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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 06/08/2022 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being 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 applicant regards as his invention.
Claims 2, 4-5, 7, and 9-10 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.
Claims 2 and 7 recite the limitation: “wherein if the current loop count does not reach the preset maximum loop count”. It is unclear what steps should occur when the current loop count does not reach the reset maximum loop. The claim recites a condition but does not specify what steps are performed when the condition is met. The limitation is incomplete and is not given patentable weight.
Claims 4 and 9 recite the variable ‘T’ in “(
U
1
T
α
i
)” and “(
V
1
T
β
i
)”. The variable ‘T is undefined. For examination purposes, Examiner interprets the variable ‘T’ as being equal to 1.
Claims 4 and 9 recite the limitations "the first auxiliary vector sets" and “the second auxiliary vector sets” in “wherein the first hidden layer includes a first fully-connected layer and a first activation function layer, wherein the first fully-connected layer output a first calculation result (Ufa1) according to the first weight matrices (Uk+1) and the first auxiliary vector sets (ai)” and “wherein the further first fully-connected layer output a further first calculation result ( Vf1)according to the third weight matrices (Vlk") and the second auxiliary vector sets (p3)”, respectively. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, examiner interprets the limitations as “a plurality of first auxiliary vector sets" and “a plurality of second auxiliary vector sets”.
Claims 5 and 10 recite the equation: “
m
i
,
j
=
u
i
T
v
j
”. The variables ‘u’, ‘v’, and ‘T’ are not defined. For examination purposes, Examiner interprets the variables as being equal to 1.
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 1-2, 5-7, and 10-12 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1
According to the first part of the analysis, in the instant case, claims 1-10 are directed to a method and claims 11-12 are directed to a device comprising at least a processor. Thus, each of the claims falls within one of the four statutory categories (i.e. process, machine, manufacture, or composition of matter).
Claim 1 recites:
Step 2A, Prong 1
“identifying, by the processor, a plurality of first entries (xij) of the incomplete matrix according to the object data” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can identify entries in an incomplete matrix. (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
“obtaining, by the processor, a plurality of second entries (mij) of a recovered complete matrix corresponding to the incomplete matrix from the analysis model, wherein values of the second entries are determined as original values of the first entries of the incomplete matrix, such that incorrect data in the incomplete matrix is recovered” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can determine predicted values of a matrix with missing values based on the original values. (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
Step 2A, Prong 2
“by a processor of the electronic device” (mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
“receiving, by a processor of the electronic device, object data, wherein the object data comprises an incomplete matrix” (insignificant extra-solution activity)
“inputting, by the processor, the first entries (xij) and a preset maximum loop count (Kmax) into an executed analysis model using Bi- Branch Neural Network (BiBNN) algorithm” (insignificant extra-solution activity)
This judicial exception is not integrated into a practical application.
Step 2B
“by a processor of the electronic device” (mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
“receiving, by a processor of the electronic device, object data, wherein the object data comprises an incomplete matrix” (This step appears to be directed to transmitting or receiving information, which is well-understood, routine, and conventional. i. Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); See MPEP 2106.05 (d) (II).)
