DETAILED ACTIONNotice 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 .
Applicant Response to Official Action
The response filed on 6/10/2026 has been entered and made of record.
Acknowledgment
Claims 4, 7-13, 17, 22-27, 31, 34-41 were canceled.
Claims 1-2, 5, 14-15, 18, 28-29, and 32, amended on 6/10/2026, are acknowledged by the examiner.
Response to Arguments
Applicant’s arguments with respect to claims 1, 14, 28, and their dependent claims have been considered but they are moot in view of the new grounds of rejection necessitated by amendments initiated by the applicant. Examiner addresses the main arguments of the Applicant as below.
Regarding the drawing objection, in the amendment filed on 6/10/2026 the Applicant canceled claims 4, 17, and 31. As a result, the drawing objection is withdrawn.
Regarding the 35 U.S.C. 112(f) interpretation, the amendment filed on 6/10/2026 does not address the issue. Claim 14 recites “a controller configured to execute the executable instructions to result in performing the following steps”. It is noted that the controller in claim 14 can be a software implementation. In addition, it performs “steps”. As a result, the 35 U.S.C. 112(f) interpretation is maintained.
Regarding the 35 U.S.C. 112(b) rejections, the amendment filed on 6/10/2026 addresses the issue. As a result, the 35 U.S.C. 112(b) rejections are withdrawn.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) ELEMENT IN CLAIM FOR A COMBINATION.—An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as "configured to" or "so that"; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”), such as in claims 14-21 and 28-33, are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
If applicant intends to have these limitations interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to remove the structure, materials, or acts that performs the claimed function; or (2) present a sufficient showing that the claim limitation(s) does/do not recite sufficient structure, materials, or acts to perform the claimed function.
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 of this title, 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.
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 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.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under pre-AIA 35 U.S.C. 103(a) 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 under pre-AIA 35 U.S.C. 103(a), the examiner presumes that the subject matter of the various claims was commonly owned at the time any inventions covered therein were made absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and invention dates of each claim that was not commonly owned at the time a later invention was made in order for the examiner to consider the applicability of pre-AIA 35 U.S.C. 103(c) and potential pre-AIA 35 U.S.C. 102(e), (f) or (g) prior art under pre-AIA 35 U.S.C. 103(a).
Claims 1-2, 4-5, 14-15, 17-18, 28-29, 31-32 are rejected under 35 U.S.C. 103 as being unpatentable over Fukuda (US Patent 10,726,326 B2), (“Fukuda”), in view of Bouzaraa et al. (US Patent US 10,666,873 B2), (“Bouzaraa”).
Regarding claim 1, Fukuda meets the claim limitations, as follows:
A computer-implemented method (a computer implemented method) [Fukuda: col. 1, line 27] of extracting high-frequency features from data (One or more embodiments according to the present invention are directed to computer implemented methods, computer systems and computer program products for learning a neural network that has a plurality of filters for extracting local features from an input) [Fukuda: col. 2, line 47-51]; (In the describing embodiment, steps S202H-S204H for the high frequency components) [Fukuda: col. 11, line 52-53; Figs. 6-7], comprising:receiving a first dataset ((receives speech signals) [Fukuda: col. 4, line 31-32]; (At step S201, the preparing module 122 reads the speech data 140 and prepares the each group of the training data with class label for each component from the speech data 140 with associated transcriptions. In the describing embodiment, high and low frequency components of the speech data 130 are divided so as to generate two groups of the training data) [Fukuda: col. 11, line 34-40; Fig. 6]; in a training phase and using the first dataset (Referring back to FIG. 6, processing of steps S202-S204 is conducted for each group of the training data prepared in step 201. In the describing embodiment, steps S202H-S204H for the high frequency components) [Fukuda: col. 11, line 50-53; Figs. 6-7], applying frequency-based guidance during training to learnable filters in a neural network ((a computer implemented method performed by a computing device for learning a neural network that has a plurality of filters for extracting local features. The method includes calculating a