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 .
This action is responsive to the filing on 002/08/2022. Claim(s) 37-38, 40 and 42 have been amended and claims 1-36, 39, 41 and 43-49 have been canceled. Claims 50-53 have been newly added. Claims 37-39, 40, 42 and 50-53 are pending in this case. Claim 37 is the only independent claim.
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.
Claim(s) 37-38, 40, 42, and 51-53 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
When considering subject matter eligibility under 35 U.S.C. 101, it must be determined whether the claim is directed to one of the four statutory categories of invention, i.e., process, machine, manufacture, or composition of matter (Step 1). If the claim does fall within one of the statutory categories, the second step in the analysis is to determine whether the claim is directed to a judicial exception (Step 2A). The Step 2A analysis is broken into two prongs. In the first prong (Step 2A, Prong 1), it is determined whether or not the claims recite a judicial exception (e.g., mathematical concepts, mental processes, certain methods of organizing human activity). If it is determined in Step 2A, Prong 1 that the claims recite a judicial exception, the analysis proceeds to the second prong (Step 2A, Prong 2), where it is determined whether or not the claims integrate the judicial exception into a practical application. If itis determined at step 2A, Prong 2 that the claims do not integrate the judicial exception into a practical application, the analysis proceeds to determining whether the claim is a patent-eligible application of the exception (Step 2B). If an abstract idea is present in the claim, any element or combination of elements in the claim must be sufficient to ensure that the claim integrates the judicial exception into a practical application, or else amounts to significantly more than the abstract idea itself. Applicant is advised to consult the 2019 PEG for more details of the analysis.
Step 1 Analysis: Is the claim to a process, machine, manufacture or composition of matter? See MPEP § 2106.03.
Claims 37-38, 40, 42, 51-53 are drawn to a method, therefore each of these claim groups falls under one of four categories of statutory subject matter (machine/products/apparatus, process/method, manufactures and compositions of mater; Step 1). Nonetheless, the claims are directed to a judicially recognized exception of an abstract idea without significant more (Step 2A, see below).
Step 2A Prong One: Recites Abstract Idea, Law Of Nature, Natural Phenomenon
Claim(s) 37-38, 40, 42, 51-53 are directed to an abstract idea, specifically, a mental process – concepts performed in the human mind (including an observation, evaluation, judgement, opinion) as well as a mathematical concepts – (mathematical relationships, mathematical formulas or equations, mathematical calculations).
Independent claim 37 recites in part:
A method for training [
obtaining training data having a label indicating a class of the training data
The limitation above is broadly and reasonably interpreted as a mental process, as a form or
mental evaluation or judgement. For example, one can obtain a dataset of images of fruits and each image is labeled with its corresponding fruit type (e.g., "apple", "banana", "orange").
“generating the predicted label based on the at least one mode parameter, the weight, and a random value obeying a distribution corresponding to the at least one mode parameter”
The limitation is broadly and reasonably interpreted as a mental process , as a form or
mental evaluation or judgement using a “pen and paper". For example, one can create a label using parameters that dictates how information is handled, the strength of the labels (weight), and a random value from a distribution of the parameters that dictates how information is handled.
“generating, [
The limitation above is broadly and reasonably interpreted as a mental process, as a form or
mental evaluation or judgement using a “pen and paper". For example, with pen and paper one can create a dataset of images of fruits and each image is labeled with its corresponding fruit type (e.g., "apple", "banana", "orange").
“determining a loss of the neural network based on the label, the predicted label, and the weight applied to the at least one candidate class”
The limitation above are broadly and reasonably interpreted as a mathematical concepts – (mathematical relationships, mathematical formulas or equations, mathematical calculations). For example, imagine a dataset of images of fruits and each image is labeled with its corresponding fruit type (e.g., "apple", "banana", "orange"). We can classify images as either “apple” (class 0) or “banana” (class 1) (class 0 (apple: Weight = 1.0) and (class 1 (banana: Weight = 2.0))). A loss function can then be used to determine the loss of the label and weight.
Step 2A Prong Two: Does Not Integrate Into Practical Application
The judicial exception is not integrated into a practical application. The method in claim 1 does not seem to incorporate any computing components other than a “neural network”, and does/do not articulate in a way that would be considered “significantly more” under the MPEP guidelines.
The limitation “obtaining an output of at least one layer before a weighted layer in the neural network as an input of the weighted layer, the input indicating a possibility that the training data belongs to at least one candidate class” as drafted, amount to insignificant extra-solution activity to the judicial exception, e.g., mere data gathering in. conjunction with a law of nature or abstract idea such as a step of obtaining information
The limitation “determining, based on at least one parameter of the weighted layer and the input of the weighted layer, at least one mode parameter associated with a type of distribution mode and a weight applied to the at least one candidate class;” as drafted, amount to insignificant extra-solution activity, as a generic computing component reciting at a high-level of generality (i.e., as a generic processor performing data gathering and mathematical calculations) such that they amount to no more than mere instructions to apply the exception using generic computer components.
