Detailed Action
This FINAL communication is in response to the amendment filed 07/08/2025.
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 . Claims 1-18 have been examined.
Response to Amendment
The amendments filed 07/08/2025 have been entered.
Claims 1-20 remain pending within the application.
The amendment filed 07/08/2025 is sufficient to overcome the disclosure’s objections.
The amendment filed 07/08/2025 is sufficient to overcome the claim objection.
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:
Claim 1 recites
A method for data processing, comprising:
determining respective degrees of influence of a plurality of augmented sample sets in a training set on a model to be trained, the plurality of augmented sample sets corresponding to a plurality of original samples;
determining, based on the degrees of influence, a first group of augmented sample sets from the plurality of augmented sample sets, the first group of augmented sample sets being to have a negative influence on the model to be trained;
determining a training loss function associated with the training set, in the training loss function, a first weight being allocated to augmented samples from the first group of augmented sample sets to reduce the negative influence; and
training the model to be trained based on the training loss function and the training set,
wherein the training the model comprises iteratively performing forward and back propagation operations between the model and the training loss function until a training loss value is less than a predetermined value.
Step 1: “Is The Claim To A Process, Machine, Manufacture Or Composition Of Matter?”
Claim 1 recites at least one step or act of determining degrees of influence pf a plurality of augmented sample sets in a training set on a model to be trained. Thus, Claim 1 is a process, which is one of the statutory categories of invention.
STEP 2A (1): “Does The Claim Recite An Abstract Idea, Law Of Nature, Or Natural Phenomenon?”
The limitation “determining……... to a plurality of original samples” “determining……... to have a negative influence on the model to be trained” “determining……... to reduce the negative influence”, as drafted, are processes that, under broadest reasonable interpretation, recite Mental Processes which are Abstract Ideas. The determining steps are evaluations that can easily be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).
The limitation “and training the model to be trained based on the training loss function and the training set, wherein the training the model comprises iteratively performing forward and back propagation operations between the model and the training loss function until a training loss value is less than a predetermined value.”, as drafted, is a process that, under broadest reasonable interpretation, recites a Mathematical Concept which is an Abstract Idea. Training loss function is a Mathematical calculation (Para [0078] defines the mathematical training loss function). See MPEP 2106.04(a)(2)(I).
All limitations within Claim 1 are directed to Abstract ideas for being Mental Process and Mathematical Concepts. Thus, PRONG ONE of STEP 2A is satisfied.
STEP 2A (2): “Does The Claim Recite Additional Elements That Integrate The Judicial Exception into A Practical Application?”
Claim 1 does not recite additional elements that integrate the judicial exception into a practical application. See MPEP 2106.04.
Thus, Claim 1 fails PRONG TWO of STEP 2A.
STEP 2B: “Does The Claim Recite Additional Elements That Amount To Significantly More Than The Judicial Exception?”
Claim 1 reciting the abstract idea does not reside additional elements that amount to significantly more than the judicial exception.
Thus, Claim 1 fails STEP 2B.
Claim 1 is not patent eligible.
Regarding dependent claims 2-9 and 19, these claims recite either insignificant extra solution activity or additional abstract ideas, and therefore, also directed toward non-statutory subject matter.
Claims 2-9 and 19 are not patent eligible.
Regarding Claim 10:
Claim 10 recites
An electronic device, comprising: at least one processing circuit configured to:
determine respective degrees of influence of a plurality of augmented sample sets in a training set on a model to be trained, the plurality of augmented sample sets corresponding to a plurality of original samples;
determine, based on the degrees of influence, a first group of augmented sample sets from the plurality of augmented sample sets, the first group of augmented sample sets being to have a negative influence on the model to be trained;
determine a training loss function associated with the training set, in the training loss function, a first weight being allocated to augmented samples from the first group of augmented sample sets to reduce the negative influence; and
train the model to be trained based on the training loss function and the training set by iteratively performing forward and back propagation operations between the model and the training loss function until a training loss value is less than a predetermined value.
Step 1: “Is The Claim To A Process, Machine, Manufacture Or Composition Of Matter?”
