DETAILED ACTION 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. Priority Acknowledgment is made of applicant’s claim for foreign priority under 35 U.S.C. 119 (a)-(d). Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55. Information Disclosure Statement The information disclosure statements submitted on September 13, 2022 and February 2, 2026 have been considered by the Examiner. Claim Rejections - 35 USC § 112 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 appl icant regards as his invention. Claim 18 is 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. In particular, in claim 18, there is not antecedent basis for “[t]he OOD data apparatus,” “the processing circuit” and “the threshold” recited therein. For examination purposes, claim 18 is interpreted to depend from claim 17 , which provides antecedence for “[t]he OOD data apparatus,” “the processing circuit” and “the threshold.” 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 1-18 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea (e.g. mathematical concept, mental process) without significantly more. As described in MPEP § 2106, the analyses as to whether a claim qualifies as eligible subject matter under 35 U.S.C. § 101 includes the following determinations: (1) Whether the claim is to a statutory category, i.e. to a process, machine, manufacture or composition of matter (“Step 1”) – see MPEP §§ 2106, subsection III, and 2106.03 (2) If the claim is to a statutory category, whether the claim recites any judicial exceptions, including certain groupings of abstract ideas (i.e., mathematical concepts, certain methods of organizing human activity, or mental processes) (“Step 2A, Prong One”) – see MPEP §§ 2106, subsection III, and 2106.04 (3) If the claim recites a judicial exception, whether the claim recites additional elements that integrate the judicial exception into a practical application (“Step 2A, Prong Two”) – see MPEP §§ 2106, subsection III, and 2106.04 (4) If the claim does not recite additional elements that integrate the judicial exception into a practical application, whether the claim recites additional elements that amount to significantly more than the judicial exception (“Step 2B”) – see MPEP §§ 2106, subsection III, and 2106.05 Claim 1 Regarding “Step 1,” independent claim 1 is to a statutory category as claim 1 is directed to a n apparatus, which can be considered a machine. Accordingly, the analysis proceeds to “Step 2A, Prong One” to determine if the claim recites a judicial exception. In this case, claim 1 recites mathematical concepts and mental processes and thus recites a judicial exception. In particular , the recitations of “calculate an intermediate output by applying a trained model to the monitoring target data” and “calculate a noise influence level of the intermediate output in a given layer, using only a parameter that has a high contribution to a task among parameters constituting the trained model,” when given their broadest reasonable interpretations, are considered mathematical concepts. The recitation of “discriminate as to whether or not the monitoring target data is OOD data based on the noise influence level” is considered a mental process. Because claim 1 recites a judicial exception (i.e. mathematical concepts, a mental process), the analysis proceeds to “Step2A, Prong Two.” But here the claim does not recite additional elements that integrate the judicial exception into a practical application. In particular, in addition to the above-noted mathematical concepts and mental process, claim 1 recites a task to “obtain monitoring target data” and that the apparatus comprises a “processing circuit” configured to implement the tasks recited in claim 1. However, “ obtain[ ing ] monitoring target data” is considered insignificant extra-solution activity, i.e. mere data gathering, and is insufficient to integrate the abstract idea into a practical application. See MPEP § 2106.5(g). That this task and the others recited in claim 1, which as noted above are considered mathematical concepts and mental processes, are implemented via a “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus also do es not integrate the judicial exception into a practical application. See MPEP § 2106.05(f). Accordingly, as claim 1 does not recite additional elements that integrate the judicial exception into a practical application, the analysis proceeds to “Step 2B” to determine whether the claims recite additional elements that amount to significantly more than the judicial exception. However, in this case, the claim does not. As noted above, in addition to the above-noted mathematical concepts and mental process, claim 1 recites a task to “obtain monitoring target data . ” As further noted above, this is indicative of insignificant extra-solution activity, i.e. mere data gathering. See MPEP § 2106.5(g). Such data gathering is also well-understood, routine and conventional. See, e.g., Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015); buySAFE , Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014). Claim 1 also recites that the tasks therein are implemented with a “processing circuit.” However, like