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 made non-final.
Claims 1-20 are pending. Claims 1 and 11 and are independent claims.
Specification
The preliminary amendment to the Specification filed on 01/12/2023 is acknowledged.
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 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1:
Step 1: This part of the eligibility analysis evaluates whether the claim falls within any statutory category. See MPEP 2106.03. Claim 1 recites: A method for ameliorating negative impacts of signals that are sparse in various data series for trainings of artificial neural network models, the method comprising…. Claim 1 is directed to a method (Step 1: YES).
Step 2A prong 1: Does the claim recite a judicial exception? Claim 1 recites: segmenting… the data samples into a plurality of batches according to the window size… (segmenting data based on window size is a mental process)… inputting… the window of data and the prediction value into the loss function such that the loss function outputs a plurality of computed loss values for each of the data samples for a respective window of data when the ANN has a first state including a set of weights by the model (calculating loss using a loss function with data and prediction values is a mathematical calculation)… determining, by the processor, a probability value for each of the data samples within a respective batch while accounting a total duration of a respective data sample, a cumulative amount of data in the respective sample that has already been processed… and an average frequency of a respective signal per the respective data sample (determining a probability value is a mathematical calculation); generating… a plurality of new sample weights based on applying the probability values to the sample weights for the respective batch (generating new sample weights based on applying probabilities to existing sample weight values is a mathematical calculation, i.e., multiplication); generating… a new set of computed loss values based on the computed loss values and the new sample weights (generating new loss values based on weights and old loss values is a mathematical calculation, i.e., multiplication)…… applying… the new set of computed loss values to the set of weights such the set of weights is changed from the first state into a second state (applying loss values to weights to change them from one state to another is a mathematical calculation, i.e. gradient descent). These steps can be performed mentally or are mathematical calculations (Step 2A prong 1: YES).
Step 2A prong 2: Does the claim recite additional elements? Do those additional elements, considered individually and in combination, integrate the judicial exception into a practical application? Claim 1 recites: accessing, by a processor, a window size, a loss function, a plurality of sample weights, and a data series, wherein the loss function has a first value and a second value, wherein the data series contains a plurality of data samples containing a plurality of signals, wherein the signals are sparse within the data samples… by the processor… for each of the batches: causing, by the processor, a model of an artificial neural network (ANN) to output a prediction value based on the first value given a window of data based on the second value, wherein the window data corresponding to the window size… by the processor… by the processor… by the processor… by the processor… and causing, by the processor, the ANN to be programmed for generating a prediction that is more accurate at the second state than the first state. Accessing various inputs like a window size, loss function, weights, and data series is insignificant extra-solution activity of data gathering that does not add a meaningful limitation to the neural network method. Performing various steps by using a processor are mere instructions to implement an abstract idea on a generic computer. Causing a model of the ANN to output a prediction value based on inputs determined via a window size, and causing the ANN to generate more accurate predictions at a second state are an attempt to use the neural network model by merely applying the abstract idea without placing any limits on how the neural network model operates. Further, the claim omits any details as to how the neural network model solves a technical problem and instead recites only the idea of a solution or outcome. See MPEP 2106.05(f). Thus, the limitation represents no more than mere instructions to implement the abstract idea which is equivalent to adding the words “apply it” to the recited judicial exception (Step 2A prong 2: NO).
Step 2B: These elements are recited at such a high level of generality that they fail to integrate the abstract idea into a practical application, since they only amount to data gathering or outputting without significantly more (MPEP 2106.05(g)) or provide nothing more than mere instructions to implement an abstract idea on a generic computer (MPEP 2106.05(f)). These limitations, taken either alone or in combination, fail to provide an inventive concept (Step 2B: NO). Thus, the claim is not patent eligible.