“inputting, by the processor, the first entries (xij) and a preset maximum loop count (Kmax) into an executed analysis model using Bi- Branch Neural Network (BiBNN) algorithm” (This step appears to be directed to transmitting or receiving information, which is well-understood, routine, and conventional. i. Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); See MPEP 2106.05 (d) (II).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 2 recites:
Step 2A, Prong 1
“generating a plurality of first auxiliary vector sets (
α
i
) corresponding to a first branch of the BiBNN and a plurality of second auxiliary vector sets (
β
j
) corresponding to a second branch of the BiBNN according to the first entries (xij)” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can generate vectors comprising numerical values corresponding to branches of a network. (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
“initializing a first weight matrices (
U
1
1
) and a second weight matrices (
U
2
1
) of the first branch, a third weight matrices (
V
1
1
) and a fourth weight matrices (
V
2
1
) of the second branch, and a loop count (k) according to the first entries (xij)” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can set matrices and a loop count to an initial value. (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
“calculating current first weight matrices (
U
1
k
+
1
) and current second weight matrices (
U
2
k
+
1
) in the current loop according to previous first weight matrices (
U
1
k
) and previous second weight matrices (
U
2
k
) in the previous loop, and calculating current third weight matrices (
V
1
k
+
1
) and current fourth weight matrices (
V
2
k
+
1
) in the current loop according to previous third weight matrices (
V
1
k
) and previous fourth weight matrices (
V
2
k
) in the previous loop” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
“determining whether the current loop count reaches the preset maximum loop count (Kmax) wherein if the current loop count does not reach the preset maximum loop count” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can determine when a loop has reached its preset maximum loop count. (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
“wherein if the current loop count reaches the preset maximum loop count, outputting the first outputs (ui) by the first branch according to the first auxiliary vector sets (
α
i
), the current first weight matrices (
U
1
k
m
a
x
) and the current second weight matrices (
U
2
k
m
a
x
), and outputting the second outputs (vj) by the second branch according to the second auxiliary vector sets (
β
j
), the current third weight matrices (
V
1
k
m
a
x
) and the current fourth weight matrices (
V
1
k
m
a
x
)” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
“calculating the values of the second entries (mij) of the recovered complete matrix according to the first outputs (ui) and the second outputs (vj)” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
Step 2A, Prong 2
“inputting the first auxiliary vector sets (ai) to the first branch, and inputting the second auxiliary vector sets (p) to the second branch according to the first entries (xij)” (insignificant extra-solution activity)
This judicial exception is not integrated into a practical application.
Step 2B
“inputting the first auxiliary vector sets (ai) to the first branch, and inputting the second auxiliary vector sets (p) to the second branch according to the first entries (xij)” (This step appears to be directed to transmitting or receiving information, which is well-understood, routine, and conventional. i. Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); See MPEP 2106.05 (d) (II).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 5 recites:
Step 2A, Prong 1
“wherein the second entries (mij) of the complete matrix are calculated by the equation below:
PNG
media_image1.png
12
59
media_image1.png
Greyscale
” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 6 recites:
Step 2A, Prong 1
“identifying, by the processor, values of first entries (xij) of the matrix according to the object data” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can identify entries in an incomplete matrix. (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
“obtaining, by the processor, a plurality of second entries (mij) of from the analysis model, wherein values of the second entries are determined as original ratings of the matrix, such that unknown ratings of the part of the first entries are predicted” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can determine predicted values of a matrix with missing values based on the original values. (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
“regarding each user ID, selecting, by the processor, one or more item IDs having ratings higher than a rating threshold, so as to determine one or more recommendation items corresponding to the selected item IDs for user corresponding to each user ID” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can select an item for recommendation by determining the rating for the item meets a certain threshold. (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
Step 2A, Prong 2
“by a processor of the electronic device” (mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
“receiving, by a processor of the electronic device, object data, wherein the object data comprises a matrix, wherein rows of the matrix correspond to user IDs of the users respectively, columns of the matrix correspond to item IDs of items respectively, and each entry of the matrix indicates a rating related to corresponding item ID and corresponding user ID” (insignificant extra-solution activity)
“inputting, by the processor, the entries (xij) and a preset maximum loop count (Kmax) into an executed analysis model using Bi-Branch Neural Network (BiBNN) algorithm” (insignificant extra-solution activity)
This judicial exception is not integrated into a practical application.
Step 2B
“by a processor of the electronic device” (mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
“receiving, by a processor of the electronic device, object data, wherein the object data comprises a matrix, wherein rows of the matrix correspond to user IDs of the users respectively, columns of the matrix correspond to item IDs of items respectively, and each entry of the matrix indicates a rating related to corresponding item ID and corresponding user ID” (This step appears to be directed to transmitting or receiving information, which is well-understood, routine, and conventional. i. Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); See MPEP 2106.05 (d) (II).)
“inputting, by the processor, the entries (xij) and a preset maximum loop count (Kmax) into an executed analysis model using Bi-Branch Neural Network (BiBNN) algorithm” (This step appears to be directed to transmitting or receiving information, which is well-understood, routine, and conventional. i. Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); See MPEP 2106.05 (d) (II).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 8 recites:
Step 2A, Prong 1 & Step 2A, Prong 2 & 2B
See rejection of claim 3. Same rationale applies.