further noted above, this represent s no more than mere instructions to apply the judicial exception on a computer, and thus also do not amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Consequently, claim 1 recites an abstract idea but does not include additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea. As a result, and for the reasons described above, claim 1 is rejected as being patent ineligible under 35 U.S.C. § 101. Claim 2 In claim 2, the recitation of “calculate, as the noise influence level, a projected component of the intermediate output to the parameter” is considered a mathematical concept. That this task is implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 2 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 2 is also patent ineligible under 35 U.S.C. § 101. Claim 3 In claim 3, the recitation of “determine a projection matrix based on a matrix decomposition of the parameter, and calculate the projected component by making the projection matrix act on the intermediate output” is considered a mathematical concept. That this task is implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 3 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 3 is also patent ineligible under 35 U.S.C. § 101. Claim 4 In claim 4, the recitation of “calculate the projection matrix by performing singular value decomposition on a weight parameter constituting the trained model” is considered a mathematical concept. That this task is implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 4 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 4 is also patent ineligible under 35 U.S.C. § 101. Claim 5 In claim 5, the recitation of “delete, as a component having a low contribution to the task, a matrix component that has a small singular value, among matrix components included in the projection matrix” is considered a mathematical concept or mental process. That this task is implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 5 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 5 is also patent ineligible under 35 U.S.C. § 101. Claim 6 In claim 6, the recitation of “search for a matrix component that satisfies a predetermined condition based on a change in a task performance of the trained model when positions and/or a number of the matrix components included in the projection matrix are changed” is considered a mathematical concept or mental process. That this task is implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 6 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 6 is also patent ineligible under 35 U.S.C. § 101. Claim 7 In claim 7, the following recitations are considered mathematical concepts: “inject a noise to the intermediate output in an intermediate hidden layer of the trained model” wherein the noise influence level is calculated by “calculates a first intermediate output by applying a hidden layer later than the intermediate hidden layer to the intermediate output to which the noise is not injected; calculates a second intermediate output by applying the later hidden layer to the intermediate output to which the noise is injected; and calculates a degree of variation between the first intermediate output and the second intermediate output.” That these tasks are implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 7 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 7 is also patent ineligible under 35 U.S.C. § 101. Claim 8 In claim 8, the recitation of “calculate a contribution to the task using a singular value of the parameter” is considered a mathematical concept. That this task is implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 8 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 8 is also patent ineligible under 35 U.S.C. § 101. Claim 9 In claim 9, the recitation of “delete, as a component having a low contribution to the task, a matrix component that has a small singular value in the parameter” is considered a mathematical concept or mental process. That this task is implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 9 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 9 is also patent ineligible under 35 U.S.C. § 101. Claim 10 In claim 10, the recitation of “search for a matrix component that satisfies a predetermined condition based on a change in a task performance of the trained model when positions and/or a number of the matrix components included in the parameter are changed” is considered a mathematical concept or mental process. That this task is implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 10 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 10 is also patent ineligible under 35 U.S.C. § 101. Claim 11 In claim 11, the recitation of “convert the noise influence level to a one-dimensional variable for discrimination” is considered a mathematical concept. The recitation of “discriminate as to whether or not the monitoring target data is OOD based on a comparison between the variable for discrimination and a threshold” is considered a mental process. That these tasks are implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 11 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 11 is