CTNF 18/477,575 CTNF 93840 DETAILED ACTION Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA. This action is responsive to the claims filed 9/29/2023. Claims 1-20 are presented for examination. Specification 07-29 AIA The disclosure is objected to because of the following informalities: In the description of the first snippet 100 of FIG. 1, the specification states that “the propositional nodes (two of them shown) are labeled with label numbers that are less than 110 and the propositional nodes (eight of them shown) are labeled with label numbers that are greater than 150”; the eight nodes numbered greater than 150 (152, 154, 156, 158, 160, 162, 164, 166) are the operational nodes, and the second occurrence of “propositional nodes” should read “operational nodes” . Appropriate correction is required. Claim Rejections - 35 USC § 112 07-30-02 AIA 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. 07-34-01 Claims 11-12 and 14-20 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 11 recites "iteratively applying J-modulated Fréchet bounds." Claim 11 depends from claim 1, which does not introduce the relative correlation coefficient "J." There is insufficient antecedent basis for the modifier "J," and "J-modulated" is not defined within the claim, so the metes and bounds of the limitation cannot be determined from the claim language. For purposes of examination, said limitations are interpreted as “ iteratively applying Fréchet bounds modulated by a relative correlation coefficient J bounded in a range of [-1, 1] ” consistent with the specification at [0055]–[0063]. Claim 12 depends from claim 11 and does not cure the deficiency of claim 11 and thus is rejected under 35 U.S.C. 112(b) for the same reasons set forth above with respect to claim 11. Claim 14 recites "predicting output made at previously labelled nodes." There is insufficient antecedent basis for "previously labelled nodes", neither claim 14 nor claim 1 recites any labelling of nodes, so it is unclear which nodes are encompassed. For purposes of examination, said limitations are interpreted as “ predicting output at the logical operational nodes ” consistent with the specification at [0074] and [00110]. Claim 15 recites "propositional nodes, logical operational nodes" and subsequently "the proposition and logical operator nodes" and "each logical operator of the logical operator nodes." There is insufficient antecedent basis for "proposition … nodes" and "logical operator nodes," which are inconsistent with the previously recited "propositional nodes" and "logical operational nodes"; and "each logical operator" is ambiguous as to whether it denotes a node, a component, or the logical operation performed. For purposes of examination, said limitations are interpreted as “ the propositional and logical operational nodes are coupled with respective belief bounds, and each logical operational node comprises a respective activation function set to a probability-respecting generalization of Fréchet inequalities ”, consistent with claims 1 and 18 and the specification [0019]–[0023]. Claim 16 recites "the operational nodes," which lacks antecedent basis because parent claim 15 recited "logical operational nodes"/"logical operator nodes," not "operational nodes." Claim 16 is further rejected for inheriting the indefiniteness of claim 15. For purposes of examination, said limitations are interpreted as the " logical operational nodes " as recited in claim 15, per the specification [0021]. Claim 17 recites "wherein the training comprises … during passes of the training." There is insufficient antecedent basis for "the training", neither claim 15 nor claim 17 previously recites training. Claim 17 is further rejected for inheriting the indefiniteness of claim 15. For purposes of examination, "the training" is interpreted as “ wherein training comprises ”. Claim 18 recites "propositional nodes, logical operational nodes" and subsequently "the proposition and logical operational nodes" and "each logical operator of the operational nodes." There is insufficient antecedent basis for "proposition … nodes" and for "logical operator," and "the operational nodes" is inconsistent with "logical operational nodes," rendering the scope indeterminate. For purposes of examination, said limitations are interpreted as “ the propositional and logical operational nodes are coupled with respective belief bounds, and each logical operational node comprises a respective activation function set to a probability-respecting generalization of Fréchet inequalities ”, consistent with claim 1 and the specification [0019]–[0023]. Claim 19 depends from claim 18 and does not cure the deficiency of claim 19 and thus is rejected under 35 U.S.C. 112(b) for the same reasons set forth above with respect to claim 18. Claim 20 recites "wherein the training comprises … during passes of the training." There is insufficient antecedent basis for "the training", neither claim 18 nor claim 20 previously recites training. Claim 20 is further rejected for inheriting the indefiniteness of claim 18. For purposes of examination, "wherein the training comprises" is interpreted as “ wherein training comprises ”. Claim Rejections - 35 USC § 103 07-06 AIA 15-10-15 In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status. 