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 . 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.
Status of the Claims
The pending claims in the present application are claims 1, 4, 5, 11, 12, 14-16, 19, 20, 22, 24-26, and 29-35 of the Amendment dated 11 February 2026.
Claim Objections
Claim 22 is objected to because of the following informalities: the claims appears to be missing the word “output” in the middle of the phrase “first comprises” (see lines 1 and 2 of the claim). Appropriate correction is required.
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.
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.
Claims 1, 4, 5, 11, 14-16, 19, 20, 22, 24, 25, and 29-34 are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Pat. App. Pub. No. 2023/0017316 A1 to Kulkarni et al. (hereinafter referred to as “Kulkarni”), further in view of Iliadis, Dimitrios, Bernard De Baets, and Willem Waegeman. "Multi-target prediction for dummies using two-branch neural networks." Machine Learning 111.2 (2022): 651-684 (hereinafter referred to as “Iliadis”), further in view of Lin, Liyu, et al. "TB-NET: A Two-Branch Neural Network for Direction of Arrival Estimation under Model Imperfections." Electronics 11.2 (2022): 220 (hereinafter referred to as “Lin”), and further in view of Jennings, Connor, Dazhong Wu, and Janis Terpenny. "Forecasting obsolescence risk and product life cycle with machine learning." IEEE Transactions on Components, Packaging and Manufacturing Technology 6.9 (2016): 1428-1439. (hereinafter referred to as “Jennings”).
Regarding claim 1, Kulkarni discloses the following limitations:
“A method comprising: receiving, by a computing device, information regarding a product from another computing device; ...” - Kulkarni discloses, “As shown in FIG. 1A, and by reference number 105, the prediction system may receive, from the user devices, project management data identifying a project management life cycle associated with development of a software product. The user devices may be associated with developers of the software product” (para. [0016]). A prediction system receiving project management data for a software product from a user device, in Kulkarni, reads on the recited limitation.
“... determining, by the computing device, a set of features from the information regarding the product that are relevant to an execution outcome and a lifespan of the product; ...” - See the aspects of Kulkarni that have been cited above. Kulkarni also discloses, “The project management data may include data identifying dependencies associated with the software product, a requirement quality of the software product, test coverage associated with the software product, defects associated with the software product, technical objects associated with the software product, rework effort to fix defects in the software product, additional work accepted for the software product, dependencies between software development teams associated with the software product, planned and available capacity associated with the software product, effort spent on development of the software product, emergency tickets associated with the software product, rework associated with the software product, code complexity of the software product, code coverage of the software product, log failures associated with the software product, cloud readiness of the software product, user stories and requirements associated with the software product, service call responses associated with the software product, performance issues associated with the software product, and/or the like” (para. [0016]), “As shown in FIG. 1B, and by reference number 110, the prediction system may process a first portion of the project management data, with a first plurality of machine learning models, to generate timeliness scores for the software product” (para. [0018]), “As shown in FIG. 1C, and by reference number 120, the prediction system may process a second portion of the project management data, with a second plurality of machine learning models, to generate quality scores for the software product” (para. [0026]), “As shown in FIG. 1D, and by reference number 130, the prediction system may process a third portion of the project management data, with a third plurality of machine learning models, to generate product readiness scores for the software product” (para. [0034]), “As further shown in FIG. 1D, and by reference number 135, the prediction system may combine the product readiness scores to determine an overall product readiness score for the software product” (para. [0039]), and “As shown in FIG. 1E, and by reference number 140, the prediction system may utilize a fourth machine learning model, with the overall timeliness score, the overall quality score, and the overall product readiness score, to generate a success probability for the software product” (para. [0040]). A prediction system that processes portions of project management data, which are used to determine a success probability for a software product, in Kulkarni, reads on the recited limitation. Alternatively, the “lifespan” limitation is taught by Jennings (see below).