plurality of projection parameter sets by analyzing one or more training data, in which the plurality of the projection parameter sets defines a projection of each training data into a new space and each projection parameter set has a same size as the filters in the neural network.) [Fukuda: col. 1, line 26-34; Figs. 6-7]; (Now referring to the series of FIGS. 1-7, there are shown computer systems and methods for learning a neural network that has a plurality of filters according to one or more embodiments of the present invention.) [Fukuda: col. 4, line 15-18; Figs 1-7]), wherein the learnable filters are caused to converge to eigenvectors of the frequency-based guidance (In a particular embodiment, calculation of the plurality of the projection parameter sets can be performed by finding the plurality of the projection parameter sets so as to maximize separability of data points each defining the acoustic features of the training data with different classes in the new space and to minimize variability of data points defining the acoustic feature of the training data with same class in the new space. More practically, the calculation may be conducted by estimating eigenvectors based on a Linear Discriminant Analysis (LDA) criterion, in which the projection parameter sets are obtained as the eigenvectors that form a projection (or LDA) matrix θ) [Fukuda: col. 6, line 38-49] and wherein the frequency-based guidance is directed to obtaining high eigenvalues associated with high-frequency eigenvectors (In particular embodiments with LDA criterion, the scaling factor is an eigenvalue corresponding to the eigenvector. The eigenvalue can tell about magnitude of distortion of transformation defined by the associated eigenvector.) [Fukuda: col. 6, line 55-59]; and in a detect phase (extracts acoustic features from the received speech signals) [Fukuda: col. 4, line 32-33; Figs 6-7], using the high-frequency eigenvectors (As shown in FIG. 5B, among the K localized filters in the convolutional layer of the neural network 150, initial weights for P localized filters are replaced by the P eigenvectors (θ1 , θ2, ... , θp)- Weights for remaining (K-P) localized filters (Φi, ... , ΦK-P) other than the weights of the localized filters that are replaced by using the eigenvectors, can be set with random initial values) [Fukuda: col. 9, line 49-55; Figs 5A-7] to extract high-frequency features from a second dataset (the initializing module 124 may select the predetermined number Q of the eigenvectors for the high frequency component) [Fukuda: col. 12, line 21-23; Figs 6-7].
Fukuda does not explicitly disclose the following claim limitations (Emphasis added).
during training to learnable filters in a neural network, wherein the learnable filters are caused to converge to.
However, in the same field of endeavor Bouzaraa further discloses the deficient claim limitations as follows:
a diagonal degree matrix (The network aims at iteratively estimating the set of
parameters of each layer (for example, filter weights W and biases B) during the training phase so that the created exposure conversion model best fits the exposure ratio of the training data. Understandably, if many networks with different architectures are trained, the training phase will learn the parameters of each network separately. During each iteration, the network updates the set of filter weights W = {W_1, W_2, . . . , W_L} and biases B= {B_1, B_2, ... , B_L} of the L convolutional layers, according to the current loss value. The loss is computed by comparing the predicted mapped images, based on the current instances of filter weights and biases, and the labels (for example, ground-truths). Next, the error loss is back-propagated through the network and used to update the network parameters using stochastic gradient and a user pre-defined learning rate ή) [Bouzaraa: col. 12, line 45-61].
It would have been obvious to one with an ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fukuda with Bouzaraa to program the system to implement a method of Bouzaraa.
Therefore, the combination of Fukuda with Bouzaraa will enable the system to such as increase the quality of a photograph, or to detect the geometry of the scene [Bouzaraa: col. 1, line 63-64, col. 2, line 32-37].
Regarding claims 2, 15, and 29, Fukuda meets the claim limitations as set forth in claims 1, 14, 28. Fukuda further meets the claim limitations as follow.
normalizing the high eigenvalues to values in ranging from 0 to 1 ((In particular embodiments with LDA criterion, the scaling factor is an eigenvalue corresponding to the eigenvector) [Fukuda: col. 6, line 55-57]; (the second example (Example 2) using top 64 eigenvectors, that was in a range from 4/10 to 7/10) [Fukuda: col. 14, line 46-48] – Note: Range from 4/10 to 7/10 is within the range from 0 to 1).