07-20-aia AIA 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. 07-21-aia AIA Claim s 1-3, 6-7, 10, 13, 15, 17-18, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Riegel et al. (hereinafter Riegel), US 2021/0365817 A1 in view of Qian et al. (hereinafter Qian) "Logical Credal Networks" (2021), and further in view of Fréchet “Generalization of the total probability theorem” (1935) . Qian and Fréchet disclosed in an IDS dated 9/29/2023. Regarding independent claim 1, Riegel teaches a computer-implemented method (Abstract, [0004], and FIG. 5, sets out systems, methods, and computer program products for configuring and using a logical neural network executed on computing hardware, FIG. 5 specifying the computer-performed method operations 503–520) , comprising: performing inferencing with a probabilistic logical neural network ([0005], [0030], Logical Neural Network (a probabilistic logical neural network) is “a neural network that has a 1-to-1 correspondence with a system of logical formulae, in which evaluation is equivalent to logical inference” and is “capable of inference in any direction, i.e. via normal evaluation, modus ponens, conjunction elimination, and all related inference rules”) , the probabilistic logical neural network comprising a probabilistic graphical model comprising propositional nodes, logical operational nodes, and directed edges ([0046], [0048], [0059], structurally Riegel's LNN is a graph comprising the syntax trees of formulae in a represented knowledgebase connected via nodes and connective edges (a probabilistic graphical model), “one neuron exists for each logical connective occurring in each formula and, additionally, one neuron for each unique proposition occurring in any formula” (the logical operational nodes and the propositional nodes), each connective neuron accepting as input the output of the neurons corresponding to its operands so that the operand-to-connective connections form the connective edges (directed edges)) , wherein the probabilistic logical neural network implements upward and downward inference ([0005], [0159], [0168], FIG. 5 operations 506 and 509, Riegel evaluates the neurons in the forward direction in a pass from leaves to root (upward inference) and backtracks for each leaf using inverse computations to update the subformula bounds (downward inference)) , the directed edges indicate a direction of upward inference ([0046], [0159], FIG. 5 operation 506, the forward pass propagates the computed bounds from leaves to root along the operand-to-connective connective edges (the directed edges), which is the direction of the forward, upward inference) , the downward inference is in an opposite direction from that of the directed edges ([0168], [0202], FIG. 5 operation 509, the backtracking pass performs weighted inverse bounds computation that updates the subformula bounds in the root-toward-leaves direction, opposite the leaf-to-root forward direction of the connective edges (the directed edges)) , the propositional and logical operational nodes are coupled with respective belief bounds ([0059], “All neurons return pairs of values in the range [0,1] representing upper and lower bounds on the truth values of their corresponding subformulae and propositions” (respective belief bounds)) , and each logical operational node comprises a respective activation function ([0049], [0059], [0095]–[0096], “Neurons corresponding to logical connectives accept as input the output of neurons corresponding to their operands and have activation functions configured to match the connectives' truth functions”, each connective neuron (each logical operational node) thus implementing a respective activation function) : Riegel does not expressly teach an set to a probability-respecting generalization However, Qian teaches a probability-respecting generalization (Abstract and Section 1, Qian’s Logical Credal Networks are “an expressive probabilistic logic that generalizes many prior models that combine logic and probability” (a generalization), the logic taking imprecise information represented by “probability bounds and conditional probability bounds of logic formulas” and specifying a set of probability distributions over all interpretations consistent with those bounds, so that every bound the logic computes is a valid probability of