces "a second neural network" in the reconstructing step. The proper reference should be "the first neural network." Because the claim introduces two neural networks (“a first neural network” and “a second neural network”), the reference to “the neural network” (without “first”) is ambiguous and lacks proper antecedent basis. For purposes of examination, “the neural network” in the dividing step is interpreted under BRI to refer to “the first neural network” introduced in the receiving step.
Claim 28 recites “the neural network device” in the recording step. However, claim 28 introduces “a first neural network”, not “a neural network device”. The terms “neural network” and “neural network device” are not identical. There is insufficient antecedent basis for the limitation “the neural network device” in the claim. For purposes of examination, “the neural network device” is interpreted under BRI to refer to the same “first neural network” introduced in the receiving step of claim 28.
Claims 29-35 are rejected as being dependent upon a rejected base claim.
Claim 29 recites “the second neural network is simpler than the first neural network”. The term "simpler" is a relative term which renders the claim indefinite. The term "simpler" is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree of simplicity, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. While the specification states that “a simpler second neural network can recreate a portion of the functioning of a more complex first neural network” (pg. 8, lns 7-8 of specification), it does not define what constitutes “simpler”, e.g., fewer nodes, fewer layers, fewer connections, reduced computational complexity, or some other metric. The metes and bounds of the claim cannot be determined. For purposes of examination, "simpler" is interpreted under BRI to mean that the second neural network has fewer parameters, fewer nodes, fewer layers, fewer connections, or reduced computational complexity compared to the first neural network.
Claim 35 recites “determining that second neural network is functioning properly or trained”. The phrase “that second neural network” lacks the definite article “the”, creating grammatical ambiguity. The proper phrasing should be “determining that the second neural network is functioning properly or trained”. There is insufficient antecedent basis for this limitation as written. For purposes of examination, “that second neural network” is interpreted under BRI as “the second neural network” introduced in claim 28.
Claim 35 recites “the input pattern”. However, claim 28 (from which claim 35 depends) introduces “an input”, not “an input pattern”. The terms “input” and “input pattern” are not identical. There is insufficient antecedent basis for the limitation “the input pattern” in the claim. For purposes of examination, “the input pattern” is interpreted under BRI to refer to the “input” introduced in claim 28.
Appropriate correction is required.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 21-27 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The analysis of the claims follows the 2019 Revised Patent Subject Matter Eligibility Guidelines ("2019 PEG").
Claim 21
Step 1: Claim 21 recites "A computer-implemented method, comprising..." which is directed to the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 21 recites following limitations that are mathematical concepts:
“dividing functional activity in the neural network that is responsive to the first input into first time bins” and “recording, for each of the first time bins, a first measure of the functional activity in the neural network device during that first time bin that is responsive to the first input”…these steps involve organizing and measuring data in preparation for mathematical computation, which are data-gathering steps but considered necessary and part of the mathematical computation/analysis that follows.
“characterizing one or more parameters of the recorded first measure of the functional activity using topological methods”…topological methods (including computation of directed flag complexes, Betti numbers, Euler characteristics, and simplex counts) are mathematical concepts involving mathematical calculations applied to data. See MPEP 2106.04(a)(2)(I).
The same mathematical limitations are repeated for the second input…“dividing functional activity in the neural network that is responsive to the second input into second time bins; recording, for each of the second time bins, a measure of the functional activity in the neural network device during that second time bin that is responsive to the first input; characterizing one or more parameters of the recorded second measure of the functional activity using topological methods”.
“distinguishing the functional response of the neural network to the first input from the functional response of the neural network to the second input based on the characterized topological parameters”…this is a mathematical comparison of two sets of topological parameters (e.g., comparing Betti numbers, simplex counts, or Euler characteristics between two functional responses), which is a mathematical calculation. See MPEP 2106.04(a)(2)(I). The additional elements are "A computer-implemented method", "receiving, at a neural network, a first input", "receiving, at the neural network, a second input", and the recitation of a "neural network" / "neural network device".
