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 .
Response to Amendment
This Office Action is in response to applicant’s communication filed 30 March 2026, in response to the Office Action mailed 7 January 2026. The applicant’s remarks and any amendments to the claims or specification have been considered, with the results that follow.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1, 5, 9, 12, 13, and 16-19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chang et al. (Prognostics-Based LED Qualification Using Similarity-Based Statistical Measure with RVM Regression Model, July 2017, pgs. 5667-5677) in view of Dillman (WO 2021/211787).
As per claim 1, Chang teaches a model training method, comprising: acquiring a training sample set comprising training sample design data and training sample test data [The developed method predicts LED remaining useful life (RUL) by calculating the accumulated sum of products of similarity weights and historical LED RUL values at the 210th hour. Specifically, a similarity weight, defined as the degree of affinity between two different LED’s degradation trends, is derived from the difference between a test unit’s degradation trend and a training unit’s degradation trend. Likewise, the RVM is used to represent a unit’s degradation behavior (pg. 5667, abstract; etc.), which includes training the RVM using training samples for test condition (design) and test result (test) data (pg. 5669, fig. 1; pg. 5671, section III; etc.); This prediction process can utilize any training data set that was conducted for all devices. This means that all historical data sets can be used for RUL prediction (pg. 5670, section II.B; etc.)]; wherein the training sample design data comprises design data of a training sample display device, the training sample test data comprising test data of the training sample display device [The developed method predicts LED remaining useful life (RUL) by calculating the accumulated sum of products of similarity weights and historical LED RUL values at the 210th hour. Specifically, a similarity weight, defined as the degree of affinity between two different LED’s degradation trends, is derived from the difference between a test unit’s degradation trend and a training unit’s degradation trend. Likewise, the RVM is used to represent a unit’s degradation behavior (pg. 5667, abstract; etc.), which includes training the RVM using training samples for test condition (design) and test result (test) data (pg. 5669, fig. 1; pg. 5671, section III; etc.); This prediction process can utilize any training data set that was conducted for all devices. This means that all historical data sets can be used for RUL prediction (pg. 5670, section II.B; etc.); where the test condition data (e.g., junction temperatures, forward voltage, etc.) is the training sample design data, and the degradation in light output and color during testing is the training sample test data; of the LED (training sample display device)]; inputting the training sample design data into a model to be trained, and training the model to be trained according to an output of the model to be trained and the training sample test data to obtain an initial prediction model [the test condition and test output data is acquired during testing (pg. 5671, table 1; etc.) and used as training samples to train the RVM (model to be trained) during training, which produces an RUL prediction (pg. 5669, section II.A and fig. 1; etc.) ; where the test condition data (e.g., junction temperatures, forward voltage, etc.) is the training sample design data, and the degradation in light output and color during testing is the training sample test data; of the LED (training sample display device)]; and determining the initial prediction model as a performance prediction model when the initial prediction model satisfies a pre-set condition [the model is trained over a specified training set, Li, for a specified number of training samples for K LEDs (pg. 5671, section II.B; etc.); where training completing after a specified number of iterations/samples is the satisfying a pre-set condition]; wherein the performance prediction model is used for predicting the performance data of the target display device according to the design data of the display device [the RVM is used to make a RUL (performance) prediction of the LED (target display device) according to the similarity of the device and test conditions to the test datasets (pg. 5669, fig. 1; etc.)].
While Chang teaches training a model as well as contemplating a fusion of multiple types of models (see, e.g. Chang: pg. 5668, second paragraph in the right column), it has not been relied upon for teaching wherein the model to be trained is a fully-connected neural network or a transformer model, and there is at least one skip connection between different network levels of a fully-connected layer of the model to be trained; wherein the at least one skip connection is used for inputting output values of network levels separated by at lest two layers into a pre-set network layer after being fused; the pre-set network layer is a deep network separated from a fused network layer by at least three layers.