the associated logic formula (a probability-respecting generalization). Because Riegel and Qian are analogous art and within the same field of endeavor, specifically neuro-symbolic and probabilistic logical reasoning under imprecise or incomplete knowledge, they address the same problem solving area of performing sound inference over logical formulae when only bounded probabilities of propositions are known, accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention, to combine Qian’s probability-respecting probabilistic-logic generalization with Riegel’s logical-connective neurons, which already return lower and upper truth-value bounds in [0,1], with a reasonable expectation of success, such that each connective neuron’s activation function aggregates those bounds as valid probability bounds over the conjunctions and disjunctions of its operand propositions, thereby teaching each logical operational node comprises a respective activation function set to a probability-respecting generalization of bounds on logical combinations of propositions. This modification would have been motivated by the desire to give bounded truth values a sound probabilistic semantics for aggregating imprecise probability bounds over logic formulas (Qian: Abstract and Section 1). Riegel and Qian do not expressly teach that the probability-respecting generalization is a generalization of Fréchet inequalities . However, Fréchet teaches Fréchet inequalities (Fréchet: pp. 379–387, Fréchet generalizes the total-probability theorem and establishes, for chance events H ₁ , …, H ₙ of respective probabilities p ₁ , …, p ₙ , the double inequality p ₁ + p ₂ + … + p ₙ − (n−1) ≤ P ≤ p ₁ + p ₂ + … + p ₙ , which bounds from below and from above the probability P of a combination of those events using only the probabilities of the individual events (the lower and upper Fréchet inequalities on the conjunction and disjunction of events), the bounds holding whatever the unknown dependence among the events). Because Riegel, in view of Qian, and Fréchet are analogous art and within the same field of endeavor, specifically probabilistic reasoning over combinations of events under bounded or imprecise probabilities, they address the same problem solving area of computing valid probability bounds on the conjunction and disjunction of events from the probabilities of the individual events, accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention, to use Fréchet’s lower and upper bounds on the conjunction and disjunction of events as the particular probability-respecting generalization computed by the connective-neuron activation functions of Riegel as modified by Qian, with a reasonable expectation of success, the Fréchet inequalities being the long-established closed-form bounds for precisely the probability-of-combination quantities that Qian’s probability-respecting generalization assigns to Riegel’s conjunction and disjunction connectives, thereby arriving at each logical operational node comprises a respective activation function set to a probability-respecting generalization of Fréchet inequalities. This modification would have been motivated by the desire to equip Riegel’s probability-respecting connective activation functions with sound, computable lower and upper probability bounds on conjunctions and disjunctions even when the dependence among the operand propositions is unknown (Fréchet: pp. 379–387). Regarding dependent claim 2, Riegel, in view of Qian and Fréchet, teach the method of claim 1, wherein the propositional nodes are associated with assertions (see Riegel [0048], proposition neurons (the propositional nodes) provide inputs that are “facts about the world”, that is, assertions of the underlying knowledgebase (assertions)). Regarding dependent claim 3, Riegel, in view of Qian and Fréchet, teach the method of claim 1, wherein the directed edges respectively point from a propositional node to a logical operational node or from one logical operational node to another logical operational node (see Riegel [0048], [0059], “Neurons corresponding to logical connectives accept as input the output of neurons corresponding to their operands”, so that a proposition neuron (a propositional node) feeds a connective neuron (a logical operational node) and a connective neuron feeds a higher-level connective neuron, the connections being the directed edges). Regarding dependent claim 6, Riegel, in view of Qian and Fréchet, teach the method of claim 1, wherein for each node of the probabilistic logical neural network the belief bounds comprise a lower bound and an upper bound, wherein the lower bound and the upper bound are both greater than or equal to zero and less than or equal to one (see Riegel [0059], “All neurons return pairs of values in the range [0,1] representing upper