“... simultaneously generating, by the computing device using a multi-target machine learning (ML) model ...” - See the aspects of Kulkarni that have been cited above. Kulkarni also discloses, “Some implementations described herein relate to a prediction system that utilizes a combination of machine learning models to determine a success probability for a software product,” “process a first portion of the project management data, with a first plurality of machine learning models,” “process a second portion of the project management data, with a second plurality of machine learning models,” “process a third portion of the project management data, with a third plurality of machine learning models,” and “utilize a fourth machine learning model, with the overall timeliness score, the overall quality score, and the overall product readiness score, to generate a success probability for the software product” (para. [0013]); and “two or more of the blocks of process 500 may be performed in parallel” (para. [0090]). Using machine learning models to generate a success probability for a software product based on processing portions of project management data, and doing so with the machine learning models running in parallel, in Kulkarni, reads on the recited limitation.
The combination of Kulkarni and Iliadis (hereinafter referred to as “Kulkarni/Iliadis”) teaches limitations below of claim 1 that do not appear to be disclosed in their entirety by Kulkarni:
The claimed “multi-target machine learning (ML) model” is one “that includes a multi-output deep neural network (DNN) with two parallel network branches connected to an input layer with at least as many neurons as the determined set of relevant features” - See the aspects of Kulkarni that have been cited above. Kulkarni also discloses a “neural network” (para. [0041]). Iliadis discloses aspects that do not appear to be disclosed by Kulkarni. Iliadis discloses a “deep learning methodology” with “a flexible multi-branch neural network architecture” (Abstract, p. 651), that includes a “two-branch neural network” with parallel aspects shown in Fig. 6 (p. 662), and shown also in Fig. 7 (p. 663) and in Fig. 8 (p. 664). Iliadis also discloses or suggests input layer aspects in each of the figures, including, “where no side information is provided (for example, the labels in a multi-label classification problem), we use a single fully-connected layer” (p. 663). Iliadis also discloses “The basic neural network will have as many input nodes as instance features” (p. 664). Iliadis also discloses, “Datasets with multiple classes (multi-class multi-task learning) could be tacked by replacing the single output node with a number of nodes that is equal to the number of classes” (p. 674). The deep learning methodology multi-branch neural network architecture, including multiple output nodes, two-branch neural network design, and input layer aspects, such as having as many input nodes as instance features, of Iliadis, reads on the recited limitation.
Iliadis discloses “machine learning tasks” for “Multi-target prediction” (Abstract, p. 651), similar to the claimed invention and to Kulkarni. It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the multiple machine learning model aspects, of Kulkarni, to include multi-branch neural network architecture aspects, of Iliadis, for its flexibility and competitive performance, per Iliadis (see Abstract, p. 651).
The combination of Kulkarni, Iliadis, and Lin (hereinafter referred to as “Kulkarni/Iliadis/Lin”) teaches limitations below of claim 1 that do not appear to be disclosed in their entirety by Kulkarni/Iliadis:
“... a first output comprising a classification response from a first of the parallel network branches as a first prediction of the execution outcome of the product based on the determined set of relevant features and ...” - See the aspects of Kulkarni and Iliadis that have been cited above. One of the branches of the two-branch neural network architecture, associated with one of the output nodes, of Iliadis, when applied as one or more of the machine learning models, in Kulkarni, reads on the recited “a first output comprising a ... response from a first of the parallel network branches as a first prediction of the execution outcome of the product based on the determined set of relevant features” limitation. While Iliadis also discloses “multi-label classification” (pp. 654 and 657), which reads on the recited “classification response,” Iliadis does not appear to directly link the multi-label classification to one of the two branches. Lin discloses a “data-driven deep-learning method” including a “two-branch neural network (TB-Net) that combines classification and regression in parallel” (Abstract, p. 1). The classification branch of the two parallel branches, in Lin, reads on the recited “classification response from a first of the parallel network branches” limitation.