Dillman teaches wherein the model to be trained is a fully-connected neural network or a transformer model [an ensemble of learning models can be used to improve prediction performance, including fully-connected neural networks and SVM models, etc., (paras. 0067, 0087, etc.), for the vector machine model of Chang, above], and there is at least one skip connection between different network levels of a fully-connected layer of the model to be trained; wherein, the at least one skip connection is used for inputting output values of network levels separated by at least two layers into a pre-set network layer after being fused; the pre-set network layer is a deep network separated from a fused network layer by at least three layers [the neural network can include multiple skip connections between different fully connected layers and fusion blocks of the neural network(s), where different channels may be fused (Dillman: paras. 0081, 0090, etc.) which can skip more than 3 layers (see, e.g., Dillman: figs. 5, 11, etc., which show skipping more than 3 layers and connecting to the fusion network layers)]
Chang and Dillman are analogous art, as they are within the same field of endeavor, namely improving performance prediction of machine learning models.
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to use an ensemble of models including SVM and fully-connected neural networks, as well as skip connections connecting separated layers of the network, as taught by Dillman, in place of- or in combination with the machine learning model of the system taught by Chang.
Dillman provides motivation as [ensemble learning models improve the prediction performance (para. 0087, etc.), the skip connections can provide summation operations and concatenation of features (para. 0081, etc.), and fusion layers can be used to combine different data channels (paras. 0090, 0097-99, etc.)].
As per claim 5, Chang/Dillman teaches wherein the step of inputting training sample design data into a model to be trained, and training the model to be trained according to an output of the model to be trained and the training sample test data comprises: inputting the output of the model to be trained and the training sample test data into a pre-set loss function to obtain a loss value; and aiming at minimizing the loss value, and adjusting parameters of the model to be trained [Specifically, a similarity weight, defined as the degree of affinity between two different LED’s degradation trends, is derived from the difference between a test unit’s degradation trend and a training unit’s degradation trend. Likewise, the RVM is used to represent a unit’s degradation behavior (Chang: pg. 5667, abstract; etc.), which includes training the RVM using training samples for test condition (design) and test result (test) data (Chang: pg. 5669, fig. 1; pg. 5671, section III; etc.); This prediction process can utilize any training data set that was conducted for all devices. This means that all historical data sets can be used for RUL prediction (Chang: pg. 5670, section II.B; etc.); where the model can be trained recursively (i.e., inputting the output of the model, to the model) (Chang: pg. 5668, section I; pg. 5669, section II; etc.), and includes adjusting model parameters by minimizing a predetermined error (loss) function (Chang: pgs. 5669-5670, section II.A; pgs. 5672-5673, section IV.A; pg. 5675, table III; etc.)].
As per claim 9, Chang/Dillman teaches wherein the design data of the training sample display device comprises at least one of: material data of the training sample display device, structural data of the training sample display device, pixel design data of the training sample display device, and process data of the training sample display device [training the RVM using training samples for test condition (design) and test result (test) data (Chang: pg. 5669, fig. 1; pg. 5671, section III; etc.); This prediction process can utilize any training data set that was conducted for all devices. This means that all historical data sets can be used for RUL prediction (Chang: pg. 5670, section II.B; etc.); where the test condition data (e.g., junction temperatures, forward voltage, etc.) is the process data of the training sample display device]; and the test data of the training sample display device comprises at least one of: a quantum dot spectrum of the training sample display device, a half-peak width of the training sample display device, a blue light absorption spectrum of the training sample display device, a color shift of the training sample display device, a luminance decay of the training sample display device, a luminance of the training sample display device, a color gamut of the training sample display device, an external quantum efficiency of the training sample display device, and a lifetime of the training sample display device [training the RVM using training samples for test condition (design) and test result (test) data, including aging tests were conducted under various test conditions to observe degradation trends in both light output (i.e., lumen maintenance) and color (i.e., seven-step SDCM). (Chang: pg. 5669, fig. 1; pg. 5671, section III; etc.); This prediction process can utilize any training data set that was conducted for all devices. This means that all historical data sets can be used for RUL prediction (Chang: pg. 5670, section II.B; etc.); where the degradation in light output (i.e., lumen maintenance) and color (i.e., seven-step SDCM) includes at least luminance, luminance decay, a color gamut, and a lifetime of the training sample display device].