Claim 10 recites of an electronic device comprising of at least one processing circuit. Thus, Claim 10 is a machine, which is one of the statutory categories of invention.
STEP 2A (1): “Does The Claim Recite An Abstract Idea, Law Of Nature, Or Natural Phenomenon?”
The limitation “determine……... to a plurality of original samples” “determine……... to have a negative influence on the model to be trained” “determine……... to reduce the negative influence”, as drafted, are processes that, under broadest reasonable interpretation, recite Mental Processes which are Abstract Ideas. The steps to determine…. are an evaluation that can easily be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).
The limitation “and train the model to be trained based on the training loss function and the training set by iteratively performing forward and back propagation operations between the model and the training loss function until a training loss value is less than a predetermined value.”, as drafted, is a process that, under broadest reasonable interpretation, recites a Mathematical Concept which is an Abstract Idea. Training loss function is a Mathematical calculation (Para [0078] defines the mathematical training loss function). See MPEP 2106.04(a)(2)(I).
All limitations within Claim 10 are directed to Abstract ideas for being Mental Process and Mathematical Concepts. Thus, PRONG ONE of STEP 2A is satisfied.
STEP 2A (2): “Does The Claim Recite Additional Elements That Integrate The Judicial Exception into A Practical Application?”
This judicial exception is not integrated into a practical application because in Claim 10 “electronic device…”, “processing circuit...” are recited at a high level of granularity (i.e. computer implemented to performing generic computer functions) such that it amounts no more than mere instructions to apply the exception using a generic computer component. See MPEP 2106.04.
Thus, Claim 10 fails PRONG TWO of STEP 2A.
STEP 2B: “Does The Claim Recite Additional Elements That Amount To Significantly More Than The Judicial Exception?”
The claim limitations reciting the abstract idea do not include additional elements that are sufficient to amount to significantly more than the judicial exception because as discussed above, the additional element of using “An electronic device, comprising: at least one processing circuit configured to:” to perform the abstract idea recited above amounts to no more than mere instructions to apply to the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. Therefore, the claim limitations do not include elements that amount to significantly more.
Thus, Claim 10 fails STEP 2B.
Claim 10 is not patent eligible.
Regarding dependent claims 11-18 and 20, these claims recite either insignificant extra solution activity or additional abstract ideas, and therefore, also directed toward non-statutory subject matter.
Claims 11-18 and 20 are not patent eligible.
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.
Claims 1-8, 10-17 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al. (US 2022/0101040 A1) in view of Sather et al. (US 11,847,567 B1).
As per Claim 1, Zhang discloses A method for data processing (Abstract Line 1 mentions about method for classification i.e. data classification) comprising:
determining respective degrees of influence of a plurality of augmented sample sets in a training set on a model to be trained, the plurality of augmented sample sets corresponding to a plurality of original samples (Para [0027] Line 1-8 mentions about finding a degree of influence on multiple images which are same as sample sets in a training set);
determining, based on the degrees of influence, a first group of augmented sample sets from the plurality of augmented sample sets, the first group of augmented sample sets being to have a negative influence on the model to be trained (Para [0027] talks about contribution calculation unit 104 which is configured to calculate the influence on multiple images. Line 7 mentions about an example for calculating a positive degree of influence of the image on the classification set. Similar logic can be used to calculate sample sets which have a negative influence on the model. Also, "without specific details” of "degree", degree of positive influence could subjectively be construed as also an indication of degree of negative influence.");
determining a training loss function associated with the training set, in the training loss function, a first weight being allocated to augmented samples from the first group of augmented sample sets to reduce the negative influence (Claim 6 Line 21 mentions about using the loss function for at least one sample group which is similar to training sets and Claim 7 Line 2 mentions about the loss function. Para [0045] mentions all about the loss function. Para [0046] Line 14 mentions about weights applied to sample groups);
and training the model to be trained based on the training loss function and the training set until a training loss value is less than a predetermined value. (Para [0025] The pre-trained classification model may be a pre-trained deep learning network model such as a pre-trained convolution neural network model. It is obvious to one of ordinary skill in the art. Para [0042] The pre-trained classification model may be obtained from an initial classification model by extracting a feature for each sample image, calculating a contribution of the sample image to a classification result, aggregating features of sample images to obtain an aggregated feature and training the initial classification model based on the loss function for the initial classification model to meet a predetermined convergence condition. Para [0043] The predetermined convergence condition may be one of: the number of training reaches a predetermined number; the loss function is minimized; and the loss function is less then or equal to a predetermined threshold. Para [0047] mentions about training model based on loss function and training data set).