“... a second output comprising a regression response from a second of the parallel network branches as a second prediction ... of the product based on the determined set of relevant features; and ...” - See the aspects of Kulkarni, Iliadis, and Lin that have been cited above. The other branch of the two-branch neural network architecture, associated with another of the outputs nodes, of Iliadis, when applied as one or more of the machine learning models, in Kulkarni, reads on the recited “a second output comprising a ... response from a second of the parallel network branches as a second prediction of the lifespan of the product based on the determined set of relevant features” limitation. While Kulkarni also discloses various “regression” models (see para. [0041]), and Iliadis also discloses “regression” (p. 668), they do not appear to directly link the regression to the other of the two branches. The regression branch of the two parallel branches, in Lin, reads on the recited “regression response from a second of the parallel network branches” limitation.
Lin discloses a “deep-learning method” and a “two-branch neural network” (Abstract, p. 1), similar to the claimed invention and to Kulkarni/Iliadis. It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the two-branch neural network architecture, of Kulkarni/Iliadis, to include the parallel classification and regression of the two-branch neural network, of Lin, for higher estimation accuracy, per Lin (see Abstract, p. 1).
The combination of Kulkarni, Iliadis, Lin, and Jennings (hereinafter referred to as Kulkarni/Iliadis/Lin/Jennings”) teaches limitations below of claim 1 that do not appear to be taught in their entirety by Kulkarni/Iliadis/Lin:
The claimed “second prediction” is “of the lifespan of the product” - Kulkarni does not appear to disclose details about product lifespan, although aspects disclosed by Kulkarni could be indicative of product lifespan. Nevertheless, the examiner cites Jennings. Jennings discloses, “the objective of this research is to develop a machine learning-based methodology capable of forecasting obsolescence risk and product life cycle accurately while minimizing maintenance and upkeep of the forecasting system. Specifically, this new methodology enables prediction of both the obsolescence risk level and the data when a part becomes obsolete” (p. 1428); “Life cycle forecasting estimates the time from creation to obsolescence of the part or element,” “Using the creation date and life cycle forecast, analysts can predict a date range when a part or element will become obsolete,” and “Obsolescence forecasting is important in both the design phase of the product and the manufacturing life cycle of the product. It is estimated that 60%-70% of cost during a product’s life cycle is caused by decisions made in the design phase” (p. 1429); and “Life Cycle Forecasting Using Machine Learning,” “Where ORML predicts the label active or obsolete, LCML uses regression to predict a numeric value of when the product/component will stop being manufactured,” and “LCML’s ability to estimate a date of obsolescence is a highly useful metric. LCML will give designers and supply chain professionals a more effective way of predicting the length of time to complete redesign” (p. 1433). Obsolescence forecasting of products, in Jennings, reads on the recited limitation.
Jennings discloses forecasting with machine learning (see title), similar to the claimed invention and to Kulkarni/Iliadis/Lin. It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the machine learning processes and predictions, of Kulkarni/Iliadis/Lin, to include the machine learning processes and associated predictions, of Jennings, to provide firms with proactive strategies for planning, among other reasons, per Jennings (see p. 1428).
Kulkarni/Iliadis/Lin/Jennings teaches limitations below of claim 1:
“... sending, by the computing device, the first and second predictions to the another computing device.” - See the aspects of Kulkarni that have been cited above. Kulkarni also discloses, “As shown in FIG. 1A, and by reference number 105, the prediction system may receive, from the user devices, project management data identifying a project management life cycle associated with development of a software product. The user devices may be associated with developers of the software product” (para. [0016]), using “Prediction system” to “Perform one or more actions based on the success probability” including “Generate and provide for display one or more inferences associated with the software product” (FIG. 1F), and “The prediction system may provide the one or more inferences to a user device and the user device may display the one or more inferences to a user of the user device” (para. [0042]). Sending, by the prediction system, the inferences to the user devices, in Kulkarni, reads on the recited limitation.