Regarding claims 2-10, they recite limitations which further narrow the abstract idea by specifying more details of the mental and mathematical process that occurs (Claim 2, determining the probability value using P=I/N is a mathematical formula; Claim 3, multiplying loss and weights is a mathematical calculation; Claim 4, describing the input data samples length is still insignificant extra-solution activity (see, e.g., CyberSource v. Retail Decisions and Electric Power Group, LLC v. Alstom S.A., both of which were found to merely perform data gathering or selecting a particular data source or type of data to be manipulated); Claim 5, using a particular type of neural network, in this case a stateful ANN, is an attempt to use the neural network model by merely applying the abstract ideas without placing any limits on how the neural network model operates; Claim 6, using a particular type of neural network, in this case a stateful RNN, is an attempt to use the neural network model by merely applying the abstract ideas without placing any limits on how the neural network model operates; Claim 7, using a particular type of neural network, in this case a stateful LSTM, is an attempt to use the neural network model by merely applying the abstract ideas without placing any limits on how the neural network model operates; Claim 8, using a particular type of neural network, in this case a stateful CNN, is an attempt to use the neural network model by merely applying the abstract ideas without placing any limits on how the neural network model operates; Claim 9, overriding a default behavior of a machine learning framework to train the ANN based on the second state is recited at a high level of generality, i.e., as a generic computer performing generic computer functions; Claim 10, describing input data as being sourced from electrical leads is attempting to limit the field of use without significantly more (MPEP 1206.05(h))).
Regarding claim 11, it is a system implementing a method similar to claim 1 and is rejected on the same grounds – see above.
Regarding claims 12-20, they recite similar limitations to claims 2-10, respectively, and are rejected on the same grounds – see above.
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) 1-4, 9, 11-14 and 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Li et al. (“VS-GRU: A Variable Sensitive Gated Recurrent Neural Network for Multivariate Time Series with Massive Missing Values”, 2019, INLCUDED IN IDS), herein Li, in view of Miroshnikov et al. (US 20210350272 A1), herein Miroshnikov and Ribera Prat et al. (US 20210073675 A1), herein Ribera Prat.
Regarding claim 1, Li teaches: A method for ameliorating negative impacts of signals that are sparse in various data series for trainings of artificial neural network models (Abstract, we propose a novel recurrent neural network called variable sensitive GRU (VS-GRU), which utilizes the different missing rate of each variable as another input and learns the feature of different variables separately, reducing the harmful impact of variables with high missing rates), the method comprising: accessing, by a processor, (pg. 6, ¶2, we use the weight vector), and a data series, wherein the loss function has a first value and a second value (pg. 8, eq. 20, loss function has inputs and outputs), wherein the data series contains a plurality of data samples containing a plurality of signals, wherein the signals are sparse within the data samples (Abstract, Multivariate time series are often accompanied with missing values, especially in clinical time series); (pg. 7, ¶1, A fully connected layer with a sigmoid activation function is added at the output of the GRU layer for the classification problem. This layer can integrate all variables and outputs the probability), a plurality of computed loss values for each of the data samples for (pg. 8, eq. 20, loss values calculated from multiple time steps, with values from previous timesteps being used in the calculation of values in the current timesteps); determining, by the processor, a probability value(pg. 4, eq. 2 and ¶4, In addition, we calculate the missing rate μ of each variable d in each time series x of length T – the series length T is a total duration, the mask indicator m is a cumulative amount of data that has been processed, and the average frequency is the 1/T multiplied by summation term); generating, by the processor, a plurality of new sample weights based on applying the probability values to the sample weights (pg. 6, eq. 11, Weights are multiplied by the missing rates to form the “missing factor”)
Li fails to teach: a window size and batch processing of the data samples.
However, in the same field of endeavor, Miroshnikov teaches: a window size and batch processing of the data samples (¶67, batch training that divides the set of training data into small batches and processes each batch of training data via different instances of the ML model 150 in parallel – dividing training data into batches having a certain size, i.e., window size).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use batches with a certain size as disclosed by Miroshnikov in the method disclosed by Li to parallelize the training process (¶67, Certain ML models 150 can be optimized to run in parallel… parallel training techniques, such as batch training)
Li in view of Miroshnikov fails to teach: generating, by the processor, a new set of computed loss values based on the computed loss values and the new sample weights; applying, by the processor, the new set of computed loss values to the set of weights such the set of weights is changed from the first state into a second state; and causing, by the processor, the ANN to be programmed for generating a prediction that is more accurate at the second state than the first state.