Zhang does not disclose:
wherein the training the model comprises iteratively performing forward and back propagation operations between the model and the training loss function.
Sather discloses wherein the training the model comprises iteratively performing forward and back propagation operations between the model and the training loss function until a training loss value is less than a predetermined value. (Spec Col 11 Line 8-26 The training model iteratively selects different input value sets with known output value sets. For each selected input value set, the training process typically (1) forward propagates the input value set through the network’s nodes to produce computed output value set and then (2) backpropagates a gradient (rate of change) of a loss function (output error) that quantifies in a particular way the difference between the input set’s known output value set, in order to adjust the network’s configurable parameters. Some embodiments use an alternating direction method of multipliers (ADMM) to train the quantized weight values (which includes performing forward and backward propagation) and ensure that at least a threshold percentage of the weight values are set to zero (interpreted as the loss error being minimized and reaches a predetermined value)).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention given the teachings of Zhang and Sather that a method for training a model which comprises iteratively performing forward and back propagation operations between the model and the training loss function. With Zhang and Sather disclosing machine learning methods for data classification, and with Sather additionally disclosing forward and backward propagation, one of ordinary skill in the art of implementing a method for training a model which comprises iteratively performing forward and back propagation operations between the model and the training loss function because forward and backward propagation, when used in image classification models, allows the network to learn from data by making prediction, measuring its error, and systematically correcting it. One would therefore be motivated to combine these teachings as in doing so would create this method for training a model which comprises iteratively performing forward and back propagation operations between the model and the training loss function.
As per Claim 2, Zhang in view of Sather discloses The method according to claim 1. Zhang additionally discloses wherein determining the degrees of influence of the plurality of augmented sample sets on the model to be trained comprises (Para [0027] Line 1-8 mentions about finding a degree of influence on multiple images which are same as sample sets in a training set):
determining a first loss value based on a first training subset of the training set, the first training subset comprising only the plurality of original samples (Para [0042] mentions about extracting different values (like feature, contribution and aggregating feature of sample image) based on loss function. Similar process can be used to extract first loss value);
determining a second loss value based on a second training subset of the training set, the second training subset comprising the plurality of original samples and at least one augmented sample set of the plurality of augmented sample sets, the at least one augmented sample set corresponding to at least one original sample of the plurality of original samples (Para [0042] mentions about extracting different values (like feature, contribution and aggregating feature of sample image) based on loss function. Similar process can be used to extract second loss value);
and determining a degree of influence of the at least one augmented sample set on the model to be trained based on the first loss value and the second loss value; (Para [0027] Line 1-8 mentions about finding a degree of influence on multiple images which are same as sample sets in a training set. Para [0043] mentions about minimizing the loss function).
As per Claim 3, Zhang in view of Sather discloses The method according to claim 2. Zhang additionally discloses wherein determining the first group of augmented sample sets further comprises: (Para [0027] Line 1-8 mentions about finding a degree of influence on multiple images which are same as sample sets in a training set):
in accordance with a determination that a difference between the first loss value and the second loss value is less than zero, determining the at least one augmented sample set to belong to the first group of augmented sample sets; (Para [0027] Line 1-8 mentions about finding a degree of influence on multiple images which are same as sample sets in a training set. Para [0043] mentions about minimizing the loss function).
and in accordance with a determination that the difference between the first loss value and the second loss value is greater than or equal to zero, determining the at least one augmented sample set to belong to a second group of augmented sample sets, the second group of augmented sample sets being to have a positive influence on the model to be trained. (Para [0027] Line 1-8 mentions about finding a degree of influence on multiple images which are same as sample sets in a training set and the calculation of positive influence on the model).