Regarding claim 4, Kulkarni/Iliadis/Lin/Jennings teaches the following limitations:
“The method of claim 1, wherein the multi-target ML model is generated using a training dataset generated from a corpus of historical product execution and lifespan data of an organization.” - “The method of claim 1, wherein the multi-target ML model is generated using a training dataset generated from a corpus of historical product execution and lifespan data of an organization.” - See the aspects of at least Kulkarni and Jennings that have been cited above. Kulkarni also discloses, “generating historical data for training the one or more of the first plurality of machine learning models, the second plurality of machine learning models, the third plurality of machine learning models, or the fourth machine learning model relative to other systems for identifying, obtaining, and/or generating historical data for training machine learning models” (para. [0047]). Jennings also discloses, “case data contain over 7000 unique models of cellular phones with known procurable or discontinued status, release year and quarter, and other technical specifications” and “the data set was split into two random groups. The first group represents 2/3 of the data set and is called the training data set” (p. 1433). The machine learning models being trained using historical data, in Kulkarni and Jennings, reads on the recited limitation. The rationales for combining the teachings of the cited references, from the rejection of independent claim 1, also apply to this rejection of claim 4.
Regarding claim 5, Kulkarni/Iliadis/Lin/Jennings teaches the following limitations:
“The method of claim 4, wherein the training dataset comprises a plurality of training/testing samples, wherein each training/testing sample of the plurality of training/testing samples includes one or more features extracted from the historical product execution and lifespan data, wherein the one or more features includes a feature indicative of a product type associated with the product, a business domain associated with the product, a language associated with the product, a database associated with the product, a consumption associated with the product, or a deployment associated with the product.” - See the aspects of at least Kulkarni and Jennings that have been cited above. The historical data of Kulkarni and the training data sets of Jennings read on the recited limitation. The rationales for combining the teachings of the cited references, from the rejection of independent claim 1, also apply to this rejection of claim 5.
Regarding claims 11, 14 and 15, while the claims are of different scope relative to claims 1, 4, and 5, the claims recite limitations similar to those recited by claims 1, 4, and 5. As such, the rationales applied to reject claims 1, 4, and 5 also apply for purposes of rejecting claims 11, 14, and 15. Limitations recited by claims 11, 14, and 15 that do not appear to have a counterpart in claims 1, 4, and 5, such as the recited “system comprising: one or more non-transitory machine-readable mediums configured to store instructions; and one or more processors configured to execute the instructions stored on the one or more non-transitory machine-readable mediums, wherein execution of the instructions causes the one or more processors to carry out a process” limitations, are taught by Kulkarni/Iliadis/Lin/Jennings (see, e.g., para. [0004] of Kulkarni). Claims 11, 14, and 15 are, therefore, also rejected under 35 USC 103 as obvious in view of Kulkarni/Iliadis/Lin/Jennings.
Regarding claims 16, 19, and 20, while the claims are of different scope relative to claims 1, 4, and 5 and to claims 11, 14, and 15, the claims recite limitations similar to those recited by claims 1, 4, 5, 11, 14, and 15. As such, the rationales applied to reject claims 1, 4, 5, 11, 14, and 15 also apply for purposes of rejecting claims 16, 19, and 20. Claims 16, 19, and 20 are, therefore, also rejected under 35 USC 103 as obvious in view of Kulkarni/Iliadis/Lin/Jennings.
Regarding claim 22, Kulkarni/Iliadis/Lin/Jennings teaches the following limitations:
“The method of claim 1, wherein the first comprises a binary output representing the first prediction and the second output comprises a numerical output representing the second prediction.” - See the aspects of Iliadis that have been cited above. Iliadis also discloses, “The final output layer consists of a single node that outputs the predicted score y_xt. In the classification-related MTP settings a sigmoid function is used before the output in order to restrict it to [0, 1]” (p. 664), and “Datasets with multiple classes (multi-class multi-task learning) could be tacked by replacing the single output node with a number of nodes that is equal to the number of classes” (p. 674). One of the multiple outputs being restricted to 0, 1, like the described single node, while another of the multiple outputs specifying predicted scores, in Iliadis, reads on the recited limitation. The rationales for combining the teachings of the cited references, from the rejection of claim 1, also apply to this rejection of claim 22.