also patent ineligible under 35 U.S.C. § 101. Claim 12 In claim 12, the recitation of “set data not used for training the trained model among a plurality of training data sets as OOD data, and sets the threshold using the OOD data” is considered a mental process. That this task is implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 12 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 12 is also patent ineligible under 35 U.S.C. § 101. Claim 13 In claim 13, the recitation of “set the threshold in such a manner that an outlier of the training data used for training the trained model can be classified into OOD data” is considered a mental process. That this task is implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 13 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 13 is also patent ineligible under 35 U.S.C. § 101. Claim 1 4 Regarding “Step 1,” independent claim 1 4 is to a statutory category as claim 1 4 is directed to a method, i.e. a process. Accordingly, the analysis proceeds to “Step 2A, Prong One” to determine if the claim recites a judicial exception. In this case, claim 1 4 recites mathematical concepts and mental processes and thus recites a judicial exception. In particular, the recitations of “calculat ing an intermediate output by applying a trained model to the monitoring target data” and “calculat ing a noise influence level of the intermediate output in a given layer, using only a parameter that has a high contribution to a task among parameters constituting the trained model,” when given their broadest reasonable interpretations, are considered mathematical concepts. The recitation of “discriminat ing as to whether or not the monitoring target data is OOD data based on the noise influence level” is considered a mental process. Because claim 1 4 recites a judicial exception (i.e. mathematical concepts, a mental process), the analysis proceeds to “Step2A, Prong Two.” But here the claim does not recite additional elements that integrate the judicial exception into a practical application. In particular, in addition to the above-noted mathematical concepts and mental process, claim 1 4 recites “obtain ing monitoring target data . ” However, “ obtain ing monitoring target data” is considered insignificant extra-solution activity, i.e. mere data gathering, and is insufficient to integrate the abstract idea into a practical application. See MPEP § 2106.5(g). Accordingly, as claim 1 4 does not recite additional elements that integrate the judicial exception into a practical application, the analysis proceeds to “Step 2B” to determine whether the claims recite additional elements that amount to significantly more than the judicial exception. However, in this case, the claim does not. As noted above, in addition to the above-noted mathematical concepts and mental process, claim 1 4 recites a task of “obtain ing monitoring target data.” As further noted above, this is indicative of insignificant extra-solution activity, i.e. mere data gathering. See MPEP § 2106.5(g). Such data gathering is also well-understood, routine and conventional. See, e.g., Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015); buySAFE , Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014). Consequently, claim 1 4 recites an abstract idea but does not include additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea. As a result, and for the reasons described above, claim 1 4 is rejected as being patent ineligible under 35 U.S.C. § 101. Claim 1 5 Regarding “Step 1,” independent claim 1 5 is to a statutory category as claim 1 5 is directed to a non-transitory computer readable storage medium, which is considered a machine, manufacture, or composition of matter . Accordingly, the analysis proceeds to “Step 2A, Prong One” to determine if the claim recites a judicial exception. In this case, claim 1 5 recites mathematical concepts and mental processes and thus recites a judicial exception. In particular, the recitations of “calculating an intermediate output by applying a trained model to the monitoring target data” and “calculating a noise influence level of the intermediate output in a given layer, using only a parameter that has a high contribution to a task among parameters constituting the trained model,” when given their broadest reasonable interpretations, are considered mathematical concepts. The recitation of “discriminating as to whether or not the monitoring target data is OOD data based on the noise influence level” is considered a mental process. Because claim 1 5 recites a judicial exception (i.e. mathematical concepts, a mental process), the analysis proceeds to “Step2A, Prong Two.” But here the claim does not recite additional elements that integrate the judicial exception into a practical application. In particular, in addition to the above-noted mathematical concepts and mental process, claim 1 5 recites “obtaining monitoring target data.” However, “ obtaining monitoring target data” is considered insignificant extra-solution activity, i.e. mere data gathering, and is insufficient to integrate the abstract idea into a practical application. See MPEP § 2106.5(g). Claim 15 also additionally