Regarding claims 4, 17, and 31, Fukuda meets the claim limitations as set forth in claims 1, 14, 28. Fukuda further meets the claim limitations as follow.
defining an operator associated with the high eigenvalues ((FIG. 5A schematically shows a scatter plot of training data projected on first two eigenvectors of the projection matrix θ = {θ1, θ2, ... θn, ... θNxM} obtained based on the LDA. The plurality of the projection parameter set, that is the projection matrix θ, defines a projection of the acoustic features of each training data into a new space depending on
class) [Fukuda: col. 6, line 55-57; col. 22, line 25-31; Figs 5A-B]; (In a particular embodiment, the initializing module 124 first sorts the eigenvectors by ascending order, and selects the predetermined number P of the eigenvectors with largest eigenvalues. In particular embodiments, the predetermined number P may be any value not more than the number of the filter K) [Fukuda: col. 9, line 27-32; Fig. 3]; (Training data were prepared by sliding a local window having 9 frames x 9 frequency bands along with a direction of frequency axis over the speech data. As for examples and comparative example, the neural network was learned by the learning process shown in FIG. 3 with various predetermined numbers P. The predetermined number P was set to be 0 (comparative example), 32 (example 1), 64 (example 2) and 81 (example 3). The localized filters in the first convolutional layer were used as targets of the novel parameter initialization. The weights for P localized filters were replaced by calculated P eigenvectors with largest eigenvalues while others were initialized with random values) [Fukuda: col. 13, line 60 – col. 14, line 6; Fig. 3]).
Regarding claims 5, 18, and 32, Fukuda meets the claim limitations as set forth in claims 1, 14, 28. Fukuda further meets the claim limitations as follow.
controlling a spectrum of the learnable filters (Log mel spectrum with 40 bands was used as acoustic feature input. A class label was aligned to each center frame by the forced alignment technique based on standard GMM/HHM system. 5000 of quin-phone HMM states were used as the class labels for the parameter initialization. Training data were prepared by sliding a local window having 9 frames x 9 frequency bands along with a direction of frequency axis over the speech data. As for examples and comparative example, the neural network was learned by the learning process shown in FIG. 3 with various predetermined numbers P. The predetermined number P was set to be 0 (comparative example), 32 (example 1), 64 (example 2) and 81 (example 3). The localized filters in the first convolutional layer were used as targets of the novel parameter initialization. The weights for P localized filters were replaced by calculated P eigenvectors with largest eigenvalues while others were initialized with random values) [Fukuda: col. 13, line 55 – col. 14, line 6; Fig. 3]).
Regarding claim 14, Fukuda meets the claim limitations, as follows:
A system (the speech recognition system) [Fukuda: col. 4, line 30; Figs. 1-2] for extracting high-frequency features from data (One or more embodiments according to the present invention are directed to computer implemented methods, computer systems and computer program products for learning a neural network that has a plurality of filters for extracting local features from an input) [Fukuda: col. 2, line 47-51]; (In the describing embodiment, steps S202H-S204H for the high frequency components) [Fukuda: col. 11, line 52-53; Figs. 6-7], comprising:
a neural network (The neural network 150 depicted in FIG. 1 includes an input layer 152, one or more convolutional layers 154, one or more fully-connected layers 156 and an output layer 158. The neural network 150 shown in FIG. 1 has a typical
configuration of the aforementioned CNN.) [Fukuda: col. 2, line 58-62; Figs. 1-2] to receive a first dataset ((receives speech signals) [Fukuda: col. 4, line 31-32]; (At step S201, the preparing module 122 reads the speech data 140 and prepares the each group of the training data with class label for each component from the speech data 140 with associated transcriptions. In the describing embodiment, high and low frequency components of the speech data 130 are divided so as to generate two groups of the training data) [Fukuda: col. 11, line 34-40; Fig. 6]); a memory for storing data and executable instructions (The computer system includes a memory tangibly storing the program instructions) [Fukuda: col. 1, line 41-43]; and a controller configured to execute the executable instructions to result in performing the following steps (The computer system is configured to calculate a plurality of projection parameter sets by analyzing one or more training data, in which the plurality of the projection parameter sets defines a projection of each training data into new space and each projection parameter set has a same size as the filters in the neural network. The computer system is further configured to set at least part of