and lower bounds on the truth values of their corresponding subformulae and propositions” (a lower bound and an upper bound that are both greater than or equal to zero and less than or equal to one)). Regarding dependent claim 7, Riegel, in view of Qian and Fréchet, teach the method of claim 1, wherein the probabilistic logical neural network includes a respective weight for each of the directed edges, wherein the weights are initialized to a value of 1 and adjusted during successive iterations of training (see Riegel [0006], [0007], teaches a weighted logic that behaves exactly classically given classical input, [0049], [0054]–[0056], each input operand of a logical neuron has “a respective assigned or computed weight” representing an importance over the neuron's inputs (a respective weight for each of the directed edges), and during the LNN learning phase the parameters of the model, including the neuron weights, are tweaked to minimize the loss function (adjusted during successive iterations of training); because classical behavior of Riegel's weighted-logic neurons corresponds to unit-valued weights, initializing each weight to a value of 1 (initialized to a value of 1) would be the classical-logic default of Riegel's framework). Regarding dependent claim 10, Riegel, in view of Qian and Fréchet, teach the method of claim 1, wherein the logical operational nodes comprise one or more implication nodes indicating beliefs about conditional probabilities between two or more inputs (see Riegel [0005], [0042]–[0046], Riegel's connective neurons (the logical operational nodes) include neurons implementing the implication connective, the network performing modus ponens, modus tollens, transposition, and related inference rules over those implication connectives (one or more implication nodes); and see Qian Abstract and Section 1, attaches “probability bounds and conditional probability bounds of logic formulas” (beliefs about conditional probabilities) to such implications). Regarding dependent claim 13, Riegel, in view of Qian and Fréchet, teach the method of claim 1, further comprising forming the probabilistic logical neural network via receiving the propositional nodes, the logical operational nodes, and the belief bounds as input, initializing weights, identifying some of the logical operational and propositional nodes as ground truth, and adjusting the weights using backpropagation to minimize loss based on the ground truth (see Riegel [0046], [0048], [0054]–[0056], Riegel constructs the LNN in a 1-to-1 correspondence with a system of logical formulae from its proposition and connective neurons (the propositional nodes, the logical operational nodes) and, for the learning phase, receives training inputs together with ground-truth bounds for the proposition nodes drawn from observed training facts and training queries (the belief bounds as input; identifying some of the logical operational and propositional nodes as ground truth), the parameters of the model then being tweaked by gradient-based optimization to minimize the loss function (initializing weights; adjusting the weights using backpropagation to minimize loss based on the ground truth)). Regarding dependent claim 14, Riegel, in view of Qian and Fréchet, teach the method of claim 1, wherein the inferencing comprises receiving new data at the propositional nodes, clamping values of the propositional nodes based on the received new data, and predicting output made at previously labelled nodes (interpreted as output at the logical operational nodes per the 35 U.S.C. 112(b) rejection set forth above) (see Riegel [0048], [0094], FIG. 5 operations 503 and 506, Riegel's proposition neurons (the propositional nodes) receive inputs that are facts about the world whose truth-value bounds are “clamped to be a value 0 or 1” based on the received facts (receiving new data at the propositional nodes; clamping values of the propositional nodes based on the received new data), and the connective neurons are then evaluated in the forward pass from leaves to root, each emitting an output truth-value bound pair (predicting output at the logical operational nodes)). Regarding claims 15 and 17, these are computer system claims that are substantially the same as the method of claims 1 and 7, respectively. Thus, claims 15 and 17 are rejected for the same reasons as claims 1 and 7. In addition, In addition, Riegel teaches a computer system comprising: one or more processors, one or more computer-readable memories, and program instructions stored on at least one of the one or more computer-readable memories for execution by at least one of the one or more processors (Abstract, [0004], [0010], [0012], [0433], and [0439] provides its logical neural network as systems, methods, and computer program products, the network