As per claim 12, Chang/Dillman teaches a performance prediction method, comprising: acquiring design data of a target display device; and inputting design data of the target display device into a performance prediction model to obtain test data of the target display device; wherein the performance prediction model is trained by using the model training method according to claim 1 [The developed method predicts LED remaining useful life (RUL) by calculating the accumulated sum of products of similarity weights and historical LED RUL values at the 210th hour. Specifically, a similarity weight, defined as the degree of affinity between two different LED’s degradation trends, is derived from the difference between a test unit’s degradation trend and a training unit’s degradation trend. Likewise, the RVM is used to represent a unit’s degradation behavior (Chang: pg. 5667, abstract; etc.), which includes training the RVM using training samples for test condition (design) and test result (test) data (Chang: pg. 5669, fig. 1; pg. 5671, section III; etc.); This prediction process can utilize any training data set that was conducted for all devices. This means that all historical data sets can be used for RUL prediction (Chang: pg. 5670, section II.B; etc.); where the test condition data (e.g., junction temperatures, forward voltage, etc.) is the training sample design data, and the degradation in light output and color during testing is the training sample test data; of the LED (training sample display device].
As per claim 13, Chang/Dillman teaches wherein the method further comprises: determining target design data as target hardware design data when the test data of the target display device is higher than a preset performance threshold [the RUL of the LED is predicted based on specified failure criteria (Chang: pg. 5671, section III and fig. 4; pg. 5672, fig. 5; etc.); where the target LED is therefore target hardware design data for the specified time period based on the performance remaining above the specified failure criteria (threshold)].
As per claim 16, Chang/Dillman teaches a computing processing device, comprising: a memory in which a computer-readable code is stored; and one or more processors, wherein when the computer-readable code is executed by the one or more processors, the computing processing device performs the method according to claim 1 [the system can be implemented on a number of computers necessary to perform the method (Chang: pg. 5671, fig. 3 and section III; etc.), which would necessarily include memory storing instructions being executed by one or more processors].
As per claim 17, Chang/Dillman teaches a non-transitory computer-readable medium, having computer-readable code stored thereon which, when run on a computing processing device, causes the computing processing device to perform the method according to claim 1 [the system can be implemented on a number of computers necessary to perform the method (Chang: pg. 5671, fig. 3 and section III; etc.), which would necessarily include memory storing instructions being executed by one or more processors].
As per claim 18, Chang/Dillman teaches a computing processing device, comprising: a memory in which a computer-readable code is stored; and one or more processors, wherein when the computer-readable code is executed by the one or more processors, the computing processing device performs the method according to claim 12 [the system can be implemented on a number of computers necessary to perform the method (Chang: pg. 5671, fig. 3 and section III; etc.), which would necessarily include memory storing instructions being executed by one or more processors].
As per claim 19, Chang/Dillman teaches a non-transitory computer-readable medium, having computer-readable code stored thereon which, when run on a computing device, causes the computing device to perform the method according to claim 12 [the system can be implemented on a number of computers necessary to perform the method (Chang: pg. 5671, fig. 3 and section III; etc.), which would necessarily include memory storing instructions being executed by one or more processors].
Claim(s) 2, 3, and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chang et al. (Prognostics-Based LED Qualification Using Similarity-Based Statistical Measure with RVM Regression Model, July 2017, pgs. 5667-5677), in view of Dillman (WO 2021/211787), and further in view of Nayar (US 2021/0365775).