Claim 9 recites:
Step 2A, Prong 1 & Step 2A, Prong 2 & 2B
See rejection of claim 4. Same rationale applies.
Claim 10 recites:
Step 2A, Prong 1 & Step 2A, Prong 2 & 2B
See rejection of claim 5. Same rationale applies.
Claim 11 recites:
Step 2A, Prong 1
See rejection of claim 1. Same rationale applies.
Step 2A, Prong 2 & 2B
The claim recites additional elements (“An electronic device for performing data recovering operation, comprising: a processor, configured to execute machine instructions to implement a computer-implemented method”). (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
This judicial exception is not integrated into a practical application.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 12 recites:
Step 2A, Prong 1
See rejection of claim 6. Same rationale applies.
Step 2A, Prong 2 & 2B
The claim recites additional elements (“An electronic device for performing data recovering operation, comprising: a processor, configured to execute machine instructions to implement a computer-implemented method”). (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
This judicial exception is not integrated into a practical application.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim Rejections - 35 USC § 103
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.
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.
Claims 1, 6, and 11-12 are rejected under 35 U.S.C. 103 as being unpatentable over Lee et al. (“How to impute missing ratings? Claims, solution, and its application to collaborative filtering.”) in view of Song et al. (US-20220350329-A1).
Regarding Claim 1,
Lee (“How to Impute Missing Ratings? Claims, Solution, and Its Application to Collaborative Filtering”) teaches a computer-implemented method for performing data recovering operation by an electronic device, comprising:
receiving, by a processor of the electronic device, object data, wherein the object data comprises an incomplete matrix (figure 1; pg. 788, section 4.4; “User vectors and item vectors are rows and columns, respectively, of rating matrix R, where missing ratings are treated as 0s.”);
identifying, by the processor, a plurality of first entries (xi,j) of the incomplete matrix according to the object data (figure 1; pg. 788, section 4.4; “In our case, x is a pair of user and item vectors. User vectors and item vectors are rows and columns, respectively, of rating matrix R, where missing ratings are treated as 0s.” values in the matrix R has first entries.);
inputting, by the processor, the first entries (xij) into an executed analysis model using Bi-Branch Neural Network (BiBNN) algorithm (figure 1; pg. 784; “First, it infers the preuse preferences on all user-item pairs based on the original rating matrix R. Second, it extracts the features (i.e., characteristics) of the user and the item using the variational autoencoder (VAE). Third, it trains the model that infers the post-use preference based on the pre-use preference and the extracted features using the multi-layer perceptron (MLP).); and
obtaining, by the processor, a plurality of second entries (mij) of a recovered complete matrix corresponding to the incomplete matrix from the analysis model, wherein values of the second entries are determined as original values of the first entries of the incomplete matrix, such that incorrect data in the incomplete matrix is recovered (pg. 783; “Data imputation aims to infer missing ratings and to impute them to the original rating matrix.” pg. 784; “First, it infers the preuse preferences on all user-item pairs based on the original rating matrix R. Second, it extracts the features (i.e., characteristics) of the user and the item using the variational autoencoder (VAE). Third, it trains the model that infers the post-use preference based on the pre-use preference and the extracted features using the multi-layer perceptron (MLP). Finally, we infer the post-use preference for every missing rating (i.e., ru,i = null) using the trained model, and impute the inferred post-use preferences to R.”).
Lee does not explicitly disclose
a preset maximum loop count (Kmax)
However, Song (US 20220350329 A1) teaches
inputting, by the processor, the first entries (xij) and a preset maximum loop count (Kmax) into an executed analysis model using Bi-Branch Neural Network (BiBNN) algorithm (fig. 2; para [0024] “S600: presetting a maximum number of iterations, taking the actual path in the I moments and a corresponding predicted path as an input of a neural network model to train the error model E, updating parameters of the model through back propagation (BP) during training, and stopping training when E≤e.sup.−5 to obtain a well-trained odometry error model;” figure two shows a neural network with at least two branches similar to figure 5 of applicant’s drawings.)