The limitation “training the neural network based on the predicted label to minimize a difference between the label and the predicted label;
updating network parameters of the neural network based on the loss to minimize the loss of the updated neural network”, as drafted, amounts to insignificant extra-solution activity, the process of training a machine learning model (i.e., neural network) is known to be required for any machine learning model. Also, providing iterative training to the trained machine learning model is also the very nature of machine learning.
Looking at the claims limitations as an ordered combination and taking the claim as a whole, there still is not integration into a practical application. The claims also don’t appear to improve the functioning of a computer or require the use of a specific machine. SEE MPEP 2106.04(d)(1) and 2106.05(a). Accordingly, these additional elements do not integrate the abstract idea into a practical application because they don’t impose any meaningful limits on practicing the abstract idea. The claim is directed to an abstract idea.
Step 2B: Does Not Amount To Significantly More
As per MPEP 2106.05(II) the considerations discussed above for mere instructions to apply the exception and merely linking to a field of use are carried over for Step 2B.
Claim 38 is dependent on claim 37, and include a clarification of what sources comprise, which doesn’t provide integration into a practical application or add significantly more to the abstract idea because including attributes to assess the data is insignificantly extra-solution activity, and therefore doesn’t break away from the reasons for the identified abstract idea
Claim 40 is dependent on claim 38, and include a clarification of what the predetermined mode is, which doesn’t provide integration into a practical application or add significantly more to the abstract idea because including attributes to assess the data is insignificantly extra-solution activity.
Claim 42 is dependent on claim 37, and include insignificant extra-solution activity, providing iterative training to the trained machine learning model is also the very nature of machine learning.
Claim 50 is dependent on claim 37, and include clarification wherein the type of distribution can be one of the following: Gaussian, normal, uniform, exponential, Poisson, Bernoulli, or Laplace, which doesn’t provide integration into a practical application or add significantly more to the abstract idea because including attributes to assess the data is insignificantly extra-solution activity.
Claim 51 is dependent on claim 37, and include clarification wherein the type of distribution can be one of the following: Gaussian, normal, uniform, exponential, Poisson, Bernoulli, or Laplace, which doesn’t provide integration into a practical application or add significantly more to the abstract idea because including attributes to assess the data is insignificantly extra-solution activity.
Claim 52 is dependent on claim 37, and include clarification wherein the neural network can be one of the following: a Deep Neural Network (DNN), a Convolutional Neural Network (CNN), a Long Short Term Memory (LSTM) Network, a Gated Recurrent Unit (GRU) network, or a Recurrent Neural Network (RNN), which doesn’t provide integration into a practical application or add significantly more to the abstract idea because including attributes to assess the data is insignificantly extra-solution activity.
Claim 53 is dependent on claim 37, and include clarification wherein the input data can be one or more of the following: an image, a video, sound, text, or a multimedia file, which doesn’t provide integration into a practical application or add significantly more to the abstract idea because including attributes to assess the data is insignificantly extra-solution activity.
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.
Claim(s) 37, 38, 42 and 52-53 are rejected under 35 U.S.C. 103 as being unpatentable over Rastrow et al. (US Patent No.9,600,764 B1), hereinafter referred to as Rastrow, in view of Karlinsky et al. (US Pub No.: 20220188639 A1), hereinafter referred to as Karlinsky.
With respect to claim 37, Rastrow disclose:
A method for training a neural network, comprising: obtaining training data having a label indicating a class of the training data (In Fig. 3 & Col. 7, lines 29-37, Rastrow discloses the process for sequence tagging using Neural-Network-Based Markov Models. The Neural-Network-Based Markov receives a series of data that are tagged (labeled).)
generating, by using a neural network, a predicted label of the training data by obtaining an output of at least one layer before a weighted layer in the neural network as an input of the weighted layer, the input indicating a possibility that the training data belongs to at least one candidate class (In Fig. 1 & Col. 4, lines 54-60, Rastrow disclosed generating an output that represents probabilities for different labels (e.g., categories, classes) associated with a specific position in a sequence. The network uses information from a previously known label or predicted label for a prior position in the sequence to inform its output for the current position. The neural network includes: input layer, output layer and hidden layer(input layers are passed through each hidden layer, where it is transformed through mathematical operations, with each layer performing calculations based on its weights and biases (i.e., weighted layer)). In Col. 5, lines 3-13 disclose obtaining output generated by the neural network for a feature vector of a previous position (input signal) in the sequence. The output corresponds to a local probability distribution over all possible labels for the current position within the sequence. In Col. 5, lines 42-50, further disclose the label for the previous position at t−1 may be set to some predetermined value (e.g., 1), as the sequential nature of the input data.)