As per claim 2, Chang/Dillman teaches wherein the step of acquiring the training sample set comprises: acquiring design data of the training sample display device and test data of the training sample display device [Likewise, the RVM is used to represent a unit’s degradation behavior (Chang: pg. 5667, abstract; etc.), which includes training the RVM using training samples for test condition (design) and test result (test) data (Chang: pg. 5669, fig. 1; pg. 5671, section III; etc.); This prediction process can utilize any training data set that was conducted for all devices. This means that all historical data sets can be used for RUL prediction (Chang: pg. 5670, section II.B; etc.); where the test condition data (e.g., junction temperatures, forward voltage, etc.) is the training sample design data, and the degradation in light output and color during testing is the training sample test data; of the LED (training sample display device)].
While Chang/Dillman teaches multiple steps for collecting design and test data and training the model (see above), it has not been relied upon for teaching pre-processing same; and performing One-Hot encoding on pre-processed design data and pre-processed test data, respectively, to obtain the training sample design data and the training sample test data.
Nayar teaches collecting data and pre-processing same [the system can include a data manipulator unit that processes/encodes input data being provided to the machine learning model(s) (fig. 2; etc.), which is pre-processing the input data]; and performing One-Hot encoding on pre-processed design data and pre-processed test data, respectively, to obtain the training sample design data and the training sample test data [the data manipulator can include encoders, which include one-hot encoding components that perform one-hot encoding on the input data (paras. 0038, 0052, 0063; figs. 2, 6, 7B; etc.); for the training sample design and test data of Chang, above].
Chang/Dillman and Nayar are analogous art, as they are within the same field of endeavor, namely using machine learning models to determine properties of input data and provide predictions.
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to utilize an encoding unit/stage on the received input data for the ML model(s), as taught by Nayar, for the training sample design and test data used by the ML model in the system taught by Chang/Dillman.
Nayar provides motivation as [the data manipulator allows conversion of the input data into a required/desired input format(s) for the machine learning model(s) (abstract; paras. 0029-30; etc.)].
As per claim 3, Chang/Dillman/Nayar teaches wherein the step of performing One-Hot encoding on pre-processed design data and pre-processed test data, respectively, to obtain the training sample design data and the training sample test data comprises: performing One-Hot fixed value encoding on the pre-processed design data, and encoding fixed value data corresponding to the pre-processed design data as the training sample design data [the data manipulator can include encoders, which include one-hot encoding components that perform fixed length and/or quantization one-hot encoding on the input data (Nayar: paras. 0038, 0052, 0063; figs. 2, 6, 7B; etc.) for training the RVM using training samples for test condition (design) and test result (test) data (Chang: pg. 5669, fig. 1; pg. 5671, section III; etc.); This prediction process can utilize any training data set that was conducted for all devices. This means that all historical data sets can be used for RUL prediction (Chang: pg. 5670, section II.B; etc.); where the test condition data (e.g., junction temperatures, forward voltage, etc.) is the training sample design data, and the degradation in light output and color during testing is the training sample test data; of the LED (training sample display device)]; and in response to the pre-processed test data being fixed value data, performing One-Hot fixed value encoding on the pre-processed test data; in response to the pre-processed test data being quantized data, performing One-Hot quantization encoding on the pre-processed test data; encoding fixed value data and/or encoding quantized data corresponding to the pre-processed test data as the training sample test data [the data manipulator can include encoders, which include one-hot encoding components that perform fixed length and/or quantization one-hot encoding on the input data (Nayar: paras. 0038, 0052, 0063; figs. 2, 6, 7B; etc.) for training the RVM using training samples for test condition (design) and test result (test) data (Chang: pg. 5669, fig. 1; pg. 5671, section III; etc.); This prediction process can utilize any training data set that was conducted for all devices. This means that all historical data sets can be used for RUL prediction (Chang: pg. 5670, section II.B; etc.); where the test condition data (e.g., junction temperatures, forward voltage, etc.) is the training sample design data, and the degradation in light output and color during testing is the training sample test data; of the LED (training sample display device].