recites that the tasks recited therein are implemented via computer executable instructions stored on the non-transitory computer readable medium. However, this represent s no more than mere instructions to apply the judicial exception on a computer, and thus also do not integrate the judicial exception into a practical application. See MPEP § 2106.05(f). Accordingly, as claim 1 5 does not recite additional elements that integrate the judicial exception into a practical application, the analysis proceeds to “Step 2B” to determine whether the claims recite additional elements that amount to significantly more than the judicial exception. However, in this case, the claim does not. As noted above, in addition to the above-noted mathematical concepts and mental process, claim 1 5 recites a task of “obtaining monitoring target data.” As further noted above, this is indicative of insignificant extra-solution activity, i.e. mere data gathering. See MPEP § 2106.5(g). Such data gathering is also well-understood, routine and conventional. See, e.g., Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015); buySAFE , Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014). Claim 15 also recites that the tasks recited therein are implemented via computer executable instructions stored on the non-transitory computer readable medium. However, like further noted above, this represent s no more than mere instructions to apply the judicial exception on a computer, and thus also do es not amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Consequently, claim 1 5 recites an abstract idea but does not include additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea. As a result, and for the reasons described above, claim 1 5 is rejected as being patent ineligible under 35 U.S.C. § 101. Claim 16 In claim 16, the recitation of “calculate, as the variable for discrimination, a norm of the noise influence level, or a ratio of a norm of the intermediate output to the norm of the noise influence level” is considered a mathematical concept. That this task is implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 16 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 16 is also patent ineligible under 35 U.S.C. § 101. Claim 17 In claim 17, the recitations of “determine a threshold for each layer of the trained model” and “discriminate as to whether or not the monitoring target data is OOD data based on a comparison between the variable for discrimination and the threshold for each layer” are considered mental processes. That these tasks are implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 17 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 17 is also patent ineligible under 35 U.S.C. § 101. Claim 18 In claim 18, the recitation of “determine the threshold based on a rank of the parameter for each layer” is considered a mental process. That this task is implemented via the “processing circuit” represent s no more than mere instructions to apply the judicial exception on a computer, and thus does not integrate the judicial exception into a practical application or amount to significantly more than the judicial exception . See MPEP § 2106.05(f). Accordingly, claim 18 fails to recite any additional elements that integrate the abstract idea into a practical application or that amount to significantly more than the abstract idea, and as a result, claim 18 is also patent ineligible under 35 U.S.C. § 101. 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. 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 1 -10, 14 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication No. 2021/0365771 to Speakman et al. (“Speakman”), and also over the article entitled “ Efficiently C ombining SVD , P runing, C lustering and R etraining for E nhanced N eural N etwork C ompression ” by Goetschalckx et al. (“ Goetschalckx ”). Regarding claim s 1 , 14 and 15 , Speakman generally describes a method and system to detect out-of-distribution data in machine learning (see e.g. paragraph 0003). Like claimed, Speakman particularly teaches: obtaining monitoring target data ( see e.g. paragraphs 0003 and 0058 : Speakman teaches obtaining a first set of activations from nodes in a hidden layer of a neural network for an input. The input can is considered “monitoring target data” like claimed.); calculating an intermediate output by applying a trained model to the monitoring target data (see e.g. paragraphs 0003 and 0058: like noted above, Speakman teaches obtaining a first set of activations from nodes in a hidden layer of a neural network for an input. The set of activations from the hidden layer of the neural network is considered an intermediate output like claimed, which is calculated by applying a trained model to the monitoring target data, i.e. to the input.); calculating a noise influence level of the intermediate output in a given layer using parameters of constituting the trained model (see e.g. paragraphs 0003 and 0059-0060: Speakman teaches next adding noise to the input, and obtaining a second set of activations from the nodes in the hidden layer of the neural network when provided the noised input. Speakman further