Lee and Song are analogous because they are directed to the field of neural networks.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the variational autoencoder of Lee with the preset maximum number of iterations of Song.
Doing so would allow for terminating the training of the neural network when a maximum number of iterations is met (Song para [0024]).
Regarding Claim 6,
Lee (“How to Impute Missing Ratings? Claims, Solution, and Its Application to Collaborative Filtering”) teaches a computer-implemented method for determining one or more recommendation items from one or more items for one or more user by an electronic device, comprising:
receiving, by a processor of the electronic device, object data, wherein the object data comprises a matrix, wherein rows of the matrix correspond to user IDs of the users respectively, columns of the matrix correspond to item IDs of items respectively, and each entry of the matrix indicates a rating related to corresponding item ID and corresponding user ID (figure 1; pg. 788, section 4.4; “User vectors and item vectors are rows and columns, respectively, of rating matrix R, where missing ratings are treated as 0s.”);
identifying, by the processor, values of first entries (xij) of the matrix according to the object data (figure 1; pg. 788, section 4.4; “In our case, x is a pair of user and item vectors. User vectors and item vectors are rows and columns, respectively, of rating matrix R, where missing ratings are treated as 0s.” values in the matrix R with are first entries.);
inputting, by the processor, the entries (xij) into an executed analysis model using Bi-Branch Neural Network (BiBNN) algorithm (figure 1; pg. 784; “First, it infers the preuse preferences on all user-item pairs based on the original rating matrix R. Second, it extracts the features (i.e., characteristics) of the user and the item using the variational autoencoder (VAE). Third, it trains the model that infers the post-use preference based on the pre-use preference and the extracted features using the multi-layer perceptron (MLP).);
obtaining, by the processor, a plurality of second entries (mij) of from the analysis model, wherein values of the second entries are determined as original ratings of the matrix, such that unknown ratings of the part of the first entries are predicted (pg. 783; “Data imputation aims to infer missing ratings and to impute them to the original rating matrix.” pg. 784; “First, it infers the preuse preferences on all user-item pairs based on the original rating matrix R. Second, it extracts the features (i.e., characteristics) of the user and the item using the variational autoencoder (VAE). Third, it trains the model that infers the post-use preference based on the pre-use preference and the extracted features using the multi-layer perceptron (MLP). Finally, we infer the post-use preference for every missing rating (i.e., ru,i = null) using the trained model, and impute the inferred post-use preferences to R.”); and
regarding each user ID, selecting, by the processor, one or more item IDs having ratings higher than a rating threshold, so as to determine one or more recommendation items corresponding to the selected item IDs for user corresponding to each user ID (pg. 785; section 3.1; “For the top-N recommendation, we vary the value of N (N = 5, 10, and 20) [9]. Three metrics are used to measure the accuracy of the top-N recommendation: precision, recall, and normalized Discounted Cumulative Gain (nDCG) [10]. For a user u, precision Pu@N and recall Ru@N are computed by |Relu ∩Recu | |Recu | and |Relu ∩Recu | |Relu | , respectively, where Recu denotes a set of N items recommended by CF to u, and Relu denotes a set of items considered relevant to u. We consider the items with rating 5 (best) as relevant items, (i.e., ground truth), following [3].” Items that meet the threshold rating of 5 are recommended to the user.).
Lee does not explicitly disclose
a preset maximum loop count (Kmax)
However, Song (US 20220350329 A1) teaches
inputting, by the processor, the entries (xij) and a preset maximum loop count (Kmax) into an executed analysis model using Bi-Branch Neural Network (BiBNN) algorithm (fig. 2; para [0024] “S600: presetting a maximum number of iterations, taking the actual path in the I moments and a corresponding predicted path as an input of a neural network model to train the error model E, updating parameters of the model through back propagation (BP) during training, and stopping training when E≤e.sup.−5 to obtain a well-trained odometry error model;” figure two shows a neural network with at least two branches similar to figure 5 of applicant’s drawings.)
Lee and Song are analogous because they are directed to the field of neural networks.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the variational autoencoder of Lee with the preset maximum number of iterations of Song.