As per claim 20, Chang/Dillman/Nayar teaches wherein a total number of the One-Hot fixed value encoding and the One-Hot quantization encoding is equal to the number of network channels of the fully-connected layer of the model to be trained [the data manipulator can include encoders, which include one-hot encoding components that perform fixed length and/or quantization one-hot encoding on the input data (Nayar: paras. 0038, 0052, 0063; figs. 2, 6, 7B; etc.) for training the RVM using training samples for test condition (design) and test result (test) data (Chang: pg. 5669, fig. 1; pg. 5671, section III; etc.); where the fixed length is selected based upon the size/number of channels (see, e.g., Nayar: paras. 0025, 0030, 0052, etc.)].
Claim(s) 4 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chang, Dillman, and Nayar as applied to claim 2 above, further in view of Sasson (US 2021/0034021), and further in view of Tuzzolino (US 11,809,727).
As per claim 4, Chang/Dillman/Nayar teaches wherein acquiring design data of the training sample display device and test data of the training sample display device and pre-processing same comprises: unifying formats and standards of the design data and the test data after the data association [the data manipulator allows conversion of the input data into a required/desired input format(s) for the machine learning model(s) (Nayar: abstract; paras. 0029-30; etc.)].
While Chang/Dillman/Nayar teaches pre-processing the data to produce a desired format/standard (see above), it has not been relied upon for teaching wherein acquiring design data of the training sample display device and test data of the training sample display device and pre-processing same comprises: performing clustering processing on the design data and the test data, so that data formats of the same type of design data are the same, and data formats of the same type of test data are the same; removing erroneous data and duplicated data in the design data and the test data after the clustering processing, and obtaining missing data in the design data and the test data after the clustering processing to obtain complete design data and complete test data; [and] normalizing the complete design data and the complete test data to unify the data scale of the design data and the test data, and performing data association on the design data and the test data after the unification of the data scale
Sasson teaches wherein acquiring design data of the training sample display device and test data of the training sample display device and pre-processing same comprises: performing clustering processing on the design data and the test data, so that data formats of the same type of design data are the same, and data formats of the same type of test data are the same [the input data for the machine learning model(s) can be preprocessed to make structural changes, and can include clustering, formatting, imputing missing values, feature adding/removal, normalization, binning, anomaly detection, etc. (paras. 0101, 0112, 0126, 0137, 0141, etc.)]; removing erroneous data in the design data and the test data after the clustering processing, and obtaining missing data in the design data and the test data after the clustering processing to obtain complete design data and complete test data [the input data for the machine learning model(s) can be preprocessed to make structural changes, and can include clustering, formatting, imputing missing values, feature adding/removal, normalization, binning, anomaly detection, etc. (paras. 0101, 0112, 0126, 0137, 0141, etc.); which includes removing erroneous data and obtaining missing data, to obtain complete data, for the design and test data of Chang/Nayar, above]; [and] normalizing the complete design data and the complete test data to unify the data scale of the design data and the test data, and performing data association on the design data and the test data after the unification of the data scale [the input data for the machine learning model(s) can be preprocessed to make structural changes, and can include clustering, formatting, imputing missing values, feature adding/removal, normalization, binning, anomaly detection, etc. (paras. 0101, 0112, 0126, 0137, 0141, etc.); which includes normalization and formatting to unify the data scale, for the design and test data of Chang/Nayar, above].
Chang/Dillman/Nayar and Sasson are analogous art, as they are within the same field of endeavor, namely optimizing machine learning model performance.
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention to include the preprocessing operations, including clustering, formatting, imputing missing values, feature adding/removal, normalization, binning, anomaly detection, etc., taught by Sasson, in the preprocessing of design and test data in the system taught by Chang/Dillman/Nayar.
Sasson provides motivation as [preprocessing allows for feature engineering (para. 0112, etc.) which can format and prepare data for specified models or systems and improve model performance (para. 0058, 0137-139, etc.)].