teaches determining a difference between the first set of activations and the second set of activations – see e.g. paragraphs 0003 and 0061. The difference can be considered a “noise influence level” of the intermediate output like claimed. This difference is calculated in part by using the parameters of the neural network, which are necessary to obtain the first and second sets of activations upon which the difference is determined.); and discriminate as to whether or not the monitoring target data is out-of-distribution (OOD) based on the noise influence level (see e.g. paragraphs 0003 , 0024, 0048 and 0062 -0063 : Speakman teaches comparing the difference with differences similarly computed with in-distribution inputs to identify whether or not the input is out-of-distribution.). Speakman thus teaches a method similar to that of claim 14. Speakman further discloses that such teachings can be implemented via computer executable instructions stored on a non-transitory computer readable storage medium for execution by a processing circuit (e.g. a processor ) of a computer (see e.g. paragraphs 0089 -0090 and 0098-0099). A non-transitory computer readable medium comprising computer executable instructions to implement the above-described teachings of Speakman is considered a non-transitory computer readable medium similar to that of claim 15, and a computer comprising a processor to execute such instructions is considered an out-of-distribution data detection apparatus similar to that of claim 1. Speakman, however does not explicitly disclose that the noise influence level is calculated “using only a parameter that has a high contribution to a task among parameters constituting the trained model,” as is recited in each of claims 1, 14 and 15. Goetschalckx nevertheless generally teaches techniques to compress neural network models, including by employing singular value decomposition (SVD) and by pruning compressed sparse matrix formats: However, the size of most common NN models can be up to tens of megabytes [8]. They hence significantly exceed embedded SRAM capacity and must be stored in DRAM. This results in many energy expensive off-chip DRAM accesses, easily summing up to unaffordable energy costs. Various techniques have been proposed to compress these DNN models, leading to smaller model sizes and, consequently, significantly lower energy usage. The two popular methods in the state-of-the-art are (1) based on singular value decomposition (SVD) [17] and (2) based on pruning and compressed sparse matrix formats , a technique also referred to as Deep Compression (DC) [6]. (Page 1 ; emphasis added ). Goetschalckx particular ly discloses that SVD can be applied to a weight matrix, and whereby the smallest, least important values of the resulting matrices can be removed but still closely approximate multiplication with the original weight matrix: SVD [17] applies singular value decomposition to a weight matrix . This factors it into two new weight matrices of which the columns of the first and rows of the second are sorted by decreasing singular value. These values can intuitively be considered as an importance metric. The rows and columns corresponding to the smallest, least important, singular values are then r em oved, resulting in two smaller weight matrices. This decreases the total number of weights to be stored , while sequential multiplication o f a vector with these smaller weight matrices still closely approximates multiplication with the large original weight matrix . (Page 2 ; emphasis added ). As noted above, the utilization of SVD decreases the total number of weights to be stored. Accordingly, it would have been obvious to one of ordinary skill in the art, having the teachings of Speakman and Goetschalckx before the effective filing date of the claimed invention, to modify the apparatus, method and non-transitory computer readable storage medium taught by Speakman so as to employ SVD like taught by Goetschalckx to one or more of the intermediate layers therein, and thereby remove the smallest, least important weights. As a result , the noise influence level (i.e. the difference between the activations of the intermediate layer when an input and a noised input are applied to the neural network) would be calculated using only parameters (i.e. weights) that have a high contribution to a task among parameters constituting the trained model. It would have been advantageous to one of ordinary skill to utilize such a combination because it can decrease the amount of weights required to be stored , as is taught by Goetschalckx (see e.g. the excerpt from page 2 provided above) . Accordingly, Speakman and Goetschalckx are considered to teach, to one of ordinary skill in the art, an OOD data detection apparatus like that of claim 1, an OOD data detection method like that of claim 14, and a non-transitory computer readable storage medium like that of claim 15. As per claim 2, Speakman further teaches that the processing circuit is configured to calculate, as the noise influence level, a projected component of the intermediate output to the parameters of the trained model ( see e.g. paragraphs 0003 and 