the plurality of the projection parameter sets as initial parameters of at least part of the plurality of the filters in the neural network for training) [Fukuda: col. 1, line 44-53]:in a training phase and using the first dataset (Referring back to FIG. 6, processing of steps S202-S204 is conducted for each group of the training data prepared in step 201. In the describing embodiment, steps S202H-S204H for the high frequency components) [Fukuda: col. 11, line 50-53; Figs. 6-7], applying frequency-based guidance during training to learnable filters in the neural network ((a computer implemented method performed by a computing device for learning a neural network that has a plurality of filters for extracting local features. The method includes calculating a plurality of projection parameter sets by analyzing one or more training data, in which the plurality of the projection parameter sets defines a projection of each training data into a new space and each projection parameter set has a same size as the filters in the neural network) [Fukuda: col. 1, line 26-34; Figs. 6-7] ; (Now referring to the series of FIGS. 1-7, there are shown computer systems and methods for learning a neural network that has a plurality of filters according to one or more embodiments of the present invention.) [Fukuda: col. 4, line 15-18; Figs 1-7]), wherein the learnable filters are caused to converge to eigenvectors of the frequency-based guidance (In a particular embodiment, calculation of the plurality of the projection parameter sets can be performed by finding the plurality of the projection parameter sets so as to maximize separability of data points each defining the acoustic features of the training data with different classes in the new space and to minimize variability of data points defining the acoustic feature of the training data with same class in the new space. More practically, the calculation may be conducted by estimating eigenvectors based on a Linear Discriminant Analysis (LDA) criterion, in which the projection parameter sets are obtained as the eigenvectors that form a projection (or LDA) matrix θ) [Fukuda: col. 6, line 38-49] and wherein the frequency-based guidance is directed to obtaining high eigenvalues associated with high-frequency eigenvectors (In particular embodiments with LDA criterion, the scaling factor is an eigenvalue corresponding to the eigenvector. The eigenvalue can tell about magnitude of distortion of transformation defined by the associated eigenvector.) [Fukuda: col. 6, line 55-59]; and in a detect phase (extracts acoustic features from the received speech signals) [Fukuda: col. 4, line 32-33; Figs 6-7], using the high-frequency eigenvectors (As shown in FIG. 5B, among the K localized filters in the convolutional layer of the neural network 150, initial weights for P localized filters are replaced by the P eigenvectors (θ1 , θ2, ... , θp)- Weights for remaining (K-P) localized filters (Φi, ... , ΦK-P) other than the weights of the localized filters that are replaced by using the eigenvectors, can be set with random initial values) [Fukuda: col. 9, line 49-55; Figs 5A-7] to extract high-frequency features from a second dataset (the initializing module 124 may select the predetermined number Q of the eigenvectors for the high frequency component) [Fukuda: col. 12, line 21-23; Figs 6-7].
Fukuda does not explicitly disclose the following claim limitations (Emphasis added).
during training to learnable filters in a neural network, wherein the learnable filters are caused to converge to.
However, in the same field of endeavor Bouzaraa further discloses the deficient claim limitations as follows:
a diagonal degree matrix (The network aims at iteratively estimating the set of
parameters of each layer (for example, filter weights W and biases B) during the training phase so that the created exposure conversion model best fits the exposure ratio of the training data. Understandably, if many networks with different architectures are trained, the training phase will learn the parameters of each network separately. During each iteration, the network updates the set of filter weights W = {W_1, W_2, . . . , W_L} and biases B= {B_1, B_2, ... , B_L} of the L convolutional layers, according to the current loss value. The loss is computed by comparing the predicted mapped images, based on the current instances of filter weights and biases, and the labels (for example, ground-truths). Next, the error loss is back-propagated through the network and used to update the network parameters using stochastic gradient and a user pre-defined learning rate ή) [Bouzaraa: col. 12, line 45-61].
It would have been obvious to one with an ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fukuda with Bouzaraa to program the system to implement a method of Bouzaraa.
Therefore, the combination of Fukuda with Bouzaraa will enable the system to such as increase the quality of a photograph, or to detect the geometry of the scene [Bouzaraa: col. 1, line 63-64, col. 2, line 32-37].