being an end-to-end differentiable neural network executed on computing hardware, gradient-based learning on which presupposes execution by one or more processors with associated memory storing the program instructions). Regarding claims 18 and 20, these are computer program product claims that are substantially the same as the method of claims 1 and 7, respectively. Thus, claims 18 and 20 are rejected for the same reasons as claims 1 and 7. In addition, Riegel teaches a computer program product comprising a computer-readable storage medium having program instructions embodied therewith ([0004], [0012], [0439], Riegel further provides its logical neural network as a computer program product whose program instructions reside on a computer-readable storage medium) . 07-21-aia AIA Claim s 4-5, 11-12, and 16 are rejected under 35 USC 103 as being unpatentable over Riegel, in view of Qian and Fréchet, and further in view of Miralles-Dolz et al. (hereinafter Miralles-Dolz) “Correlated Boolean Operators for Uncertainty Logic” (2022) . Regarding dependent claim 4, Riegel, in view of Qian and Fréchet, teach all the elements of claim 1. Riegel, Qian, and Fréchet do not expressly teach wherein the logical operational nodes incorporate relative correlation coefficients bounded in a range of [-1, 1] that modulate the Fréchet inequalities, and wherein the relative correlation coefficients are taken as input and provided as output at the logical operational nodes, respectively. However, Miralles-Dolz teaches wherein the logical operational nodes incorporate relative correlation coefficients bounded in a range of [-1, 1] that modulate the Fréchet inequalities, and wherein the relative correlation coefficients are taken as input and provided as output at the logical operational nodes, respectively (Abstract and Sections 2–3, presents a correlated AND gate parameterized by a Pearson correlation coefficient ρ (relative correlation coefficients) forming “a complete copula family, as it includes the two Fréchet-Hoeffding bounds W and M, corresponding to minimal (when ρ = −1) and maximal (when ρ = 1) correlation respectively, and the independence copula Π(u, v) = uv when ρ = 0”, thereby continuously modulating the Fréchet-Hoeffding bounds across [-1, 1] (modulate the Fréchet inequalities), the correlated AND gate taking the correlation coefficient ρ as its parameter and returning the joint probability of the two events as its output (the relative correlation coefficients taken as input and provided as output)). Because Riegel, in view of Qian and Fréchet, and Miralles-Dolz are analogous art and within the same field of endeavor, specifically probabilistic logical reasoning under bounded or imprecise probabilities, they address the same problem solving area of computing sound probability bounds on conjunctions and disjunctions when the dependence between events is only partially known, accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention, to combine Miralles-Dolz's ρ-modulated correlated AND gate with Riegel's logical-connective neurons, which return lower and upper truth-value bounds in [0,1], with a reasonable expectation of success, such that each connective neuron computes its bounds as the correlation-parameterized copula bounds to teach wherein the logical operational nodes incorporate relative correlation coefficients bounded in a range of [-1, 1] that modulate the Fréchet inequalities. This modification would have been motivated by the desire to obtain tighter per-neuron probability bounds under arbitrary correlation, Miralles-Dolz expressly framing its correlated AND gate as a continuous bridge between the Fréchet-Hoeffding bounds and the independence copula, the results of which “generalize previous results by Fréchet on the conjunction of two events with unknown dependencies” (Miralles-Dolz: Abstract and Sections 2–3). Regarding dependent claim 5, Riegel, in view of Qian, Fréchet, and Miralles-Dolz, teach the method of claim 4, wherein the relative correlation coefficients interpolate between a maximum anti-correlation represented by -1, statistical independence represented by 0, and maximum correlation represented by 1 (see Miralles-Dolz Sections 2–3, Miralles-Dolz's copula family places the Fréchet-Hoeffding lower bound W at ρ = −1, representing minimal or anti-correlation (a maximum anti-correlation represented by -1), the independence copula at ρ = 0 (statistical independence represented by 0), and the Fréchet-Hoeffding upper bound M at ρ = +1, representing maximal correlation (maximum correlation represented by 1), continuously interpolating across the range between these three regimes (interpolate between)). Regarding dependent claim 11, Riegel, in view of Qian and Fréchet, teach the method of claim 1, further comprising forming the probabilistic