As per Claim 4, Zhang in view of Sather discloses The method according to claim 3. Zhang additionally discloses wherein determining the difference comprises: determining the difference at least based on a pre-trained model related to the model to be trained, the at least one original sample and the at least one augmented sample set, the pre-trained model being trained using only the plurality of original samples. (Abstract Line 1 mentions about pre-trained classification model. Para [0011] mentions about Fig. 1 showing a pre-trained classification model).
Regarding Claim 5, Zhang in view of Sather discloses The method according to claim 4, wherein determining the difference at least based on the pre-trained model related to the model to be trained (Abstract Line 1 mentions about pre-trained classification model. Para [0011] mentions about Fig. 1 showing a pre-trained classification model).
Zhang does not disclose
the at least one original sample and the at least one augmented sample set further comprises:
determining the difference based on a Hessian matrix, the Hessian matrix being predetermined by using the pre-trained model.
Sather discloses determining the difference based on a Hessian matrix (Spec Col 3 Line 42 mentions about the Hessian matrix and difference between actual output and expected output).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention given the teachings of Zhang and Sather that a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include determining the difference based on a Hessian matrix, the Hessian matrix being predetermined by using the pre-trained model. With Zhang and Sather disclosing machine learning methods for data classification, and with Sather additionally disclosing Hessian matrix, one of ordinary skill in the art of implementing a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include Hessian Matrix because a Hessian matrix, when used in image classification models, offers advantages like capturing local structures, separating edges from stable points and enables faster convergence in optimization particularly in neural networks. One would therefore be motivated to combine these teachings as in doing so would create this method for training a model with determining degrees of influence and a training loss function with a difference in loss values.
Regarding Claim 6, Zhang in view of Sather discloses The method according to claim 1, (Abstract Line 1 mentions about method for classification i.e. data classification).
Zhang does not disclose every aspect of wherein training the model to be trained comprises:
determining, based on the degrees of influence, probabilities that individual augmented samples in the first group of augmented sample sets are selected
determining a training subset from the training set and based on the probabilities;
and training the model to be trained at least based on the training loss function associated with the training subset.
Sather discloses training the model to be trained on the loss function with the sample sets based on the probabilities (Spec Col 17 Line 46 mentions about determining sample distributions based on probability. Spec Col 17 Line 49 mentions about loss function).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention given the teachings of Zhang and Sather that a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include determining, based on the degrees of influence, probabilities that individual augmented samples in the first group of augmented sample sets are selected; determining a training subset from the training set and based on the probabilities; and training the model to be trained at least based on the training loss function associated with the training subset. With Zhang and Sather disclosing machine learning methods for data classification, and with Sather additionally disclosing training the model on training loss function and training subset which is based on probabilities, one of ordinary skill in the art of implementing a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include training the model on training loss function and training subset which is based on probabilities because training the model based on probabilities to determine the sample sets would offer better predictions. One would therefore be motivated to combine these teachings as in doing so would create this method for training a model with determining degrees of influence and a training loss function with a difference in loss values.
Regarding Claim 7, Zhang in view of Sather discloses The method according to claim 6, (Para [0047] mentions about training model based on loss function and training data set).
Zhang does not disclose every aspect of
wherein determining the training loss function further comprises:
for an augmented sample from the first group of augmented sample sets in the training subset, determining the first weight based on the probabilities.
Sather discloses calculating the weight based on probabilities and are used to compute loss matrix values. (Spec Col 17 Line 52 mentions about determining the gradients of all the weights).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention given the teachings of Zhang and Sather that a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include for an augmented sample from the first group of augmented sample sets in the training subset, determining the first weight based on the probabilities. With Zhang and Sather disclosing machine learning methods for data classification, and with Sather additionally disclosing determining gradients of all weights, one of ordinary skill in the art of implementing a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include determining gradients of all weights because determining gradients based on probability in image classification models would offer better predictions than in rigid classifications. One would therefore be motivated to combine these teachings as in doing so would create this method for training a model with determining degrees of influence and a training loss function with a difference in loss values.