0058 -0060 : like noted above, Speakman teaches obtaining a first set of activations, i.e. an intermediate output, from nodes in a hidden layer of a neural network for an input , adding noise to the input, and obtaining a second set of activations from the same nodes in the hidden layer of the neural network. Like further noted above, Speakman teaches determining a difference between the first set of activations and the second set of activations and using such a difference to determine whether the input is OOD data – see e.g. paragraphs 0003, 0024, 0048 and 0061-0063. Speakman suggests that such a procedure can be employed for multiple hidden layers of the neural network – see e.g. paragraphs 0055 and 0073. In such embodiments , the intermediate output of a hidden layer would be applied to a subsequent hidden layer to obtain the activations therefrom. Speakman thus teaches , inter alia : (a) receiving a first intermediate output from a hidden layer of a neural network in response to an input ; (b) subjecting the intermediate output to a subsequent hidden layer of the neural network to obtain a subsequent intermediate output ; (c) adding noise to the input; (d) receiving a first perturbed intermediate output from the hidden layer of the neural network in response to the noised input, and (e) subjecting the perturbed intermediate output to the subsequent layer of the neural network to obtain a subsequent perturbed intermediate output . The output of the subsequent hidden layer is a projected component of the intermediate output of the previous hidden layer to parameters of the trained model, i.e. to the parameters of the subsequent hidden layer. The original or perturbed output of the subsequent hidden layer can additionally or alternative b e considered a noise influence level of the intermediate output, which is used to determine whether or not the input is OOD data . ). As described above, it would have been obvious to modify the apparatus taught by Speakman so as to employ SVD like taught by Goetschalckx to one or more of the intermediate layers therein, and thereby remove the smallest, least important weights. As a result, the noise influence level (i.e. the original or perturbed output of the subsequent hidden layer when the intermediate output of the previous hidden layer is applied thereto) would be a projected component of the intermediate output to the parameter (i.e. to the parameters of the subsequent hidden layer that have a high contribution). Accordingly, the above-described combination of Speakman and Goetschalckx is further considered to teach an OOD data detection apparatus like that of claim 2. As per claim 3, it would have been obvious, as is described above, to modify the apparatus taught by Speakman so as to employ SVD like taught by Goetschalckx to one or more of the intermediate layers therein, and thereby remove the smallest, least important weights. Goetschalckx particularly teaches that applying SVD to neural network weights comprises, inter alia, determining a projection matrix based on a matrix decomposition of the neural network layer weights, whereby the output of the neural network layer is calculated by making the projection matrix act on the input (i.e. an input vector) to the neural network layer: SVD [17] applies singular value decomposition to a weight matrix . This factors it into two new weight matrices of which the columns of the first and rows of the second are sorted by decreasing singular value. These values can intuitively be considered as an importance metric. The rows and columns corresponding to the smallest, least important, singular values are then removed, resulting in two smaller weight matrices. This decreases the total number of weights to be stored, while sequential multiplication of a vector with these smaller weight matrices still closely approximates multiplication with the large original weight matrix . (Page 2 ; emphasis added ). It thus follows that the combination of Speakman and Goetschalckx entails determining a projection matrix based on a matrix decomposition of the parameter (i.e. based on a matrix decomposition of the neural network weights of a subsequent hidden layer), whereby the projected component (i.e. the output of the subsequent hidden layer) is calculated in part by making the projection matrix act on the intermediate output (i.e. the output of a previous hidden layer). Accordingly, the above-described combination of Speakman and Goetschalckx is further considered to teach an OOD data detection apparatus like that of claim 3. As per claim 4, it would have been obvious, as is described above, to modify the apparatus taught by Speakman so as to employ SVD like taught by Goetschalckx to one or more of the intermediate layers therein, and thereby remove the smallest, least important weights. As further described above (see the rejection for claim 3), Goetschalckx particularly teaches that applying SVD to neural network weights comprises, inter alia, determining a projection matrix based on a matrix decomposition of the neural network layer weights. It thus follows that the projection matrix is calculated by performing singular value