determining, based on at least one parameter of the weighted layer and the input of the weighted layer, at least one mode parameter associated with a type of distribution mode and a weight applied to the at least one candidate class ( In Col. 2, lines 39-42, Rastrow discloses that the input layer may consist of a set of parameters (e.g., a feature vector) for a given time instance, such as a frame of audio data. In Col. 5, lines 10-13, Rastrow disclose a neural network performs a forward pass through its hidden layers to produce an output, called “y.sub.t” (the data variable from hidden layer and input layer of the hidden layer). In Col. 5, lines 28-55, Rastrow disclose the neural network generates a probability distribution y.sub.t for the feature vector corresponding to the next position at time t within the input signal.)
and generating the predicted label based on the at least one mode parameter, the weight, and a random value obeying a distribution corresponding to the at least one mode parameter (In Fig. 1 & Col. 4, lines 54-60, Rastrow disclosed generating an output that represents probabilities for different labels (e.g., categories, classes) associated with a specific position in a sequence. The network uses information from a previously known label or predicted label for a prior position in the sequence to inform its output for the current position. In Col. 4-5, lines 66-7, disclose the output generated by the neural network 100 for a feature vector (parameter)of a previous position in the sequence. For example, the previous label data 122 may be named s.sub.t-1, where t−1 corresponds to the time instance or other position immediately preceding t in the sequence.)
With respect to claim 37, Rastrow do not explicitly disclose:
training the neural network based on the predicted label to minimize a difference between the label and the predicted label by: determining a loss of the neural network based on the label, the predicted label and the weight applied to the at least one candidate class
and updating network parameters of the neural network based on the loss to minimize the loss of the updated neural network
However, it is known by Karlinsky to disclose:
Training the neural network based on the predicted label to minimize a difference between the label and the predicted label by: determining a loss of the neural network based on the label, the predicted label and the weight applied to the at least one candidate class (In paragraph [0016], Karlinsky discloses using a cross-entropy loss measuring the quality of predicted labels relative to generated labels, receiving predicted labels from a gradient-generating subnetwork 106 weighted by weights produced by the weight-predicting network. In paragraph [0020], further disclose the training model, computing a loss of common subnetwork based on training with the labeled data, based on predicted labels from the actual label. In paragraph [0022], they disclose that training the model using gradient loss helps reduce the common subnetwork.)
Updating network parameters of the neural network based on the loss to minimize the loss of the updated neural network (In paragraph [0025], Karlinsky discloses that the training module may permanently update the gradient-generating subnetwork with the gradient based on the training loss, reducing the loss of the gradient-generating subnetwork.)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to incorporate Rastrow’s invention in view of Karlinsky in order to reduce the total processing time and improving performance of the neural network as taught by Rastrow (see(Col. 8, lines 26-34 ).)
Regarding claim 38, Rastrow in view of Karlinsky disclose elements of claim 37. In addition, Rastrow disclose:
The method according to claim 37, wherein the weighted layer further determines at least one mode parameter associated with a predetermined mode to generate a predicted result to obey the predetermined mode (In Col. 5, lines 42-50, Rastrow discloses initializing the output label for the previous position to a specific value (e.g., 1) while setting outputs for other positions to a default value (e.g., 0), predicting the current output based on the pervious one.)
Regarding claim 42, Rastrow in view of Karlinsky disclose elements of claim 37. In addition, Rastrow disclose:
The method according to claim 37, wherein updating the network parameters of the neural network based on the loss comprises: updating at least one parameter of the weighted layer based on the loss to minimize the loss of the updated neural network (In paragraph [0025], Karlinsky discloses that the training module 110 may reduce the training loss with respect to gradient-generating subnetwork 106 to generate the reduction of the main loss by means of the update of main network parameters with the gradient that gradient-generating subnetwork 104 generates. )
Regarding claim 52, Rastrow in view of Karlinsky disclose elements of claim 37. In addition, Rastrow disclose:
The method according to claim 37, wherein the neural network comprises one of: a Deep Neural Network (DNN), a Convolutional Neural Network (CNN), a Long Short Term Memory (LSTM) Network, a Gated Recurrent Unit (GRU) network, and a Recurrent Neural Network (RNN) (In Col. 2, lines 49-50, Rastrow discloses that the neural network used in structured prediction is a recurrent neural network (“RNN”).)
Regarding claim 53, Rastrow in view of Karlinsky disclose elements of claim 37. In addition, Rastrow disclose:
The method according to claim 37, wherein the input data comprises at least one of: an audio (In Col. 2, lines 39-42, disclose the input layer consists of audio data.)
Claim(s) 40 and 50-51 are rejected under 35 U.S.C. 103 as being unpatentable over Rastrow in view of Karlinsky and further in view of Chao et al. (US Patent No.12,033,089 B2), hereinafter referred to as Chao.