Regarding claim 28, Fukuda meets the claim limitations, as follows:
A non-transitory computer-readable medium including instructions (a memory tangibly storing the program instructions) [Fukuda: col. 1, line 42-43], that when executed by a processor (The computer system is configured to calculate a plurality of projection parameter sets by analyzing one or more training data, in which the plurality of the projection parameter sets defines a projection of each training data into new space and each projection parameter set has a same size as the filters in the neural network. The computer system is further configured to set at least part of the plurality of the projection parameter sets as initial parameters of at least part of the plurality of the filters in the neural network for training) [Fukuda: col. 1, line 44-53], causes a system for extracting high-frequency features from data to perform the following steps (One or more embodiments according to the present invention are directed to computer implemented methods, computer systems and computer program products for learning a neural network that has a plurality of filters for extracting local features from an input) [Fukuda: col. 2, line 47-51]; (In the describing embodiment, steps S202H-S204H for the high frequency components) [Fukuda: col. 11, line 52-53; Figs. 6-7]:
receiving a first dataset ((receives speech signals) [Fukuda: col. 4, line 31-32]; (At step S201, the preparing module 122 reads the speech data 140 and prepares the each group of the training data with class label for each component from the speech data 140 with associated transcriptions. In the describing embodiment, high and low frequency components of the speech data 130 are divided so as to generate two groups of the training data) [Fukuda: col. 11, line 34-40; Fig. 6]; in a training phase and using the first dataset (Referring back to FIG. 6, processing of steps S202-S204 is conducted for each group of the training data prepared in step 201. In the describing embodiment, steps S202H-S204H for the high frequency components) [Fukuda: col. 11, line 50-53; Figs. 6-7], applying frequency-based guidance during training to learnable filters in a neural network ((a computer implemented method performed by a computing device for learning a neural network that has a plurality of filters for extracting local features. The method includes calculating a plurality of projection parameter sets by analyzing one or more training data, in which the plurality of the projection parameter sets defines a projection of each training data into a new space and each projection parameter set has a same size as the filters in the neural network.) [Fukuda: col. 1, line 26-34; Figs. 6-7]; (Now referring to the series of FIGS. 1-7, there are shown computer systems and methods for learning a neural network that has a plurality of filters according to one or more embodiments of the present invention.) [Fukuda: col. 4, line 15-18; Figs 1-7]), wherein the learnable filters are caused to converge to eigenvectors of the frequency-based guidance (In a particular embodiment, calculation of the plurality of the projection parameter sets can be performed by finding the plurality of the projection parameter sets so as to maximize separability of data points each defining the acoustic features of the training data with different classes in the new space and to minimize variability of data points defining the acoustic feature of the training data with same class in the new space. More practically, the calculation may be conducted by estimating eigenvectors based on a Linear Discriminant Analysis (LDA) criterion, in which the projection parameter sets are obtained as the eigenvectors that form a projection (or LDA) matrix θ) [Fukuda: col. 6, line 38-49] and wherein the frequency-based guidance is directed to obtaining high eigenvalues associated with high-frequency eigenvectors (In particular embodiments with LDA criterion, the scaling factor is an eigenvalue corresponding to the eigenvector. The eigenvalue can tell about magnitude of distortion of transformation defined by the associated eigenvector.) [Fukuda: col. 6, line 55-59]; and in a detect phase (extracts acoustic features from the received speech signals) [Fukuda: col. 4, line 32-33; Figs 6-7], using the high-frequency eigenvectors (As shown in FIG. 5B, among the K localized filters in the convolutional layer of the neural network 150, initial weights for P localized filters are replaced by the P eigenvectors (θ1 , θ2, ... , θp)- Weights for remaining (K-P) localized filters (Φi, ... , ΦK-P) other than the weights of the localized filters that are replaced by using the eigenvectors, can be set with random initial values) [Fukuda: col. 9, line 49-55; Figs 5A-7] to extract high-frequency features from a second dataset (the initializing module 124 may select the predetermined number Q of the eigenvectors for the high frequency component) [Fukuda: col. 12, line 21-23; Figs 6-7].
Fukuda does not explicitly disclose the following claim limitations (Emphasis added).
during training to learnable filters in a neural network, wherein the learnable filters are caused to converge to.