decomposition (i.e. SVD) on a weight parameter constituting the trained model. Accordingly, the above-described combination of Speakman and Goetschalckx is further considered to teach an OOD data detection apparatus like that of claim 4. As per claim 5, it would have been obvious, as is described above, to modify the apparatus taught by Speakman so as to employ SVD like taught by Goetschalckx to one or more of the intermediate layers therein, and thereby remove the smallest, least important weights. As further described above (see the rejection for claim 3), Goetschalckx particularly teaches that applying SVD to neural network weights comprises, inter alia, determining a projection matrix based on a matrix decomposition of the neural network layer weights. Goetschalckx further teaches deleting, as components having a low contribution to the task, a matrix component that has a small singular value among matrix components included in the projection matrix: SVD [17] applies singular value decomposition to a weight matrix. This factors it into two new weight matrices of which the columns of the first and rows of the second are sorted by decreasing singular value. These values can intuitively be considered as an importance metric. The rows and columns corresponding to the smallest, least important, singular values are then removed, resulting in two smaller weight matrices . This decreases the total number of weights to be stored, while sequential multiplication of a vector with these smaller weight matrices still closely approximates multiplication with the large original weight matrix . (Page 2; emphasis added). Accordingly, the above -described combination of Speakman and Goetschalckx is further considered to teach an OOD data detection apparatus like that of claim 5. As per claim 6, it would have been obvious, as is described above, to modify the apparatus taught by Speakman so as to employ SVD like taught by Goetschalckx to one or more of the intermediate layers therein, and thereby remove the smallest, least important weights. As further described above (see the rejections for claims 3 and 4), Goetschalckx particularly teaches that applying SVD to neural network weights comprises, inter alia, determining a projection matrix based on a matrix decomposition of the neural network layer weights, and deleting, as components having a low contribution to the task, a matrix component that has a small singular value among matrix components included in the projection matrix. Goetschalckx further teaches, after deleting matrix components having small singular values and thereby compressing the neural network , retraining the compressed neural network until e.g. a target error rate is reached : The fourth algorithmic building block represents retraining of the compressed neural network . This can be applied after any of the previously described lossy compression blocks [e.g. SVD] in order to regain possible lost inference accuracy. Such retraining block can have stop conditions on a minimal improvement of training loss over a certain number of epochs, on a maximal number of epochs and/or on reaching a target error rate (see section3.3). (Page 2; emphasis added). Retraining the compressed neural network can be considered searching for matrix components that satisfy a predetermined condition (e.g. a target error rate) based on a change in a task performance (i.e. error rate) of the trained model when positions and/or a number of the matrix components included in the projection matrix are changed (i.e. when the small singular values therein are removed). Accordingly, the above-described combination of Speakman and Goetschalckx is further considered to teach an OOD data detection apparatus like that of claim 6. As per claim 7, Speakman further teaches injecting a noise to the intermediate output in an intermediate hidden layer of the trained model, wherein, to calculate the noise influence level, the processing circuit: (a) calculates a first intermediate output by applying a hidden layer later than the intermediate hidden layer to the intermediate output to which the noise is not injected ; (b) calculates a second intermediate output by applying the later hidden layer to the intermediate output to which the noise is injected; and (c) calculates a degree of variation between the first intermediate output and the second intermediate output as the noise influence level ( see e.g. paragraphs 0003 and 0058-0060: like noted above, Speakman teaches obtaining a first set of activations, i.e. an intermediate output, from nodes in a hidden layer of a neural network when given an input , adding noise to the input, and obtaining a second set of activations from the same nodes in the hidden layer of the neural network. Like further noted above, Speakman teaches determining a difference between the first set of activations and the second set of activations and using such a difference to determine whether the input is OOD data – see e.g. paragraphs 0003, 0024, 0048 and 0061-0063. Speakman suggests that such a procedure can be employed for multiple hidden layers of the neural network – see e.g. paragraphs 0055 and 0073. In such embodiments, the