Regarding Claim 8, Zhang in view of Sather discloses The method according to claim 1, further comprising: obtaining input data; (Para [0027] LINE 4 mentions about finding a degree of influence on multiple images which are analogous to sample sets in training models. Para [0042] Line 3 mentions about training sample sets).
Zhang does not disclose every aspect of
obtaining input data;
determining a prediction result for the input data by using the trained model.
Sather discloses determining a prediction result for the input data using the trained model. (Spec Col 5 Line 5-25 mentions about determining the process of calculating projected value which is analogous to prediction result).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention given the teachings of Zhang and Sather that a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include determining a prediction result for the input data by using the trained model. With Zhang and Sather disclosing machine learning methods for data classification, and with Sather additionally disclosing determining the projected value, one of ordinary skill in the art of implementing a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include determining the projected value because determining projections in image classification models would offer better efficiency than in rigid classifications. One would therefore be motivated to combine these teachings as in doing so would create this method for training a model with determining degrees of influence and a training loss function with a difference in loss values
As per Claim 10, Zhang discloses An electronic device, comprising: at least one processing circuit configured to: (Abstract Line 1 mentions about method for classification method for data processing i.e. processing unit) comprising:
determine respective degrees of influence of a plurality of augmented sample sets in a training set on a model to be trained, the plurality of augmented sample sets corresponding to a plurality of original samples (Para [0027] Line 1-8 mentions about finding a degree of influence on multiple images which are same as sample sets in a training set);
determine, based on the degrees of influence, a first group of augmented sample sets from the plurality of augmented sample sets, the first group of augmented sample sets being to have a negative influence on the model to be trained (Para [0027] talks about contribution calculation unit 104 which is configured to calculate the influence on multiple images. Line 7 mentions about an example for calculating a positive degree of influence of the image on the classification set. Similar logic can be used to calculate sample sets which have a negative influence on the model. Also, "without specific details” of "degree", degree of positive influence could subjectively be construed as also an indication of degree of negative influence.");
determine a training loss function associated with the training set, in the training loss function, a first weight being allocated to augmented samples from the first group of augmented sample sets to reduce the negative influence (Claim 6 Line 21 mentions about using the loss function for at least one sample group which is similar to training sets and Claim 7 Line 2 mentions about the loss function. Para [0045] mentions all about the loss function. Para [0046] Line 14 mentions about weights applied to sample groups);
and training the model to be trained based on the training loss function and the training set until a training loss value is less than a predetermined value. (Para [0025] The pre-trained classification model may be a pre-trained deep learning network model such as a pre-trained convolution neural network model. It is obvious to one of ordinary skill in the art. Para [0042] The pre-trained classification model may be obtained from an initial classification model by extracting a feature for each sample image, calculating a contribution of the sample image to a classification result, aggregating features of sample images to obtain an aggregated feature and training the initial classification model based on the loss function for the initial classification model to meet a predetermined convergence condition. Para [0043] The predetermined convergence condition may be one of: the number of training reaches a predetermined number; the loss function is minimized; and the loss function is less then or equal to a predetermined threshold. Para [0047] mentions about training model based on loss function and training data set).
Zhang does not disclose:
wherein the training the model comprises iteratively performing forward and back propagation operations between the model and the training loss function.