Step 2A, Prong 2: The judicial exception is not integrated into a practical application. The additional elements, considered individually and as an ordered combination, do not integrate the judicial exception into a practical application.
The additional element “A computer-implemented method” is recited at a high level of generality and recites mere instructions to implement the abstract idea on a computer. Mere recitation that a judicial exception is to be performed using generic computer equipment in its ordinary capacity cannot meaningfully integrate the judicial exception into a practical application. See MPEP 2106.05(f).
The additional elements "receiving, at a neural network, a first input" and "receiving, at the neural network, a second input" constitute insignificant extra-solution activity in the form of data gathering (pre-solution activity). Receiving inputs at a neural network is mere data gathering necessary for the mathematical analysis to follow. See MPEP 2106.05(g). The recitation of a "neural network" / "neural network device" generally links the use of the judicial exception to the particular technological environment of neural networks without meaningfully limiting how the judicial exception is applied. See MPEP 2106.05(h).
Considering the claim as a whole, the ordered combination of additional elements, receiving two inputs at a neural network, performing topological mathematical analysis on functional activity data recorded during time bins, and producing a mathematical comparison of the topological parameters, does not transform the mathematical analysis into a practical application. The ordered combination follows a conventional pattern of data input, mathematical computation, and output of analytical results. The specific sequence of steps does not impose meaningful limits on the judicial exception beyond specifying the particular mathematical technique (topological methods) applied to a particular type of data (neural network functional activity). See McRO, Inc. v. Bandai Namco Games Am. Inc., 837 F.3d 1299, 1315, 120 USPQ2d 1091, 1102-03 (Fed. Cir. 2016) (distinguishing claims reciting specific rules that improve technology from claims that merely use mathematical analysis in a particular field).
With respect to the Desjardins improvement analysis (see MPEP 2106.05(a); Ex Parte Desjardins, Appeal No. 2024-000567 (PTAB September 26, 2025, Appeals Review Panel Decision)), the specification describes that topological methods can be used for construction, reconstruction, testing, and analysis of neural networks, and notes that in complex neural networks, “human oversight is lost and the function of subnetworks inscrutable” (pg. 1). The specification describes two categories of potential improvements: (1) improvements to neural network construction and reconstruction using topological characterizations, and (2) improvements to neural network testing and analysis by enabling characterization of neural network functional activity that would otherwise be inscrutable.
Regarding the first category, claim 21 does not recite any steps directed to constructing, reconstructing, or modifying a neural network. Rather, claim 21 is directed to analyzing neural network responses and producing a distinction between them, a purely analytical result. The improvements described in the specification relating to neural network construction and reconstruction are not reflected in claim 21.
Regarding the second category, the specification describes that topological methods can provide characterizations of neural network functional activity useful for testing and analysis. However, claim 21 recites only the mathematical process of topological characterization and comparison of functional responses, without reciting any steps that apply the results of the analysis to improve, test, validate, or modify the neural network. The claim produces a mathematical distinction between two sets of topological parameters but does not recite using that distinction to determine whether the neural network is functioning properly, to guide training, or to otherwise improve the neural network's operation. A PHOSITA in the art would recognize that claim 21, as a whole, recites a mathematical analysis process applied to neural network data, producing an analytical comparison, rather than reflecting the testing or analysis improvements described in the specification. The claim does not include the components or steps that provide the improvement identified in the specification. See MPEP 2106.05(a); Ex Parte Desjardins, Appeal No. 2024-000567 (PTAB September 26, 2025, Appeals Review Panel Decision).
Step 2B: The claim does not include additional elements that amount to significantly more than the judicial exception. The additional element "A computer-implemented method" amounts to using a computer to perform mathematical calculations and data analysis, which is well-understood, routine, and conventional. See Alice Corp. Pty. Ltd. v. CLS Bank Int'l, 573 U.S. 208, 225-26 (2014); performing repetitive calculations (Flook, 437 U.S. at 594, 198 USPQ at 199; Bancorp Services v. Sun Life, 687 F.3d 1266, 1278, 103 USPQ2d 1425, 1433 (Fed. Cir. 2012)); electronic recordkeeping (Alice Corp., 573 U.S. at 225, 110 USPQ2d at 1984; Ultramercial, 772 F.3d at 716, 112 USPQ2d at 1755). See MPEP 2106.05(d)(II).