While Chang/Dillman/Nayar/Sasson teaches preprocessing the input data, including erroneous data removal (see above), it has not been relied upon for teaching removing duplicated data in the design data and the test data.
Tuzzolino teaches removing erroneous data and duplicated data in the design data and the test data [Data deduplication is a data compression technique whereby duplicate copies of repeating data are eliminated. Through the use of data deduplication techniques, a unique chunk of data (e.g., the master copy) may be stored once in the storage system (402) and all additional occurrences of the chunk of data are replaced with a small reference that points to the stored chunk (col. 44, line 64 to col. 45, line 3; etc.) for the design and test data of Chang/Nayar/Sasson, above].
Chang/Dillman/Nayar/Sasson and Tuzzolino are analogous art, as they are within the same field of endeavor, namely using machine learning models to make predictions for device failures.
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to include the preprocessing of device data, including removing of duplicate data items, as taught by Tuzzolino, in the preprocessing of the device data including removing erroneous data items, etc., in the system taught by Chang/Dillman/Nayar/Sasson.
Tuzzolino provides motivation as [deduplication may correct multiple copies of data produces by data sampling from multiple devices, etc. (col. 30, lines 8-39; etc.) and allows for compression to reduce the size (and resource usage, etc.) of the data elements (col. 44, line 60 to col. 45, line 8; etc.)].
Claim(s) 6 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chang et al. (Prognostics-Based LED Qualification Using Similarity-Based Statistical Measure with RVM Regression Model, July 2017, pgs. 5667-5677), in view of Dillman (WO 2021/211787), and further in view of Chia-Yen et al. (Data science framework for variable selection, metrology prediction, and process control in TFT-LCD manufacturing, Feb 2019, pgs. 76-87).
As per claim 6, Chang/Dillman teaches the model training method according to claim 5.
While Chang/Dillman teaches using an appropriate loss function during training (see above), it has not been relied upon for teaching wherein the loss function is:
PNG
media_image1.png
43
162
media_image1.png
Greyscale
wherein Loss is the loss value, Y is the training sample test data, Y’ is an output value of the model to be trained, and n is a number of iterations.
Chia-Yen teaches wherein the loss function is:
PNG
media_image1.png
43
162
media_image1.png
Greyscale
wherein Loss is the loss value, Y is the training sample test data, Y’ is an output value of the model to be trained, and n is a number of iterations [the model configuration can be optimized during training according to the mean squared error (MSE) and R-squared (pg. 80, section 3.3; pg. 81, section 3.4; etc.), where the loss function formula given is the mean squared error (MSE) function].
Chang/Dillman and Chia-Yen are analogous art, as they are within the same field of endeavor, namely using machine learning models to make predictions for display devices.
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to utilize the MSE loss function taught by Chia-Yen, with the loss function utilized in the system taught by Chang/Dillman, for optimization of the machine learning model(s).
Chia-Yen provides motivation as [the MSE allows for building a robust model by calculating the gap between real and predicted values while reducing the computational burden (pg. 80, section 3.3; etc.)].
Claim(s) 10 and 11 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chang et al. (Prognostics-Based LED Qualification Using Similarity-Based Statistical Measure with RVM Regression Model, July 2017, pgs. 5667-5677), in view of Dillman (WO 2021/211787), and further in view of well-known practices in the art.