Sather discloses wherein the training the model comprises iteratively performing forward and back propagation operations between the model and the training loss function until a training loss value is less than a predetermined value. (Spec Col 11 Line 8-26 The training model iteratively selects different input value sets with known output value sets. For each selected input value set, the training process typically (1) forward propagates the input value set through the network’s nodes to produce computed output value set and then (2) backpropagates a gradient (rate of change) of a loss function (output error) that quantifies in a particular way the difference between the input set’s known output value set, in order to adjust the network’s configurable parameters. Some embodiments use an alternating direction method of multipliers (ADMM) to train the quantized weight values (which includes performing forward and backward propagation) and ensure that at least a threshold percentage of the weight values are set to zero (interpreted as the loss error being minimized and reaches a predetermined value)).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention given the teachings of Zhang and Sather that a method for training a model which comprises iteratively performing forward and back propagation operations between the model and the training loss function. With Zhang and Sather disclosing machine learning methods for data classification, and with Sather additionally disclosing forward and backward propagation, one of ordinary skill in the art of implementing a method for training a model which comprises iteratively performing forward and back propagation operations between the model and the training loss function because forward and backward propagation, when used in image classification models, allows the network to learn from data by making prediction, measuring its error, and systematically correcting it. One would therefore be motivated to combine these teachings as in doing so would create this method for training a model which comprises iteratively performing forward and back propagation operations between the model and the training loss function.
As per Claim 11, Zhang in view of Sather discloses The device according to claim 10. Zhang additionally discloses wherein the at least one processing circuit is further configured to: (Abstract Line 1 mentions about method for classification method for data processing i.e. processing unit)
determine a first loss value based on a first training subset of the training set, the first training subset comprising only the plurality of original samples (Para [0042] mentions about extracting different values (like feature, contribution and aggregating feature of sample image) based on loss function. Similar process can be used to extract first loss value);
determine a second loss value based on a second training subset of the training set, the second training subset comprising the plurality of original samples and at least one augmented sample set of the plurality of augmented sample sets, the at least one augmented sample set corresponding to at least one original sample of the plurality of original samples (Para [0042] mentions about extracting different values (like feature, contribution and aggregating feature of sample image) based on loss function. Similar process can be used to extract second loss value);
and determine a degree of influence of the at least one augmented sample set on the model to be trained based on the first loss value and the second loss value; (Para [0027] Line 1-8 mentions about finding a degree of influence on multiple images which are same as sample sets in a training set. Para [0043] mentions about minimizing the loss function)
As per Claim 12, Zhang in view of Sather discloses The device according to claim 11. Zhang additionally discloses wherein the at least one processing circuit is further configured to: (Abstract Line 1 mentions about method for classification method for data processing i.e. processing unit)
in accordance with a determination that a difference between the first loss value and the second loss value is less than zero, determine the at least one augmented sample set to belong to the first group of augmented sample sets; (Para [0027] Line 1-8 mentions about finding a degree of influence on multiple images which are same as sample sets in a training set. Para [0043] mentions about minimizing the loss function).
and in accordance with a determination that the difference between the first loss value and the second loss value is greater than or equal to zero, determine the at least one augmented sample set to belong to a second group of augmented sample sets, the second group of augmented sample sets being to have a positive influence on the model to be trained. (Para [0027] Line 1-8 mentions about finding a degree of influence on multiple images which are same as sample sets in a training set and the calculation of positive influence on the model).
As per Claim 13, Zhang in view of Sather discloses The device according to claim 11. Zhang additionally discloses wherein the at least one processing circuit is further configured to: (Abstract Line 1 mentions about method for classification method for data processing i.e. processing unit)
determine the difference at least based on a pre-trained model related to the model to be trained, the at least one original sample and the at least one augmented sample set, the pre-trained model being trained using only the plurality of original samples. (Abstract Line 1 mentions about pre-trained classification model. Para [0011] mentions about Fig. 1 showing a pre-trained classification model).
Regarding Claim 14, Zhang in view of Sather discloses The device according to claim 13, wherein the at least one processing circuit is further configured to: (Abstract Line 1 mentions about classification method for data processing i.e. processing unit).
Zhang does not disclose
determine the difference based on a Hessian matrix, the Hessian matrix being predetermined by using the pre-trained model.
Sather discloses determining the difference based on a Hessian matrix (Spec Col 3 Line 42 mentions about the Hessian matrix and difference between actual output and expected output).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention given the teachings of Zhang and Sather that a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include determining the difference based on a Hessian matrix, the Hessian matrix being predetermined by using the pre-trained model. With Zhang and Sather disclosing machine learning methods for data classification, and with Sather additionally disclosing Hessian matrix, one of ordinary skill in the art of implementing a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include Hessian Matrix because a Hessian matrix, when used in image classification models, offers advantages like capturing local structures, separating edges from stable points and enables faster convergence in optimization particularly in neural networks. One would therefore be motivated to combine these teachings as in doing so would create this method for training a model with determining degrees of influence and a training loss function with a difference in loss values.