The additional elements "receiving, at a neural network, a first input" and "receiving, at the neural network, a second input" were found to be insignificant extra-solution activity (data gathering) at Step 2A, Prong 2. Re-evaluating under Step 2B, these data-gathering steps of providing input data to a computational system for processing amount to mere data gathering, which is well-understood, routine, and conventional. See OIP Techs., Inc. v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (presenting offers to potential customers and gathering statistics generated based on a testing algorithm—i.e., data gathering for a computational process); storing and retrieving information in memory, Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015); OIP Techs., 788 F.3d at 1363, 115 USPQ2d at 1092-93. See MPEP 2106.05(d)(II).
The recitation of a "neural network" / "neural network device" as a field-of-use limitation does not add significantly more to the abstract idea itself. See MPEP 2106.05(h). The ordered combination of the additional elements does not amount to significantly more than the judicial exception. While the claim recites a specific sequence of mathematical operations (receiving inputs, recording functional activity in time bins, characterizing using topological methods, and distinguishing responses), these steps represent a mathematical analysis process applied in the technological environment of neural networks. The combination does not interact in an unconventional way that would transform the mathematical analysis into significantly more than the abstract idea.
Accordingly, claim 21 is patent ineligible under 35 U.S.C. 101.
Claim 22
Step 1: Claim 22 depends from claim 21 and recites a method, which falls within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 22 additionally recites:
“the topological methods comprise determining associated directed flag complexes”…Directed flag complexes are oriented simplicial complexes that encode the connectivity and direction of the underlying directed graph, determining them involves computation of simplicial complexes from adjacency matrices, which is a mathematical concept. See MPEP 2106.04(a)(2)(I).
Step 2A, Prong 2 & Step 2B: No additional elements introduced, analysis from parent claim maintained.
Claim 23
Step 1: Claim 23 depends from claim 21 and recites a method, which falls within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 23 additionally recites:
“the functional activity in the neural network that is responsive to the first and second inputs includes signal transmission along edges of the neural network”….This limitation further describes the type of data being measured and analyzed, specifying that the functional activity data comprises signal transmissions along edges. This is a characterization of the data subject to the mathematical analysis. See MPEP 2106.04(a)(2).
Step 2A, Prong 2 & Step 2B: No additional elements introduced, analysis from parent claim maintained. The additional limitation merely specifies the type of data being analyzed (signal transmission along edges), which characterizes the data subject to the mathematical analysis.
Claim 24
Step 1: Claim 24 depends from claim 21 and recites a method, which falls within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 24 additionally recites:
“the duration of the time bins is constant”…This is a mathematical constraint on the temporal parameters used in the data organization step, specifying a constant duration for time bins is a mathematical relationship governing how data is partitioned. See MPEP 2106.04(a)(2)(I).
Step 2A, Prong 2 & Step 2B: No additional elements introduced, analysis from parent claim maintained.
Claim 25
Step 1: Claim 25 depends from claim 21 and recites a method, which falls within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 25 additionally recites:
”the measure of the functional activity is a functional connectivity matrix”… A functional connectivity matrix is a mathematical representation, e.g., a binary matrix where active and inactive edges are denoted that captures the activity in the neural network. Specifying the measure as a functional connectivity matrix is specifying a mathematical data structure used in the mathematical analysis. See MPEP 2106.04(a)(2)(I).
Step 2A, Prong 2 & Step 2B: No additional elements introduced, analysis from parent claim maintained.
Claim 26
Step 1: Claim 26 depends from claim 21 and recites a method, which falls within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 26 additionally recites:
“the first input and the second input are known inputs”….This limitation constrains the type of data used in the analysis to known (as opposed to unknown) inputs. Specifying the inputs as known inputs is a data characterization that further describes the conditions under which the mathematical analysis is performed, and does not alter the abstract nature of the claim. See MPEP 2106.04(a)(2).