Regarding claim 40, Rastrow in view of Karlinsky disclose elements of claim 38. Rastrow in view of Karlinsky do not explicitly disclose:
The method according to claim 38 , wherein the predetermined mode is a Gaussian distribution, and the at least one mode parameter comprises a mean value and a variance of the Gaussian distribution
However, Chao disclose the limitation (In Col. 6, lines 58-64, Chao discloses a Gaussian distribution given parameters representing the mean and covariance, respectively.)
Accordingly, it would have been obvious to a person having ordinary skills in art before the effective filling date of the claimed invention, having the teaching of Rastrow and Karlinsky before them, to include Chao’s to reduce the complexity of training the deep convolutional deep convolutional factor analyzer (DCFA) as taught by Chao (see(Col. 4, lines 64-65).)
Regarding claim 50, Rastrow in view of Karlinsky disclose elements of claim 37. Rastrow in view of Karlinsky do not explicitly disclose:
The method according to claim 37, wherein the type of distribution mode is one of: a Gaussian distribution, a normal distribution, a uniform distribution, an exponential distribution, a Poisson distribution, a Bernoulli distribution, and a Laplace distribution
However, Chao disclose the limitation (In Col. 6, lines 58-64, Chao discloses a Gaussian distribution given parameters representing the mean and covariance, respectively.)
Accordingly, it would have been obvious to a person having ordinary skills in art before the effective filling date of the claimed invention, having the teaching of Rastrow and Karlinsky before them, to include Chao’s to reduce the complexity of training the deep convolutional deep convolutional factor analyzer (DCFA) as taught by Chao (see(Col. 4, lines 64-65).)
Regarding claim 51, Rastrow in view of Karlinsky disclose elements of claim 37. Rastrow in view of Karlinsky do not explicitly disclose:
The method according to claim 37, wherein the type of distribution mode is a Gaussian distribution, and the at least one mode parameter comprises a mean value and a variance of the Gaussian distribution
However, Chao disclose the limitation (In Col. 6, lines 58-64, Chao discloses a Gaussian distribution given parameters representing the mean and covariance, respectively.)
Accordingly, it would have been obvious to a person having ordinary skills in art before the effective filling date of the claimed invention, having the teaching of Rastrow and Karlinsky before them, to include Chao’s to reduce the complexity of training the deep convolutional deep convolutional factor analyzer (DCFA) as taught by Chao (see(Col. 4, lines 64-65).)
Response to Arguments
The applicant's arguments filed 0/17/2025 have been fully considered, but in part are not persuasive.
Pertaining to Rejection under 101
On Page 7, the Examiner respectfully remains convinced that the amended claim 1 still do not overcome rejection under 35 rejection under 35 U.S.C 101. For instance, in claim 1, under step 2A: Prong One recites in part an abstract idea under as a mental concept… “generating the predicted label based on the at least one mode parameter, the weight, and a random value obeying a distribution corresponding to the at least one mode parameter”, one can create with either pen and paper or the use of a generic computer a dataset of images of fruits and each image is labeled with its corresponding fruit type (e.g., "apple", "banana", "orange"). The weight of each image can be given based on the random occurrence of each image in the dataset (e.g., apple:2, banana: 3, orange:1). Therefore, directed to an abstract idea.
Furthermore, the applicant also discusses on page 8 (paragraph 2) the improvement in computer technology, under Step 2A: Prong Two, claim 1 amount to insignificant extra-solution activity. In part of claim 1, the limitation of “training the neural network based on the predicted label to minimize a difference between the label and the predicted label” as well as “updating network parameters of the neural network based on the loss to minimize the loss of the updated neural network” are found to be insignificant. Simply appending well-understood, routine, conventional activities previously known to the industry, specified at a high level of generality, to the judicial exception, e.g., a claim to an abstract idea requiring no more than a generic computer to perform generic computer functions. Also, the process of training the machine learning model and updating the machine learning model are incident to the very nature of machine learning. It is the examiner's opinion that nowhere in claim(s) shows real improvement other than directing the improvement to an abstract idea. Improvements can be provided by one or more additional elements. In this case, there are no additional elements.
Pertaining to Rejection under 103
Applicant’s arguments in regard to the examiner’s rejections under 35 USC 103 are moot in view of the new grounds of rejection.
Conclusion
THIS ACTION IS MADE FINAL. 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 nonprovisional extension fee (37 CFR 1.17(a)) 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 mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to EVEL HONORE whose telephone number is (703) 756- 1179. The examiner can normally be reached on Monday through Friday from 8am to 5pm. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Ashish Thomas can be reached on (571) 272-0631. 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 application may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and mange patent submissions in Patent Center, 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.
/EVEL HONORE/Examiner, Art Unit 2142
/HAIMEI JIANG/ Primary Examiner, Art Unit 2142