Regarding Claim 15, Zhang in view of Sather discloses The device according to claim 10,
wherein the at least one processing circuit is further configured to: (Abstract Line 1 mentions about classification method for data processing i.e. processing unit).
Zhang does not disclose every aspect to
determine, based on the degrees of influence, probabilities that individual augmented samples in the first group of augmented sample sets are selected
determine a training subset from the training set and based on the probabilities;
and train the model to be trained at least based on the training loss function associated with the training subset.
Sather discloses training the model to be trained on the loss function with the sample sets based on the probabilities (Spec Col 17 Line 46 mentions about determining sample distributions based on probability. Spec Col 17 Line 49 mentions about loss function).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention given the teachings of Zhang and Sather that a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include determining, based on the degrees of influence, probabilities that individual augmented samples in the first group of augmented sample sets are selected; determining a training subset from the training set and based on the probabilities; and training the model to be trained at least based on the training loss function associated with the training subset. With Zhang and Sather disclosing machine learning methods for data classification, and with Sather additionally disclosing training the model on training loss function and training subset which is based on probabilities, one of ordinary skill in the art of implementing a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include training the model on training loss function and training subset which is based on probabilities because training the model based on probabilities to determine the sample sets would offer better predictions. One would therefore be motivated to combine these teachings as in doing so would create this method for training a model with determining degrees of influence and a training loss function with a difference in loss values.
Regarding Claim 16, Zhang in view of Sather discloses The device according to claim 15, wherein the at least one processing circuit is further configured to: (Abstract Line 1 mentions about classification method for data processing i.e. processing unit).
Zhang does not disclose every aspect of
for an augmented sample from the first group of augmented sample sets in the training subset, determine the first weight based on the probabilities.
Sather discloses calculating the weight based on probabilities and are used to compute loss matrix values. (Spec Col 17 Line 52 mentions about determining the gradients of all the weights).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention given the teachings of Zhang and Sather that a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include for an augmented sample from the first group of augmented sample sets in the training subset, determining the first weight based on the probabilities. With Zhang and Sather disclosing machine learning methods for data classification, and with Sather additionally disclosing determining gradients of all weights, one of ordinary skill in the art of implementing a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include determining gradients of all weights because determining gradients based on probability in image classification models would offer better predictions than in rigid classifications. One would therefore be motivated to combine these teachings as in doing so would create this method for training a model with determining degrees of influence and a training loss function with a difference in loss values.
Regarding Claim 17, Zhang in view of Sather discloses The device according to claim 10, wherein the at least one processing circuit is further configured (Abstract Line 1 mentions about classification method for data processing i.e. processing unit).
Zhang does not disclose every aspect to
obtain input data;
determine a prediction result for the input data by using the trained model.
Sather discloses determining a prediction result for the input data using the trained model. (Spec Col 5 Line 5-25 mentions about determining the process of calculating projected value which is analogous to prediction result).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention given the teachings of Zhang and Sather that a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include determining a prediction result for the input data by using the trained model. With Zhang and Sather disclosing machine learning methods for data classification, and with Sather additionally disclosing determining the projected value, one of ordinary skill in the art of implementing a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include determining the projected value because determining projections in image classification models would offer better efficiency than in rigid classifications. One would therefore be motivated to combine these teachings as in doing so would create this method for training a model with determining degrees of influence and a training loss function with a difference in loss values.
Claims 9 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al. (US 2022/0101040 A1), Sather et al. (US 11,847,567 B1) and Filoche et al. (US 2023/0275812 A1).
Regarding Claim 9, Zhang in view of Sather discloses The method according to claim 8, (Zhang Abstract Line 1 mentions about model training and data processing).