Step 2A, Prong 2 & Step 2B: No additional elements introduced, analysis from parent claim maintained.
Claim 27
Step 1: Claim 27 depends from claim 21 and recites a method, which falls within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 27 additionally recites:
“determining that neural network device is functioning properly or trained based on the distinguishing of the functional response to the first input from the functional response to the second input”…This step of making a determination about whether the neural network is functioning properly or is trained, based on the results of the distinguishing step, is a mental process involving evaluation, judgment, and opinion based on the results of the topological analysis. A person could mentally evaluate whether a distinction between two functional responses indicates proper functioning or successful training. See MPEP 2106.04(a)(2)(III).
Step 2A, Prong 2: The judicial exception is not integrated into a practical application. Claim 27 does not introduce any new additional elements beyond those in claim 21. The additional limitation of determining whether the neural network is functioning properly or trained is itself part of the abstract idea (a mental evaluation based on mathematical analysis results).
With respect to the Desjardins improvement analysis (See Ex Parte Desjardins, Appeal No. 2024-000567 (PTAB September 26, 2025, Appeals Review Panel Decision) (precedential); MPEP 2106.05(a)), the specification describes that distinguishing functional responses of a neural network using topological methods can indicate whether the neural network is "functioning properly" (during testing) and can provide "an indication that training is complete" (during training). See pg. 18 of specification. These descriptions relate to the utility of the mathematical analysis results, not to a technological improvement in how the neural network itself operates. Claim 27 recites using the mathematical comparison result to make a determination about the neural network's status, but does not recite any steps that modify, adjust, or improve the neural network based on that determination. The claim produces a determination (an evaluative output) rather than effecting a technological change. Unlike Ex Parte Desjardins, where the claims recited specific steps that adjusted model parameters to protect prior task knowledge (thereby improving how the ML model itself operated), claim 27 merely adds a mental evaluation step on top of the mathematical analysis of claim 21. A person of ordinary skill in the art would recognize that claim 27 recites the use of mathematical analysis results to form a judgment, not a technological improvement to neural network technology.
For the same reasons as discussed in the Step 2A, Prong 2 analysis of claim 21, the additional elements of the parent claim do not integrate the judicial exception into a practical application.
Step 2B: The claim does not include additional elements that amount to significantly more than the judicial exception. No new additional elements are introduced. For the same reasons as discussed in the Step 2B analysis of claim 21, the claim remains patent ineligible.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 21-27 are rejected under 35 USC 103 as being unpatentable over The topology of the directed clique complex as a network invariant to Masulli et al. (hereinafter Masulli) in view of Clique topology reveals intrinsic geometric structure in neural correlations to Giusti et al. (hereinafter Giusti).