intermediate output of a hidden layer would be applied to a subsequent hidden layer to obtain the activations therefrom. Speakman thus teaches, inter alia: (a) receiving an intermediate output from an intermediate hidden layer of a neural network in response to an input; (b) calculating a first intermediate output by applying a subsequent hidden layer to the intermediate output; (c) adding noise to the input, which would thus inject noise in the intermediate output in the intermediate hidden layer; (d) calculating a second intermediate output by applying the subsequent hidden layer to the intermediate output to which noise is injected; and (e) calculating a degree of variation between the first intermediate output and the second intermediate output. This degree of variation can additionally or alternatively be considered a noise influence level like claimed, which is used to determine whether or not the input is OOD data.) . Accordingly, the above-described combination of Speakman and Goetschalckx is further considered to teach an OOD data detection apparatus like that of claim 7. As per claim 8, it would have been obvious, as is described above, to modify the apparatus taught by Speakman so as to employ SVD like taught by Goetschalckx to one or more of the intermediate layers therein, and thereby remove the smallest, least important weights. Goetschalckx particularly teaches that applying SVD to neural network weights comprises, inter alia, determining matrices of singular values based on a matrix decomposition of the neural network layer weights, and deleting, as components having a low contribution to the task, matrix components that have a small singular value among matrix components included in the matrices: SVD [17] applies singular value decomposition to a weight matrix . This factors it into two new weight matrices of which the columns of the first and rows of the second are sorted by decreasing singular value. These values can intuitively be considered as an importance metric. The rows and columns corresponding to the smallest, least important, singular values are then removed, resulting in two smaller weight matrices . This decreases the total number of weights to be stored, while sequential multiplication of a vector with these smaller weight matrices still closely approximates multiplication with the large original weight matrix . (Page 2; emphasis added). Goetschalckx thus teaches calculating a contribution (i.e. importance) to the task using a singular value of the parameter. Accordingly, the above-described combination of Speakman and Goetschalckx is further considered to teach an OOD data detection apparatus like that of claim 8. As per claim 9, it would have been obvious, as is described above, to modify the apparatus taught by Speakman so as to employ SVD like taught by Goetschalckx to one or more of the intermediate layers therein, and thereby remove the smallest, least important weights. As further described above (see the rejection for claim 3), Goetschalckx particularly teaches that applying SVD to neural network weights comprises, inter alia, determining matrices based on a matrix decomposition of the neural network layer weights. Goetschalckx further teaches deleting, as components having a low contribution to the task, a matrix component that has a small singular value in the parameter: SVD [17] applies singular value decomposition to a weight matrix. This factors it into two new weight matrices of which the columns of the first and rows of the second are sorted by decreasing singular value. These values can intuitively be considered as an importance metric. The rows and columns corresponding to the smallest, least important, singular values are then removed, resulting in two smaller weight matrices . This decreases the total number of weights to be stored, while sequential multiplication of a vector with these smaller weight matrices still closely approximates multiplication with the large original weight matrix . (Page 2; emphasis added). Accordingly, the above-described combination of Speakman and Goetschalckx is further considered to teach an OOD data detection apparatus like that of claim 9. As per claim 10, it would have been obvious, as is described above, to modify the apparatus taught by Speakman so as to employ SVD like taught by Goetschalckx to one or more of the intermediate layers therein, and thereby remove the smallest, least important weights. As further described above (see the rejection for claim 9), Goetschalckx particularly teaches that applying SVD to neural network weights comprises, inter alia, determining matrices based on a matrix decomposition of the neural network layer weights, and deleting, as components having a low contribution to the task, matrix components that have a small singular value among matrix components included in the matrices. Goetschalckx further teaches, after deleting matrix components having small singular values and thereby compressing the neural network, retraining the compressed neural network until e.g. a target error rate is reached: The fourth algorithmic building block represents retraining of the compressed neural net