logical neural network via: receiving the propositional nodes, the logical operational nodes, and the belief bounds as input, initializing weights, and systematically tightening the belief bounds using iterations of upward and downward inference (see Riegel [0046], [0059], [0202], FIG. 5 operations 503–515, constructs the network from its proposition and connective neurons, initializes the neurons with starting truth-value bounds (receiving the propositional nodes, the logical operational nodes, and the belief bounds as input; initializing weights), and iteratively evaluates the network by forward and inverse bounds computation, aggregating the tightest bounds at each proposition until the bounds converge (systematically tightening the belief bounds using iterations of upward and downward inference)). Riegel, Qian, and Fréchet do not expressly teach iteratively applying J-modulated Fréchet bounds (interpreted as iteratively applying Fréchet bounds modulated by a relative correlation coefficient J bounded in a range of [-1, 1] per the 35 U.S.C. 112(b) rejection set forth above). However, Miralles-Dolz teaches iteratively applying Fréchet bounds modulated by a relative correlation coefficient J bounded in a range of [-1, 1] (see Miralles-Dolz Abstract and Sections 2–3, Miralles-Dolz's Pearson correlation coefficient ρ taking any value in the range [-1, 1] (a relative correlation coefficient J bounded in a range of [-1, 1]) parameterizes a copula family that modulates the Fréchet-Hoeffding bounds, which, applied across Riegel's iterative forward and inverse bound-tightening, yields the iterative application of the modulated bounds). Because Riegel, in view of Qian and Fréchet, and Miralles-Dolz are analogous art and within the same field of endeavor, specifically iterative probabilistic logical reasoning and bound propagation under bounded or imprecise probabilities, they address the same problem solving area of progressively tightening the probability bounds on conjunctions and disjunctions across successive inference passes when the dependence between the operand events is only partially known, accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention, to apply Miralles-Dolz’s ρ-modulated correlated AND gate at each pass of Riegel’s iterative forward-and-inverse bound-tightening loop, as modified by Qian and Fréchet to compute probability-respecting Fréchet bounds at the connective neurons, with a reasonable expectation of success, such that each iteration recomputes the Fréchet bounds as modulated by the relative correlation coefficient J in [-1, 1] and aggregates the resulting tightened bounds at each proposition until the bounds converge, to teach iteratively applying Fréchet bounds modulated by a relative correlation coefficient J bounded in a range of [-1, 1] . This modification would have been motivated by the desire to obtain progressively tighter probability bounds as the network converges by accounting for arbitrary or partially known correlation between the operand events at every iteration rather than only once, Miralles-Dolz expressly framing its correlated AND gate as a continuous bridge between the Fréchet-Hoeffding bounds and the independence copula, the results of which “generalize previous results by Fréchet on the conjunction of two events with unknown dependencies” (Miralles-Dolz: Abstract and Sections 2–3). Regarding dependent claim 12, Riegel, in view of Qian, Fréchet, and Miralles-Dolz, teach the method of claim 11, wherein the receiving occurs via ingesting a neural network (see Riegel [0030], [0046], Riegel's logical neural network is itself a neural network having a 1-to-1 correspondence with the underlying system of logical formulae, such that its proposition and connective neurons are received by ingesting that neural network (ingesting a neural network)). Regarding claim 16, this is a computer system claim that is substantially the same as the method of claim 4. Thus, claim 16 is rejected for the same reason as claim 4 . 07-21-aia AIA Claim s 8, 9, and 19 are rejected under 35 USC 103 as being unpatentable over Riegel in view of Qian and Fréchet, and further in view of Fahlman et al. (hereinafter Fahlman) “The Cascade-Correlation Learning Architecture” (1989) . Regarding dependent claim 8, Riegel, in view of Qian and Fréchet, teach all the elements of claim 1, Riegel, Qian, and Fréchet do not expressly teach further comprising performing node spawning during training in response to identifying non-unital weights amongst inputs, the node spawning adding a new node to the probabilistic graphical model. However, Fahlman teaches performing node spawning during training in response to identifying non-unital weights amongst inputs, the node spawning adding a new node to the probabilistic graphical model (Sections 