Per claim 21, Masulli discloses A computer-implemented method (Abstract…algebro-topological invariants of directed networks implemented on a computer, "We introduce new algebro-topological invariants of directed networks, based on the topological construction of the directed clique complex…Two different cases illustrate the application of the Euler characteristic. We investigate how the evolution of a Boolean recurrent artificial neural network is influenced by its topology in a dynamics involving pruning and strengthening of the connections, and to show that the topological features of the directed clique complex influence the dynamical evolution of the network. The second application considers the directed clique complex in a broader framework, to define an invariant of directed networks, the network degree invariant, which is constructed by computing the topological invariant on a sequence of subnetworks filtered by the minimum in- or out-degree of the nodes"), comprising:
receiving, at a neural network, a first input (pg. 3…input nodes of the recurrent artificial neural network receive a first input pattern, "We considered a directed graph representing a simplified model of feedforward neural network with convergent/divergent layered structure with few embedded recurrent connections. In this model, the nodes represent individual neurons and the connections between them are oriented edges with a weight given by the connection strength…The nodes of the input layer are activated at regular time intervals");
dividing functional activity in the neural network that is responsive to the first input into first time bins (pg. 3…network activity is sampled at successive discrete time steps, e.g., first time bins, "We have computed the Euler characteristic and its variation during the evolution of such networks, both for the entirety of the nodes in the network and for the sub-network induced by the nodes that are active at each time step in order to detect how the structure changes as the network evolves");
recording, for each of the first time bins, a first measure of the functional activity in the neural network device during that first time bin that is responsive to the first input (Section p2…active-node set at each time step of the network evolution is recorded, "We have computed the Euler characteristics and its variations during the evolution of such networks…for the sub-network induced by the nodes that are active at each time step");
characterizing one or more parameters of the recorded first measure of the functional activity using topological methods (Abstract…"…based on the topological construction of the directed clique complex"; pg. 2…the Euler characteristic and Betti numbers of the directed clique complex of the active subgraph are computed as topological parameters, "In the current study we introduce a mathematical object, called directed clique complex, encoding the link structure of networks in which the edges (or links) have a given orientation. This object is a simplicial complex that can be studied with the techniques of algebraic topology to obtain invariants such as the Euler characteristic and the Betti numbers"); and
receiving, at the neural network, a second input; dividing functional activity in the neural network that is responsive to the second input into second time bins; recording, for each of the second time bins, a measure of the functional activity in the neural network device during that second time bin that is responsive to the first input; characterizing one or more parameters of the recorded second measure of the functional activity using topological methods (Abstract and Methods…the same procedure is repeated on a second input stimulus so that the evolution of the network under different input patterns can be compared, "the topological features of the directed clique complex influence the dynamical evolution of the network").
Masulli further teaches comparing the functional activity of a neural network via invariants of an associated simplicial/clique complex (Abstract…"The second application considers the directed clique complex in a broader framework, to define an invariant of directed networks, the network degree invariant, which is constructed by computing the topological invariant on a sequence of subnetworks filtered by the minimum in- or out-degree of the nodes"; pg. 7…the monotonicity and comparison of filtrations derived from the clique complex can distinguish the responses of different network topologies, "the monotonicity and the comparison between the in- and the out-degree filtrations can be used to distinguish the different types of networks").
Masulli does not expressly map the topological characterization to two distinct input-stimulus responses being distinguished by Betti curves of the neural correlation/connectivity matrix, but with Giusti does teach: distinguishing the functional response of the neural network to the first input from the functional response of the neural network to the second input based on the characterized topological parameters (Giusti: pg. 4 and fig. 1(e)…Betti curves computed from the order complex of the neural activity correlation matrix reliably distinguish random structure from geometric structure and therefore reliably discriminate between distinct neural activity patterns, "the Betti curves …display a characteristic unimodal shape… makes it possible to robustly distinguish random from non-random structure"; pg. 9, fig. 4…example of Betti curves distinguish among different behavioral/stimulus conditions such as spatial navigation, wheel running and REM sleep "Geometric organization in hippocampus during non-spatial behaviors").
Masulli and Giusti are analogous art because they are from both within the same field of endeavor, specifically the topological (algebraic-topology/persistent-homology) analysis of neural network activity. They address the same problem solving area of characterizing and comparing the functional activity of a neural network via invariants of an associated simplicial (clique) complex. Masulli builds its formalism directly on "Betti curves" introduced by Giusti (Masulli: pg. 5, "A distinct topological invariant defined for non-directed networks, referred to the Betti curves, was recently proposed by Giusti et al (2015)"), which is the core subject of Giusti (Giusti: Abstract, "a novel approach to matrix analysis, called clique topology").
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to apply Giusti's clique-topology method of distinguishing different neural-activity patterns via Betti-curve invariants of a correlation matrix to the Masulli framework of computing directed-clique-complex invariants on a recurrent neural network responding to successive input stimuli, because Masulli itself directs the PHOSITA to Giusti for the distinguishing-invariant and because Giusti demonstrates the concrete result (discrimination of behavioral/stimulus-driven neural states) that a PHOSITA would want from Masulli's formalism.