As per claim 10, Chang/Dillman teaches wherein the step of determining the initial prediction model as a performance prediction model when the initial prediction model satisfies a preset condition comprises: inputting test sample design data into the initial prediction model to obtain initial prediction data; wherein the test sample design data is design data of a test sample display device [Specifically, a similarity weight, defined as the degree of affinity between two different LED’s degradation trends, is derived from the difference between a test unit’s degradation trend and a training unit’s degradation trend. Likewise, the RVM is used to represent a unit’s degradation behavior (Chang: pg. 5667, abstract; etc.), which includes training the RVM using training samples for test condition (design) and test result (test) data (Chang: pg. 5669, fig. 1; pg. 5671, section III; etc.); This prediction process can utilize any training data set that was conducted for all devices. This means that all historical data sets can be used for RUL prediction (initial prediction data) (Chang: pg. 5670, section II.B; etc.); where the test condition data is test sample design data]; obtaining a determined result according to an error value of the initial prediction data with respect to test sample test data, comprising: [Specifically, a similarity weight, defined as the degree of affinity between two different LED’s degradation trends, is derived from the difference between a test unit’s degradation trend and a training unit’s degradation trend. Likewise, the RVM is used to represent a unit’s degradation behavior (Chang: pg. 5667, abstract; etc.), which includes training the RVM using training samples for test condition (design) and test result (test) data (Chang: pg. 5669, fig. 1; pg. 5671, section III; etc.); This prediction process can utilize any training data set that was conducted for all devices. This means that all historical data sets can be used for RUL prediction (Chang: pg. 5670, section II.B; etc.); where the model can be trained recursively (i.e., inputting the output of the model, to the model) (Chang: pg. 5668, section I; pg. 5669, section II; etc.), and includes adjusting model parameters by minimizing a predetermined error (loss) function (Chang: pgs. 5669-5670, section II.A; pgs. 5672-5673, section IV.A; pg. 5675, table III; etc.)]; wherein the test sample test data is test data of the test sample display device [Specifically, a similarity weight, defined as the degree of affinity between two different LED’s degradation trends, is derived from the difference between a test unit’s degradation trend and a training unit’s degradation trend. Likewise, the RVM is used to represent a unit’s degradation behavior (Chang: pg. 5667, abstract; etc.), which includes training the RVM using training samples for test condition (design) and test result (test) data (Chang: pg. 5669, fig. 1; pg. 5671, section III; etc.); This prediction process can utilize any training data set that was conducted for all devices. This means that all historical data sets can be used for RUL prediction (initial prediction data) (Chang: pg. 5670, section II.B; etc.); where the test result data is the test data of the test sample device (the LED(s))]; obtaining a prediction accuracy rate of the initial prediction model according to at least one determined result [an error percentage is calculated for the predictions of the model (Chang: pg. 5674, table II; etc.); which is a prediction accuracy rate]; and determining the initial prediction data as a performance prediction model [the RVM is used to make a RUL (performance) prediction of the LED (target display device) according to the similarity of the device and test conditions to the test datasets (Chang: pg. 5669, fig. 1; etc.); and an error percentage is calculated for the predictions of the model (Chang: pg. 5674, table II; etc.); which is a prediction accuracy rate].
While Chang/Dillman teaches a specified training phase for the machine learning model, including calculating an error value of the prediction data (see above), it has not been relied upon for teaching when the error value of the initial prediction data with respect to the test sample test data is less than or equal to a first pre-set threshold, determining that the initial prediction model predicts accurately, otherwise determining that the initial prediction model predicts incorrectly; and determining the initial prediction data as a performance prediction model when the prediction accuracy rate is greater than or equal to a second pre-set threshold value.
However, the examiner takes official notice that utilizing pre-determined accuracy/error metric thresholds to determine when a model has completed training or can be used for inference is old and well-known within the art. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention to, when the error value of the initial prediction data with respect to the test sample test data is less than or equal to a first pre-set threshold, determining that the initial prediction model predicts accurately, otherwise determining that the initial prediction model predicts incorrectly; and determining the initial prediction data as a performance prediction model when the prediction accuracy rate is greater than or equal to a second pre-set threshold value, in the system taught by Chang, utilizing the error rate calculations of Chang, to achieve the predictable result of ensuring that the model is sufficiently accurate before utilizing its predictions, and provide a desired tradeoff between resource usage/training time and model accuracy.