Doing so would allow for terminating the training of the neural network when a maximum number of iterations is met (Song para [0024]).
Regarding Claim 11,
Claim 11 is the device corresponding the method of claim 1. Claim 11 is substantially similar to claim 1 and is rejected on the same grounds.
Regarding Claim 12,
Claim 12 is the device corresponding the method of claim 6. Claim 12 is substantially similar to claim 6 and is rejected on the same grounds.
Claims 3 and 8 are rejected under 35 U.S.C. 103 as being unpatentable over the combination of Lee/Song, as applied above, and further in view of Klamain et al. (US-20210350176-A1) and Singh et al. (US-20220292356-A1).
Regarding Claim 3,
Lee and Song teach the method of claim 1. Song further teaches wherein architecture of the BiBNN comprises the first branch, the second branch and an output layer,
wherein the first branch includes an input layer, a first hidden layer and a second hidden layer, and the second branch includes a further input layer, a further first hidden layer (fig. 2; para [0049] “S510: assuming that the neural network structurally includes an input layer having q neurons, a hidden layer having l neurons, and an output layer having m neurons, connections between the neurons of the input layer and the neurons of the hidden layer each have a weight value of w.sub.ij, connections between the neurons of the hidden layer and the neurons of the output layer each have a weight value of w.sub.jk , the hidden layer has a bias term of a, and the output layer has a bias term of b ;” Fig. 2 shows at least two branches.),
wherein the first hidden layer is connected from the input layer, the second hidden layer is connected from the first hidden layer, the further first hidden layer is connected from the further input layer (fig. 2; para [0049]),
Lee and Song do not explicitly disclose
and a further second hidden layer
the further second hidden layer is connected from the further first hidden layer, and the output layer is connected from the second hidden layer and the further second hidden layer,
wherein the first weight matrices (
U
1
k
+
1
) are calculated between the input layer and the first hidden layer, the second weight matrices (
U
2
k
+
1
) are calculated between the first hidden layer and the second hidden layer, the third weight matrices (
V
1
k
+
1
) are calculated between the further input layer and the further first hidden layer, the fourth weight matrices (
V
2
k
+
1
) are calculated between the further first hidden layer and the further second hidden layer.
However, Klamain (US 20210350176 A1) teaches
wherein the first branch includes an input layer, a first hidden layer and a second hidden layer, and the second branch includes a further input layer, a further first hidden layer and a further second hidden layer (para [0275] “The Siamese network 900 consists of two identical sub networks 902, 903 joined at their output layer 924. Each network comprises an input layer 905, 915 adapted to receive a single digital image (e.g. a tile) 954, 914 as input. Each sub-network comprises a plurality of hidden layers 906, 916, 908, 918. A one-dimensional feature vector 910, 920 is extracted from one of the two input images by a respective one of the two sub networks. Thereby, the last hidden layer 908, 918 of each network is adapted to compute the feature vector and provide the feature vector to the output layer 924.”),
wherein the first hidden layer is connected from the input layer, the second hidden layer is connected from the first hidden layer, the further first hidden layer is connected from the further input layer, the further second hidden layer is connected from the further first hidden layer, and the output layer is connected from the second hidden layer and the further second hidden layer (fig. 9; para [0275]),
Lee, Song, and Klamain are analogous because they are directed to the field of neural networks.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify neural networks of Lee and Song with the Siamese network architecture of Klamain.
Doing so would allow for improving the accuracy of the predictive model (Singh para [0126]).
Klamain teaches the first branch includes an input layer, a first hidden layer and a second hidden layer, and the second branch includes a further input layer, a further first hidden layer and a further second hidden layer. Singh further teaches that weight matrices are calculated between the input layer, first hidden layers, and second hidden layers.