The suggestion/motivation for doing so would have been Masulli's explicit citation of Giusti as a complementary topological invariant for the same neural-network class (Masulli: pg. 5) combined with Giusti's demonstration that Betti-curve invariants of correlation matrices reliably separate different neural-activity regimes (Giusti: figs. 3 - 4), yielding the predictable result of distinguishing two input-driven response regimes of the neural network by comparing their topological parameters.
Per claim 22, Masulli combined with Giusti discloses claim 21. Masulli further teaches wherein the topological methods comprise determining associated directed flag complexes (Masulli: Abstract…the directed clique complex (equivalent to the directed flag complex) is computed from the directed neural-network graph, "the topological construction of the directed clique complex"; pg. 6…" a directed clique (or a simplex in our complex) is a fully-connected directed sub-network (Figure 3) this means that the nodes are ordered and there is one source and one sink in the sub-network, and the presence of the directed clique in the network means that the former is connected to the latter in all the possible ways within the sub-network").
Per claim 23, Masulli combined with Giusti discloses claim 21. Masulli further teaches wherein the functional activity in the neural network that is responsive to the first and second inputs includes signal transmission along edges of the neural network (Masulli: pg. 1…Active nodes activation propagates along directed edges according to a rule, "Active nodes are those nodes whose state depend on a set of precise rules that depend on network topology and dynamics").
Per claim 24, Masulli combined with Giusti discloses claim 21. Masulli further teaches wherein the duration of the time bins is constant (Masulli: pg. 3…input-layer nodes are activated at regular (fixed-duration) time intervals and evolution proceeds in uniform discrete time steps, "The nodes of the input layer are activated at regular time intervals").
Per claim 25, Masulli combined with Giusti discloses claim 21. Giusti further teaches: wherein the measure of the functional activity is a functional connectivity matrix (Giusti: fig. 3…a pairwise correlation (functional connectivity) matrix is computed from spike trains and used as the measure of functional activity, "Betti curves of the pairwise correlation matrix for the activity of N = 88 place cells in hippocampus"). The rationale to combine Giusti with Masulli is the same as the parent claim.
Per claim 26, Masulli combined with Giusti discloses claim 21. Masulli further teaches wherein the first input and the second input are known inputs (Masulli: pg. 3…the input patterns presented to the network are predetermined/known by the experimenter, "The nodes of the input layer are activated at regular time intervals").
Per claim 27, Masulli combined with Giusti discloses claim 21. Masulli combined with Giusti further teaches wherein the method further comprises determining that neural network device is functioning properly or trained based on the distinguishing of the functional response to the first input from the functional response to the second input (Masulli: Abstract…the Euler-characteristic evolution is used to assess that the trained/evolved network has reached a stable functional regime, "topological features of the directed clique complex influence the dynamical evolution of the network"; Giusti: Abstract…the topological signature confirms proper (geometric) organization of the neural activity). The rationale to combine Giusti with Masulli is the same as the parent claim.
Claims 28 and 31-35 are rejected under 35 USC 103 as being unpatentable over Masulli in view of Distilling the Knowledge in a Neural Network to Hinton et al. (hereinafter Hinton).