However, in the same field of endeavor Bouzaraa further discloses the deficient claim limitations as follows:
a diagonal degree matrix (The network aims at iteratively estimating the set of
parameters of each layer (for example, filter weights W and biases B) during the training phase so that the created exposure conversion model best fits the exposure ratio of the training data. Understandably, if many networks with different architectures are trained, the training phase will learn the parameters of each network separately. During each iteration, the network updates the set of filter weights W = {W_1, W_2, . . . , W_L} and biases B= {B_1, B_2, ... , B_L} of the L convolutional layers, according to the current loss value. The loss is computed by comparing the predicted mapped images, based on the current instances of filter weights and biases, and the labels (for example, ground-truths). Next, the error loss is back-propagated through the network and used to update the network parameters using stochastic gradient and a user pre-defined learning rate ή) [Bouzaraa: col. 12, line 45-61].
It would have been obvious to one with an ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fukuda with Bouzaraa to program the system to implement a method of Bouzaraa.
Therefore, the combination of Fukuda with Bouzaraa will enable the system to such as increase the quality of a photograph, or to detect the geometry of the scene [Bouzaraa: col. 1, line 63-64, col. 2, line 32-37].
Claims 3, 16, and 30 are rejected under 35 U.S.C. 103 as being unpatentable over Fukuda (US Patent 10,726,326 B2), (“Fukuda”), in view of Bouzaraa et al. (US Patent US 10,666,873 B2), (“Bouzaraa”), in view of Chen et al. (CN Patent Application Publication CN106323636A), (“Chen”).
Regarding claims 3, 16, and 30, Fukuda and Bouzaraa meet claim limitations as set forth in claims 1, 14, and 28. Fukuda and Bouzaraa do not explicitly disclose the following claim limitations (Emphasis added).
comprising normalizing a frequency spectrum to values ranging from 0 to 1.
However, in the same field of endeavor Chen further discloses the claim limitations and the deficient claim limitations, as follows:
comprising normalizing a frequency spectrum to values ranging from 0 to 1 (calculate the spectrum and normalize it to make the amplitude range [0, 1], and obtain training samples and test samples) [Chen: page 3].
It would have been obvious to one with an ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fukuda and Bouzaraa with Chen to program the system to implement a method of Chen.
Therefore, the combination of Fukuda and Bouzaraa with Chen will enable the system to improve the robustness of feature learning, through layer-by-layer unsupervised adaptive learning and supervised fine-tuning of the original input complex data through multi-layer sparse automatic coding [Chen: Abstract].
Claims 6, 19-21, and 33 are rejected under 35 U.S.C. 103 as being unpatentable over Fukuda (US Patent 10,726,326 B2), (“Fukuda”), in view of Bouzaraa et al. (US Patent US 10,666,873 B2), (“Bouzaraa”), in view of Harris et al. (US Patent US 12,621,310 B2), (“Harris”).
Regarding claims 6, 19, and 33, Fukuda meets the claim limitations as set forth in claims 1, 14, and 28. Fukuda further meets the claim limitations as follow.
generating a normalized NxN (As shown in FIG. 4, the local window 200 with a size of N frames x N frequency bands, which has a size identical to that of the target localized filters cp, in the neural network, are slid along the direction of the frequency axis at the central frame to generate training data each having an NxN dimensional input feature vector and a class label associated with the central frame) [Fukuda: col. 7, line 65 – col. 8, line 4] Laplacian Matrix for at least one learnable filter ((a computer implemented method performed by a computing device for learning a neural network that has a plurality of filters for extracting local features. The method includes calculating a plurality of projection parameter sets by analyzing one or more training data, in which the plurality of the projection parameter sets defines a projection of each training data into a new space and each projection parameter set has a same size as the filters in the neural network.) [Fukuda: col. 1, line 26-34; Figs. 6-7]; (Now referring to the series of FIGS. 1-7, there are shown computer systems and methods for learning a neural network that has a plurality of filters according to one or more embodiments of the present invention.) [Fukuda: col. 4, line 15-18; Figs 1-7]).
Fukuda and Bouzaraa do not explicitly disclose the following claim limitations (Emphasis added).
Laplacian Matrix.