As per claim 11, Chang/Dillman teaches, after the step of determining that the initial prediction model predicts accurately, further comprising: regarding the test sample design data as the training sample design data, regarding the test sample test data as the training sample test data, and updating the training sample set; and training the performance prediction model according to an updated training sample set [This method recursively updates the health degradation data with the online prognostics result in real time. (Chang: pg. 5669, section II and fig. 1; etc.) which can also include recursively updating the model with online updating (Chang: pg. 5668; section I; etc.)].
Response to Arguments
The objections to the drawings and to the title have been withdrawn due to the amendments filed.
The rejections of claims 10, 11, 16, and 18 under 35 U.S.C. 112(b) have been withdrawn due to the amendments filed.
Applicant's arguments filed 30 March 2026 have been fully considered but they are not persuasive.
Applicant argues that the cited art does not teach wherein the model to be trained is a fully-connected neural network or a transformer model, and there is at least one skip connection between different network levels of a fully-connected layer of the model to be trained; wherein, the at least one skip connection is used for inputting output values of network levels separated by at least two layers into a pre-set network layer after being fused; the pre-set network layer is a deep network separated from a fused network layer by at least three layers
However, as described above, Dillman teaches an ensemble of learning models can be used to improve prediction performance, including fully-connected neural networks and SVM models, etc., (paras. 0067, 0087, etc.), for the vector machine model of Chang, above, and where the neural network can include multiple skip connections between different fully connected layers and fusion blocks of the neural network(s), where different channels may be fused (Dillman: paras. 0081, 0090, etc.) which can skip more than 3 layers (see, e.g., Dillman: figs. 5, 11, etc., which show skipping more than 3 layers and connecting to the fusion network layers). Here the skip connections are skipping at least 3 layers in the fully connected network layers, providing outputs to the fusion network layers.
Examiner also notes that the applicant has not traversed examiner’s assertion of official notice. Therefore, the common knowledge or well-known in the art statement is taken to be admitted prior art. See MPEP § 2144.03(C).
Conclusion
The following is a summary of the treatment and status of all claims in the application as recommended by M.P.E.P. 707.07(i): claims 7, 8, 14, and 15 are cancelled; claims 1-6, 9-13, and 16-20 are rejected.
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Kim (US 2018/0150609) – discloses a system for predicting health trends, which includes various pre-processing operations, including normalization, clustering according to data types, etc.
Kubota (US 2022/0147829) – discloses a system/method of analysis utilizing machine learning models, and includes preprocessing operations such as one-hot vectorization and clustering, etc.
Kubo (US 2019/0369598) – discloses a system/method for predicting life of consumable device components, including LCDs, etc.
Huang (CN 113988389-A and US 2024/0232477) – discloses a system/method for predicting performance of LEDs utilizing machine learning models.
Liu et al. (Lifetime prediction of a multi-chip high-power LED light source based on artificial neural networks, Nov 2018, pgs. 361-367) – discloses utilizing an ANN to make lifetime predictions for LEDs based upon test and design data of the LED(s).
Orhan et al. (Skip Connections Eliminate Singularities, March 2018, pgs. 1-22) – discloses skip connections between layers in a fully connected neural network.
The examiner requests, in response to this Office action, that support be shown for language added to any original claims on amendment and any new claims. That is, indicate support for newly added claim language by specifically pointing to page(s) and line number(s) in the specification and/or drawing figure(s). This will assist the examiner in prosecuting the application.
When responding to this office action, Applicant is advised to clearly point out the patentable novelty which he or she thinks the claims present, in view of the state of the art disclosed by the references cited or the objections made. He or she must also show how the amendments avoid such references or objections. See 37 CFR 1.111(c).
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any 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 GEORGE GIROUX whose telephone number is (571)272-9769. The examiner can normally be reached M-F 10am-6pm.
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, Omar Fernandez Rivas can be reached at 571-272-2589. 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.
/GEORGE GIROUX/Primary Examiner, Art Unit 2128
Prosecution Insights