However, in the same field of endeavor, Ribera Prat teaches: generating, by the processor, a new set of computed loss values based on the computed loss values and the new sample weights (¶50, weighting the loss function… to generate a weighted loss function… is performed by multiplying the loss for any given data point by its corresponding weight); applying, by the processor, the new set of computed loss values to the set of weights such the set of weights is changed from the first state into a second state; and causing, by the processor, the ANN to be programmed for generating a prediction that is more accurate at the second state than the first state (¶52, The training performed in operation 370 may be performed using standard techniques for learning the parameters of a machine learning model 372, such as performing gradient descent to minimize the loss function. The result of the training is a trained model 374 that includes one or more values that configure the underlying model 372 to compute predictions or inferences that are consistent with the training data set 312 – performing gradient descent or other parameter learning techniques involves applying loss values to update weights, i.e., change them from a first state to a second state).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use a weighted loss function as disclosed by Ribera Prat in the method disclosed by Li in view of Miroshnikov to improve accuracy (¶4, thereby enabling the training of accurate regression models when the training data is highly imbalanced or highly non-uniform).
Regarding claim 2, Li further teaches: The method of claim 1, wherein the probability value is determined based on P = I/N (pg. 4, eq. 2 and ¶4, In addition, we calculate the missing rate μ of each variable d in each time series x of length T – The summation term is being interpreted as I and T is being interpreted as N).
Regarding claim 3, Li in view of Miroshnikov fails to teach: The method of claim 1, wherein the new set of computed loss values is generated based on the computed loss values and the new sample weights being multiplied.
However, in the same field of endeavor, Ribera Prat teaches: wherein the new set of computed loss values is generated based on the computed loss values and the new sample weights being multiplied (¶50, weighting the loss function… to generate a weighted loss function… is performed by multiplying the loss for any given data point by its corresponding weight).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use a weighted loss function as disclosed by Ribera Prat in the method disclosed by Li in view of Miroshnikov to improve accuracy (¶4, thereby enabling the training of accurate regression models when the training data is highly imbalanced or highly non-uniform).
Regarding claim 4, Li further teaches: The method of claim 1, wherein the data samples are different from each other in a sequence length (pg. 2, fig. 1, variable d1 has a sequence length of 2, while d3 has a sequence length of 12).
Regarding claim 9, Li in view of Miroshnikov fails to teach: The method of claim 1, further comprising: overriding, by the processor, a default behavior of a machine learning framework such that the model of the ANN is trained based on the second state.
However, in the same field of endeavor, Ribera Prat teaches: further comprising: overriding, by the processor, a default behavior of a machine learning framework such that the model of the ANN is trained based on the second state (¶52, The training performed in operation 370 may be performed using standard techniques for learning the parameters of a machine learning model 372, such as performing gradient descent to minimize the loss function. The result of the training is a trained model 374 that includes one or more values that configure the underlying model 372 to compute predictions or inferences that are consistent with the training data set 312 – the default behavior of a machine learning framework can be interpreted as producing a prediction or inference output from input data, while the model is capable of being trained via updating parameters as the “overriding” behavior).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use a weighted loss function to train a neural network as disclosed by Ribera Prat in the method disclosed by Li in view of Miroshnikov to improve network accuracy on imperfect data (¶4, thereby enabling the training of accurate regression models when the training data is highly imbalanced or highly non-uniform).
Regarding claim 11, it is a system that implements a method similar to claim 1 and is rejected on the same grounds – see above.
Regarding claims 12, 13, 14 and 19, they recite similar limitations to claims 2, 3, 4 and 9, respectively, and are rejected on the same grounds.
Claim(s) 5-7 and 15-17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Li in view of Miroshnikov and Ribera Prat as applied to claims 1 and 11 above, and further in view of Thomas (US 20190385748 A1).
Regarding claim 5, Li in view of Miroshnikov and Ribera Prat fails to explicitly teach: The method of claim 1, wherein the ANN is a stateful ANN.
However, in the same field of endeavor, Thomas teaches: wherein the ANN is a stateful ANN (¶39, In addition to typical weights and biases, networks may include gates to hold memory as well as gates to remove data from memory such as in a Long Short-Term Memory (LSTM) network. A stateful network such as the LSTM…).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to utilize a stateful network like an LSTM as disclosed by Thomas in the method disclosed by Li in view of Miroshnikov and Ribera Prat to allow the network to learn from prior events (¶39, allows the network to understand the context of current data based on prior events).