Singh (US 20220292356 A1) teaches
wherein the first weight matrices (
U
1
k
+
1
) are calculated between the input layer and the first hidden layer, the second weight matrices (
U
2
k
+
1
) are calculated between the first hidden layer and the second hidden layer, the third weight matrices (
V
1
k
+
1
) are calculated between the further input layer and the further first hidden layer, the fourth weight matrices (
V
2
k
+
1
) are calculated between the further first hidden layer and the further second hidden layer (para [0015] “FIG. 3 shows a neural network containing an input layer, hidden layers, an output layer, respective weight matrices for the connections between the layers, according to an embodiment;” para [0034] “The weights of the connections between the layers may then be represented by a weight matrix. As shown, a weight matrix W.sub.1 may include the weights of the first set of connections 308a between the input layer 302 and the first hidden layer 304a. Similarly, weight matrix W.sub.2 may include the weights of a second set of connections 308b between the first hidden layer 304a and a second hidden layer 304b. Furthermore, weight matrix W.sub.3 may include of the weights of a third set of connections between the second hidden layer 304b and a third hidden layer 304c. Finally, weight matrix W.sub.3 may include of the weights of a fourth set of connections 308d between the third hidden layer 304c and the output layer 306.”).
Lee, Song, and Singh are analogous because they are directed to the field of neural networks.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the neural networks of Lee and Song with the network architecture of Singh.
Doing so would allow for an improved neural network that is trained to be more robust against adversarial attacks by controlling condition numbers of weight matrices forming the neural network (Singh para [0024]).
Regarding Claim 8,
Claim 8 is substantially similar to claim 3 and is rejected on the same grounds.
Claims 4 and 9 are rejected under 35 U.S.C. 103 as being unpatentable over the combination of Lee/Song/Klamain/Singh, as applied above, and further in view of Mayer et al. (US-20220198711-A1).
Regarding Claim 4,
Lee, Song, Klamain and Singh teach the method of claim 3.
Lee, Song, Klamain and Singh do not explicitly disclose wherein
the first hidden layer includes a first fully-connected layer and a first activation function layer, wherein the first fully-connected layer output a first calculation result (
U
1
T
α
i
) according to the first weight matrices (
U
1
k
+
1
) and the first auxiliary vector sets (
α
i
);
the second hidden layer includes a second fully-connected layer, wherein the second fully-connected layer output the first outputs (
u
i
);
the further first hidden layer includes a further first fully-connected layer and a further first activation function layer, wherein the further first fully-connected layer output a further first calculation result (
V
1
T
β
i
) according to the third weight matrices (
V
1
k
+
1
) and the second auxiliary vector sets (
β
j
); and
the further second hidden layer includes a further second fully- connected layer, wherein the further second fully-connected layer output the second outputs (
u
j
).
Mayer (US 20220198711 A1) teaches
the first hidden layer includes a first fully-connected layer and a first activation function layer (fig. 1.1, 2.2; para [0047] “Following the convolutional layers are a series of 2 fully-connected neuron layers, each with 200 neurons that have hyberbolic tangent activation. These fully connected layers are labeled fc_a1 and fc_a2 in FIG. 1.1. A final fully-connected layer with neurons and a softmax is used to identify the input camera model, where is the number of camera models in the training camera model set.” Fc_a1 is the first fully connected layer with a activation function as shown in figure 1.1.), wherein the first fully-connected layer output a first calculation result (
U
1
T
α
i
) according to the first weight matrices (
U
1
k
+
1
) and the first auxiliary vector sets (
α
i
) (para [0109] “1) Architecture: The first layer of neurons, labeled by ‘fcA’ in FIG. 2.2, contains 2048 neurons with ReLU activation. This layer maps an input feature vector f(X) to a new, intermediate feature space f.sub.inter(X).” para [0111] “which is the weighted summation, with weights w.sub.k,0 through w.sub.k,N, of the N=200 elements in the deep-feature vector f(X), bias term bk and subsequent activation by ReLU function φ(.).”);
the second hidden layer includes a second fully-connected layer, wherein the second fully-connected layer output the first outputs (
u
i
) (para [0047] “To extract features f(X) from an image patch X, the system(s) feed X forward through the trained network and record the neuron values, preactivation, of layer fc_a2. The feature vector f(X) has dimension 200 and encodes information about the source camera model of X.” fc_a2 is the second-fully connected layer that outputs f(x).);