However, in the same field of endeavor Harris further discloses the claim limitations and the deficient claim limitations, as follows:
Laplacian Matrix (In some embodiments, an adjacency matrix and a degree matrix can be used together to construct a Laplacian matrix of a graph. A degree matrix may be a diagonal matrix. For example, given a graph G=(V, E) where V are the vertices (i.e., nodes) and E are the edges, and where the magnitude of V is equal to the total number of nodes n, the degree matrix D for the graph G can be a NxN diagonal matrix) [Harris: col. 5, line 32-49].
It would have been obvious to one with an ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fukuda and Bouzaraa with Harris to program the system to implement a method of Harris.
Therefore, the combination of Fukuda and Bouzaraa with Harris will enable the system to reduce the number of total nodes in the network data, thus improving downstream computation times while retaining accuracy of the data [Harris: col. 33, line 54-56].
Regarding claim 20, Fukuda meets the claim limitations as set forth in claim 1. Fukuda further meets the claim limitations as follow.
generating (As shown in FIG. 4, the local window 200 with a size of N frames x N frequency bands, which has a size identical to that of the target localized filters cp, in the neural network, are slid along the direction of the frequency axis at the central frame to generate training data each having an NxN dimensional input feature vector and a class label associated with the central frame) [Fukuda: col. 7, line 65 – col. 8, line 4] an adjacency matrix.
Fukuda and Bouzaraa do not explicitly disclose the following claim limitations (Emphasis added).
an adjacency matrix.
However, in the same field of endeavor Harris further discloses the claim limitations and the deficient claim limitations, as follows:
an adjacency matrix (In some embodiments, an adjacency matrix and a degree matrix can be used together to construct a Laplacian matrix of a graph. A degree matrix may be a diagonal matrix. For example, given a graph G=(V, E) where V are the vertices (i.e., nodes) and E are the edges, and where the magnitude of V is equal to the total number of nodes n, the degree matrix D for the graph G can be a NxN diagonal matrix) [Harris: col. 5, line 32-49].
It would have been obvious to one with an ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fukuda and Bouzaraa with Harris to program the system to implement a method of Harris.
Therefore, the combination of Fukuda and Bouzaraa with Harris will enable the system to reduce the number of total nodes in the network data, thus improving downstream computation times while retaining accuracy of the data [Harris: col. 33, line 54-56].
Regarding claim 21, Fukuda meets the claim limitations as set forth in claim 1. Fukuda further meets the claim limitations as follow.
generating (As shown in FIG. 4, the local window 200 with a size of N frames x N frequency bands, which has a size identical to that of the target localized filters cp, in the neural network, are slid along the direction of the frequency axis at the central frame to generate training data each having an NxN dimensional input feature vector and a class label associated with the central frame) [Fukuda: col. 7, line 65 – col. 8, line 4] a diagonal degree matrix.
Fukuda and Bouzaraa do not explicitly disclose the following claim limitations (Emphasis added).
a diagonal degree matrix.
However, in the same field of endeavor Harris further discloses the claim limitations and the deficient claim limitations as follows:
a diagonal degree matrix (In some embodiments, an adjacency matrix and a degree matrix can be used together to construct a Laplacian matrix of a graph. A degree matrix may be a diagonal matrix. For example, given a graph G=(V, E) where V are the vertices (i.e., nodes) and E are the edges, and where the magnitude of V is equal to the total number of nodes n, the degree matrix D for the graph G can be a NxN diagonal matrix) [Harris: col. 5, line 32-49].
It would have been obvious to one with an ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fukuda and Bouzaraa with Harris to program the system to implement a method of Harris.
Therefore, the combination of Fukuda and Bouzaraa with Harris will enable the system to reduce the number of total nodes in the network data, thus improving downstream computation times while retaining accuracy of the data [Harris: col. 33, line 54-56].
Reference Notice
Additional prior arts, included in the Notice of Reference Cited, made of record and not relied upon is considered pertinent to applicant's disclosure.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action.
Contact Information
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Philip Dang whose telephone number is (408) 918-7529. The examiner can normally be reached on Monday-Thursday between 8:30 am - 5:00 pm (PST).
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, Sath Perungavoor can be reached on 571-272-7455. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000. /Philip P. Dang/Primary Examiner, Art Unit 2488
Prosecution Insights