Per claim 28, Masulli discloses A computer-implemented method (Abstract…algebro-topological invariants of directed networks implemented on a computer, "We introduce new algebro-topological invariants of directed networks, based on the topological construction of the directed clique complex…Two different cases illustrate the application of the Euler characteristic. We investigate how the evolution of a Boolean recurrent artificial neural network is influenced by its topology in a dynamics involving pruning and strengthening of the connections, and to show that the topological features of the directed clique complex influence the dynamical evolution of the network. The second application considers the directed clique complex in a broader framework, to define an invariant of directed networks, the network degree invariant, which is constructed by computing the topological invariant on a sequence of subnetworks filtered by the minimum in- or out-degree of the nodes"), comprising:
receiving, at a first neural network, an input (pg. 3…input nodes of the recurrent artificial neural network receive a first input pattern, "We considered a directed graph representing a simplified model of feedforward neural network with convergent/divergent layered structure with few embedded recurrent connections. In this model, the nodes represent individual neurons and the connections between them are oriented edges with a weight given by the connection strength…The nodes of the input layer are activated at regular time intervals");
dividing functional activity in the neural network that is responsive to the input into time bins (pg. 3…network activity is sampled at successive discrete time steps, e.g., time bins, "We have computed the Euler characteristic and its variation during the evolution of such networks, both for the entirety of the nodes in the network and for the sub-network induced by the nodes that are active at each time step in order to detect how the structure changes as the network evolves");
recording, for each of the time bins, a measure of the functional activity in the neural network device during that time bin that is responsive to the input (Section p2…active-node set at each time step of the network evolution is recorded, "We have computed the Euler characteristics and its variations during the evolution of such networks…for the sub-network induced by the nodes that are active at each time step");
characterizing one or more parameters of the recorded measure of the functional activity using topological methods (Abstract…"…based on the topological construction of the directed clique complex"; pg. 2…the Euler characteristic and Betti numbers of the directed clique complex of the active subgraph are computed as topological parameters, "In the current study we introduce a mathematical object, called directed clique complex, encoding the link structure of networks in which the edges (or links) have a given orientation. This object is a simplicial complex that can be studied with the techniques of algebraic topology to obtain invariants such as the Euler characteristic and the Betti numbers").
Masulli does not expressly disclose, but with Hinton does teach: reconstructing at least some of functioning of the first neural network in a second neural network using the characterizations provided by the topological methods (Hinton: Sections 1…the behavior of a first (cumbersome) neural network is distilled/reconstructed in a second (smaller) neural network by using soft-target characterizations produced by the first network, "we can then use a different kind of training, which we call ”distillation” to transfer the knowledge from the cumbersome model to a small model"; Section 2…the distilled model reproduces the functional input-output mapping of the first model).
Masulli and Hinton are analogous art because they are from both within the same field of endeavor, specifically the functional characterization and transfer of neural-network behavior using compact summary representations. They address the same problem solving area of compactly representing the functioning of a first neural network so that it can be reproduced or analyzed by another model. Masulli teaches extracting a compact topological summary, e.g., Euler characteristic / Betti numbers of the directed clique complex, of the functional activity of a first neural network (Masulli: Abstract), and Hinton teaches using a compact summary of a first neural network's activity as the supervisory signal for constructing a second neural network that replicates the first's function (Hinton: Section 2).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to use the topological characterization obtained by Masulli as the compact summary of the first network's functioning, and then to follow Hinton's distillation procedure to reconstruct that functioning in a second neural network, because Hinton generally teaches that any compact functional characterization of a first network can be used as a transfer signal to train or configure a smaller/second network that mimics the first.
The suggestion/motivation for doing so would have been Hinton's explicit teaching that compressing a complex model into a simpler one is desirable for deployment (Hinton: Abstract, "making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive"), combined with Masulli's demonstration that the topological invariants of the directed clique complex capture the functional behavior of a recurrent neural network (Masulli: Abstract…”We investigate how the evolution of a Boolean recurrent artificial neural network is influenced by its topology in a dynamics involving pruning and strengthening of the connections, and to show that the topological features of the directed clique complex influence the dynamical evolution of the network”), yielding the predictable result of a second neural network whose functioning is reconstructed from the topological characterization of the first.
Per claim 31, Masulli combined with Hinton discloses claim 28. Masulli further teaches wherein the topological methods comprise determining associated directed flag complexes (Masulli: Abstract…the directed clique complex (equivalent to the directed flag complex) is computed from the directed neural-network graph, "the topological construction of the directed clique complex"; pg. 6…" a directed clique (or a simplex in our complex) is a fully-connected directed sub-network (Figure 3) this means that the nodes are ordered and there is one source and one sink in the sub-network, and the presence of the directed clique in the network means that the former is connected to the latter in all the possible ways within the sub-network").