Zhang and Sather do not disclose every aspect
wherein the input data is data of an image,
the trained model is one of: an image classification model, a semantic segmentation model and a target recognition model,
and the prediction result is a corresponding one of: an image classification result, a semantic segmentation result and a target recognition result.
Filoche discloses the trained model being an image classification model with an image input, (Para [0269] Line 6 mentions about image as an input and the trained model as image classification model. The semantic segmentation and target model are similar types of models like image classification model to analyze images.) and the prediction result is an output image. (Para [0269] Line 6 mentions about prediction result as an output image).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention given the teachings of Zhang, Sather and Filoche that a method for training a model with determining degrees of influence and a training loss function with a difference in loss values) would include wherein the input data is data of an image, the trained model is one of: an image classification model, a semantic segmentation model and a target recognition model, and the prediction result is a corresponding one of: an image classification result, a semantic segmentation result and a target recognition result. With Zhang, Sather and Filoche disclosing machine learning methods for data classification, and with Filoche additionally disclosing an image classification model or semantic segmentation model or a target recognition model one of ordinary skill in the art of implementing a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include image classification model or semantic segmentation model or a target classification model because these models would help to automate tasks, improve accuracy and efficiency and can be used across various industries especially when the inputs are images. One would therefore be motivated to combine these teachings as in doing so would create this method for training a model with determining degrees of influence and a training loss function with a difference in loss values.
Regarding Claim 18, Zhang in view of Sather discloses The device according to claim 17, (Zhang Abstract Line 1 mentions about model training and data processing).
Zhang and Sather do not disclose every aspect
wherein the input data is data of an image,
the trained model is one of: an image classification model, a semantic segmentation model and a target recognition model,
and the prediction result is a corresponding one of: an image classification result, a semantic segmentation result and a target recognition result.
Filoche discloses the trained model being an image classification model with an image input, (Para [0269] Line 6 mentions about image as an input and the trained model as image classification model. The semantic segmentation and target model are similar types of models like image classification model to analyze images.) and the prediction result is an output image. (Para [0269] Line 6 mentions about prediction result as an output image).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention given the teachings of Zhang, Sather and Filoche that a method for training a model with determining degrees of influence and a training loss function with a difference in loss values) would include wherein the input data is data of an image, the trained model is one of: an image classification model, a semantic segmentation model and a target recognition model, and the prediction result is a corresponding one of: an image classification result, a semantic segmentation result and a target recognition result. With Zhang, Sather and Filoche disclosing machine learning methods for data classification, and with Filoche additionally disclosing an image classification model or semantic segmentation model or a target recognition model one of ordinary skill in the art of implementing a method for training a model with determining degrees of influence and a training loss function with a difference in loss values would include image classification model or semantic segmentation model or a target classification model because these models would help to automate tasks, improve accuracy and efficiency and can be used across various industries especially when the inputs are images. One would therefore be motivated to combine these teachings as in doing so would create this method for training a model with determining degrees of influence and a training loss function with a difference in loss values.
Claims 19 is rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al. (US 2022/0101040 A1), Sather et al. (US 11,847,567 B1) and Filoche et al. (US 2023/0275812 A1) in further view of Risman et al. (US 2018/0082443 A1).
Regarding Claim 19, Zhang in view of Sather in view of Filoche discloses The method according to claim 9, (Zhang Abstract Line 1 mentions about model training and data processing).
Zhang and Sather and Filoche do not disclose every aspect
comprising performing image cropping, rotation and flipping on the image to obtain an augmented training set among the plurality of augmented sample sets.
Risman discloses comprising performing image cropping, rotation and flipping on the image to obtain an augmented training set among the plurality of augmented sample sets. (Para [0071] Line 1-8 mentions about in S830, the slice-level model is trained on the training images and labels and validated on the validation images and labels for a set number of epochs. Every X epochs, augmentation can be applied to the training images. Examples of augmentation include, without limitation, one or more of: rotating, translating, and/or flipping the images by a random amount, randomly zooming or cropping images, and/or applying a shear transformation.)
It would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed inventi