1–2, Cascade-Correlation begins with a minimal network and automatically trains and adds new hidden units one at a time during training, “hidden units are added to the network one at a time and do not change after they have been added” (performing node spawning during training; the node spawning adding a new node to the graphical model), each newly added unit becoming a permanent node of the network with its input-side weights frozen; Fahlman adds a new candidate unit when continued training of the existing output weights approaches an asymptote and those weights cease to reduce the residual output error, that is, when the existing connection weights amongst the inputs have stopped driving the network toward unit, saturated performance (in response to identifying non-unital weights amongst inputs)). Because Riegel, in view of Qian and Fréchet, and Fahlman are analogous art and within the same field of endeavor, specifically the training of neural networks whose units represent structured, here logical, computations, they address the same problem solving area of growing a network's capacity during training so that it acquires the units needed to fit the training data, accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention, to combine Fahlman's cascade-correlation node-addition technique with Riegel's gradient-trained logical neural network, with a reasonable expectation of success, such that Riegel's connective-neuron count grows during training when the existing weights stop reducing the residual error to teach further comprising performing node spawning during training in response to identifying non-unital weights amongst inputs, the node spawning adding a new node to the probabilistic graphical model. This modification would have been motivated by the desire to let the network determine its own size and topology and add feature-detecting units as needed (Fahlman: Abstract and Sections 1–2) and the use of Fahlman's residual-error node-addition trigger in place of the recited non-unital-weight trigger is a simple substitution of one known training trigger for another to obtain the predictable result of capacity growth during training. Regarding dependent claim 9, Riegel, in view of Qian, Fréchet, and Fahlman, teach the method of claim 8, wherein the node spawning is performed in an inner loop of the training and additional loss minimization is performed in an outer loop of the training (see Fahlman Sections 1–2, Fahlman's training comprises an inner loop that trains a pool of candidate units to maximize “the magnitude of the correlation (or, more precisely, the covariance) between V, the candidate unit's value, and Eo, the residual output error” and then installs the best candidate as a new unit (the node spawning performed in an inner loop), and an outer loop that retrains the network's output weights to minimize the loss given the updated architecture (additional loss minimization performed in an outer loop)). Regarding claim 19, this is a computer program product claim that is substantially the same as the method of claim 8. Thus, claim 19 is rejected for the same reason as claim 8. Conclusion Any inquiry concerning this communication or earlier communications from the examiner should be directed to KUANG FU CHEN whose telephone number is (571)272-1393. The examiner can normally be reached M-F 9:00-5:30pm ET. 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, Jennifer Welch can be reached on (571) 272-7212. 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. /KC CHEN/Primary Patent Examiner, Art Unit 2143 Application/Control Number: 18/477,575 Page 2 Art Unit: 2143 Application/Control Number: 18/477,575 Page 3 Art Unit: 2143 Application/Control Number: 18/477,575 Page 4 Art Unit: 2143 Application/Control Number: 18/477,575 Page 5 Art Unit: 2143 Application/Control Number: 18/477,575 Page 6 Art Unit: 2143 Application/Control Number: 18/477,575 Page 7 Art Unit: 2143 Application/Control Number: 18/477,575 Page 8 Art Unit: 2143 Application/Control Number: 18/477,575 Page 9 Art Unit: 2143 Application/Control Number: 18/477,575 Page 10 Art Unit: 2143 Application/Control Number: 18/477,575 Page 11 Art Unit: 2143 Application/Control Number: 18/477,575 Page 12 Art Unit: 2143 Application/Control Number: 18/477,575 Page 13 Art Unit: 2143 Application/Control Number: 18/477,575 Page 14 Art Unit: 2143 Application/Control Number: 18/477,575 Page 15 Art Unit: 2143 Application/Control Number: 18/477,575 Page 16 Art Unit: 2143 Application/Control Number: 18/477,575 Page 17 Art Unit: 2143 Application/Control Number: 18/477,575 Page 18 Art Unit: 2143 Application/Control Number: 18/477,575 Page 19 Art Unit: 2143 Application/Control Number: 18/477,575 Page 20 Art Unit: 2143 Application/Control Number: 18/477,575 Page 21 Art Unit: 2143
Prosecution Insights