Per claim 32, Masulli combined with Hinton discloses claim 28. Masulli further teaches wherein the functional activity in the neural network that is responsive to the input includes signal transmission along edges of the neural network (Masulli: pg. 1…Active nodes activation propagates along directed edges according to a rule, "Active nodes are those nodes whose state depend on a set of precise rules that depend on network topology and dynamics").
Per claim 33, Masulli combined with Hinton discloses claim 28. Masulli further teaches wherein the duration of the time bins is constant (Masulli: pg. 3…input-layer nodes are activated at regular (fixed-duration) time intervals and evolution proceeds in uniform discrete time steps, "The nodes of the input layer are activated at regular time intervals").
Per claim 34, Masulli combined with Hinton discloses claim 28. Hinton further teaches wherein the measure of the functional activity is a functional connectivity matrix (Hinton: Section 2…the soft-target distribution produced by the first network over its units is a functional-connectivity/activation summary, "knowledge is transferred to the distilled model by training it on a transfer set and using a soft target distribution for each case in the transfer set that is produced by using the cumbersome model with a high temperature in its softmax"). The rationale to combine Hinton with Masulli is the same as the parent claim.
Per claim 35, Masulli combined with Hinton discloses claim 28. Hinton further teaches wherein the method further comprises determining that second neural network is functioning properly or trained based on a functional response of the second neural network to the input pattern (Hinton: Abstract and Section 3, experiments verify that the distilled second network is properly trained by evaluating its response to validation input patterns, "we achieve some surprising results on MNIST"). The rationale to combine Hinton with Masulli is the same as the parent claim.
Claims 29-30 are rejected under 35 USC 103 as being unpatentable over Masulli in view of Hinton and further in view of Do Deep Nets Really Need to be Deep? to Ba et al. (hereinafter Ba).
Per claim 29, Masulli combined with Hinton discloses claim 28.
Masulli combined with Hinton does not expressly disclose, but with Ba does teach: wherein the second neural network is simpler than the first neural network (Ba: Abstract…a shallow neural network (second/simpler network) is trained to mimic a deeper neural network (first/more-complex network), "shallow feed-forward nets can learn the complex functions previously learned by deep nets and achieve accuracies previously only achievable with deep models").
Per claim 30, Masulli combined with Hinton discloses claim 28. Masulli combined with Hinton does not expressly disclose, but with Ba does teach: wherein the functioning of the first neural network is reconstructed in the second neural network without training the second neural network (Ba: pg.2…the shallow mimic net is fit directly to the logits of the deep teacher in one regression pass rather than being trained from labels in the conventional supervised sense, "The shallow mimic models, however, instead of being trained with cross-entropy…are trained directly on the 183 log probability values z, also called logit, before the softmax activation"; Hinton: Sections 1-2…the second model's parameters are set by matching soft targets from the first, which removes conventional label-based training).
For both claims 29 and 30, Masulli, Hinton and Ba are analogous art because all three fall within the same field of endeavor of characterizing, compressing, and transferring the functional behavior of artificial neural networks. They share the problem of reproducing a first neural network's behavior in a second model that is smaller/simpler or that is configured without conventional label training. Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to apply Ba's teaching that the mimic (second) network may be simpler than the first and may be fit directly from the teacher's functional characterization, to the Masulli combined with Hinton reconstruction framework.
The suggestion/motivation for doing so would have been the explicit teachings of Ba that a shallow (simpler) second network suffices to reproduce a deeper first network's function (Ba: Abstract) and training the second network directly on the first network's characterizations (rather than from labels) achieves the same accuracy (Ba: Section 3).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Patents and/or related publications are cited in the Notice of References Cited (Form PTO-892) attached to this action to further show the state of the art with respect to characterizing neural network activity using topological methods to distinguish between different inputs.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALAN CHEN whose telephone number is (571)272-4143. The examiner can normally be reached M-F 10-7.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Kamran Afshar can be reached at (571) 272-7796. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/ALAN CHEN/Primary Examiner, Art Unit 2125
Prosecution Insights