Regarding claim 6, Li in view of Miroshnikov and Ribera Prat fails to explicitly teach: The method of claim 5, wherein the stateful ANN is a stateful recurrent neural network (RNN).
However, in the same field of endeavor, Thomas teaches: wherein the stateful ANN is a stateful recurrent neural network (RNN) (¶39, In addition to typical weights and biases, networks may include gates to hold memory as well as gates to remove data from memory such as in a Long Short-Term Memory (LSTM) network. A stateful network such as the LSTM – a LSTM is a type of RNN).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to utilize a stateful RNN like an LSTM as disclosed by Thomas in the method disclosed by Li in view of Miroshnikov and Ribera Prat to allow the network to learn from prior events (¶39, allows the network to understand the context of current data based on prior events).
Regarding claim 7, Li in view of Miroshnikov and Ribera Prat fails to explicitly teach: The method of claim 5, wherein the stateful ANN is a stateful long short-term memory (LSTM).
However, in the same field of endeavor, Thomas teaches: wherein the stateful ANN is a stateful long short-term memory (LSTM) (¶39, In addition to typical weights and biases, networks may include gates to hold memory as well as gates to remove data from memory such as in a Long Short-Term Memory (LSTM) network. A stateful network such as the LSTM…).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to utilize a stateful network like an LSTM as disclosed by Thomas in the method disclosed by Li in view of Miroshnikov and Ribera Prat to allow the network to learn from prior events (¶39, allows the network to understand the context of current data based on prior events).
Regarding claims 15-17, they recite similar limitations to claims 5-7, respectively, and are rejected on the same grounds – see above.
Claim(s) 8 and 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Li in view of Miroshnikov, Ribera Prat and Thomas as applied to claims 5 and 15 above, and further in view of Munkberg et al. (US 20200126192 A1), herein Munkberg.
Regarding claim 8, Li in view of Miroshnikov, Ribera Prat and Thomas fails to teach: The method of claim 5, wherein the stateful ANN is a stateful convolutional neural network (CNN).
However, in the same field of endeavor, Munkberg teaches: wherein the stateful ANN is a stateful convolutional neural network (CNN) (¶54, Neurons or nodes (represented by circles) in a first recurrent convolutional layer generate outputs to neurons or nodes of a second convolutional layer. The recurrent convolutional layer is a stateful machine).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to utilize a stateful CNN as disclosed by Munkberg in the method disclosed by Li in view of Miroshnikov, Ribera Prat and Thomas to better process long input sequences (¶54, learn how to use past information generated by the recurrent convolutional layer and stored in hidden state… to process a new input… while generating hidden state... RNN layers process arbitrarily long sequences of inputs, such as image sequences, and are natural candidates for temporally stable image reconstruction).
Regarding claim 18, it recites similar limitations to claim 8 and is rejected on the same grounds – see above.
Claim(s) 10 and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Li in view of Miroshnikov and Ribera Prat as applied to claims 1 and 11 above, and further in view of Kumar et al. (US 20190303713 A1), herein Kumar.
Regarding claim 10, Li in view of Miroshnikov and Ribera Prat fails to teach: The method of claim 1, wherein the data samples are sourced from a plurality of electrical leads.
However, in the same field of endeavor, Kumar teaches: wherein the data samples are sourced from a plurality of electrical leads (¶82, electrical leads 1000 and 1002 are applied to a person 1004 and generate analog electrical signals that are provided to a computing device 1005. An analog-to-digital convertor 1006 converts the).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use signals from electrical leads as disclosed by Kumar in the method disclosed by Li in view of Miroshnikov and Ribera Prat (¶82, that automatically determines the state of a person based on electrical signals generated by the person).
Regarding claim 20, it recites similar limitations to claim 10 and is rejected on the same grounds – see above.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Wang et al. (US 20200093455 A1), which discloses using a weighted loss function in machine learning that resulted in better performance on an imbalanced dataset.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HARRISON CHAN YOUNG KIM whose telephone number is (571)272-0713. The examiner can normally be reached Monday - Thursday 10:00 am - 7:00 pm.
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, CESAR PAULA can be reached at (571) 272-4128. 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.
/HARRISON C KIM/ Examiner, Art Unit 2145
/CHAU T NGUYEN/ Primary Examiner, Art Unit 2145