the further first hidden layer includes a further first fully-connected layer and a further first activation function layer (fig. 1.1, 2.2; para [0047] “Following the convolutional layers are a series of 2 fully-connected neuron layers, each with 200 neurons that have hyberbolic tangent activation. These fully connected layers are labeled fc_a1 and fc_a2 in FIG. 1.1. A final fully-connected layer with neurons and a softmax is used to identify the input camera model, where is the number of camera models in the training camera model set.” Figure 1.1 shows there is another fc_a1 in the twin network.), wherein the further first fully-connected layer output a further first calculation result (
V
1
T
β
i
) according to the third weight matrices (
V
1
k
+
1
) and the second auxiliary vector sets (
β
j
) (para [0109] “1) Architecture: The first layer of neurons, labeled by ‘fcA’ in FIG. 2.2, contains 2048 neurons with ReLU activation. This layer maps an input feature vector f(X) to a new, intermediate feature space f.sub.inter(X).” para [0111] “which is the weighted summation, with weights w.sub.k,0 through w.sub.k,N, of the N=200 elements in the deep-feature vector f(X), bias term bk and subsequent activation by ReLU function φ(.).”); and
the further second hidden layer includes a further second fully- connected layer, wherein the further second fully-connected layer output the second outputs (
u
j
) (para [0047] “To extract features f(X) from an image patch X, the system(s) feed X forward through the trained network and record the neuron values, preactivation, of layer fc_a2. The feature vector f(X) has dimension 200 and encodes information about the source camera model of X.” para [0051] “The features f(X1) and f(X2) are separately connected to two layers fc b1 and fc b2. Both layers contain 1024 neurons with a parametric ReLU activation function. These layers learn to map the input features to a new feature space. During training, the weights are shared between fc_b1 and fc_b2.” Each of the feature extractors shown in figure 1.1 have their own fully connected layers and produces it’s separate outputs f(X1) and f(X2).).
Lee, Song, Klamain, Singh, and Mayer are analogous because they are directed to the field of neural networks.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the neural networks of Lee, Song, Klamain and Singh with the fully connected layers of Mayer.
Doing so would improve the system architecture and training procedure, which leads to a reduction in classification error (Mayer para [0076]).
Regarding Claim 9,
Claim 9 is substantially similar to claim 4 and is rejected on the same grounds.
Claims 5 and 10 are rejected under 35 U.S.C. 103 as being unpatentable over the combination of Lee/Song, as applied above, and further in view of Fauber et al. (US-11223543-B1).
Regarding Claim 5,
Lee and Song teach the method of claim 6.
Lee and Song do not explicitly disclose
wherein the second entries (mij) of the complete matrix are calculated by the equation below:
PNG
media_image1.png
12
59
media_image1.png
Greyscale
However, Fauber (US 11223543 B1)
wherein the second entries (mij) of the complete matrix are calculated by the equation below:
PNG
media_image1.png
12
59
media_image1.png
Greyscale
(col. 10 lines 58-63; “A complete signal vector x*∈[AltContent: rect].sup.j is assumed to be damaged such that it is missing k values, where k<<j, and the damaged vector is termed {acute over (x)}. The measurement matrix is defined as A∈[AltContent: rect].sup.m×j. Given A and the observations
PNG
media_image2.png
20
56
media_image2.png
Greyscale
the goal is to find a reconstruction {circumflex over (x)} as similar to x* as possible.”)
Lee, Song, and Fauber are analogous because they are directed towards the field of missing data imputation.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the neural networks of Lee and Song with the data imputation method of Fauber.
Doing so would allow for strong resistance to noise and rapid convergence when reconstructing missing signals (Fauber col. 10 lines 55-56;).
Regarding Claim 10,
Claim 10 is substantially similar to claim 5 and is rejected on the same grounds.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Cui et al. (US-20220058663-A1) – discloses imputing a matrix with missing values for user data ratings for items.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HENRY K NGUYEN whose telephone number is (571)272-0217. The examiner can normally be reached Mon - Fri 7:00am-4:30pm.
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, Li B Zhen can be reached at 5712723768. 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.
/HENRY NGUYEN/Examiner, Art Unit 2121