Regarding claim 24, Kulkarni/Iliadis/Lin/Jennings teaches the following limitations:
“The method of claim 1, wherein the first network branch and the second network branch comprise neurons with rectified linear unit (ReLU) activation functions.” - See the aspects of Iliadis and Lin that have been cited above. The branches including input nodes with “Leaky ReLU” or “standard ReLU” as the “activation function” in Iliadis (p. 663), and/or “ReLU” in Lin (p. 3), reads on the recited limitation. The rationales for combining the teachings of the cited references, from the rejection of independent claim 1, also apply for purposes of rejecting claim 24.
Regarding claim 25, Kulkarni/Iliadis/Lin/Jennings teaches the following limitations:
“The method of claim 1, wherein the first network branch comprises a first output layer with a single neuron with a sigmoid activation function for outputting the classification response, and the second network branch comprises a second output layer with a single neuron that does not contain an activation function.” - See the aspects of Iliadis and Lin that have been cited above. The multiple branches leading to multiple outputs, including a binary output stemming from use of the sigmoid function, and another output without the sigmoid function that provides a predicted score for classification, in Iliadis, reads on the recited limitation. The rationales for combining the teachings of the cited references, from the rejection of independent claim 1, also apply for purposes of rejecting claim 25.
Regarding claim 29, Kulkarni/Iliadis/Lin/Jennings teaches the following limitations:
“The method of claim 4, wherein the set of features relevant to the execution outcome and the lifespan of the product are identified through preprocessing of the corpus of historical product execution and lifespan data of the organization.” - See the aspects of Kulkarni and Jennings that have been cited above. Kulkarni also discloses, “a machine learning model may be trained using a set of observations. The set of observations may be obtained from historical data, such as data gathered during one or more processes described herein” (para. [0051]), and “the set of observations includes a feature set. The feature set may include a set of variables, and a variable may be referred to as a feature” (para. [0052]). The feature sets associated with forecasting success probabilities and obsolescence, obtained from historical data by being gathered during processes performed, in Kulkarni, reads on the recited limitation.
Regarding claims 30 and 31, while the claims are of different scope relative to claims 22 and 25, the claims recite limitations similar to those recited by claims 22 and 25. As such, the rationales applied to reject claims 22 and 25 also apply for purposes of rejecting claims 30 and 31. Claims 30 and 31 are, therefore, also rejected under 35 USC 103 as obvious in view of Kulkarni/Iliadis/Lin/Jennings.
Regarding claim 32, while the claim is of different scope relative to claims 22 and 25 and to claims 30 and 31, the claim recites limitations similar to those recited by claims 22, 25, 30, and 31. As such, the rationales applied to reject claims 22, 25, 30 and 31 also apply for purposes of rejecting claim 32. Claim 32 is, therefore, also rejected under 35 USC 103 as obvious in view of Kulkarni/Iliadis/Lin/Jennings.
Regarding claim 33, Kulkarni/Iliadis/Lin/Jennings teaches the following limitations:
“The method of claim 1, wherein the first of the parallel network branches is trained as a binary classification model that outputs the first prediction and the second of the parallel network branches is trained as a regression model that outputs the second prediction.” - See the aspects of Iliadis that have been cited above. The two-branch neural network including a branch trained for classification and outputting a 0, 1, and another branch for regression and outputting a score, in Iliadis, reads on the recited limitation. The rationales for combining the teachings of the cited references, from the rejection of claim 1, also apply to this rejection of claim 33.
Regarding claim 34, Kulkarni/Iliadis/Lin/Jennings teaches the following limitations:
“The method of claim 1, wherein the multi-output DNN is trained to simultaneously predict a classification response and a regression response using a training dataset generated from historical product execution and lifespan data.” - See the aspects of Kulkarni, Iliadis, and Jennings that have been cited above. The two branches leading to two output nodes if the neural network, with one branch being trained to perform classification with binary outputs and the other branch for performing regression, using training data, in Iliadis, wherein the training data includes the set of observations of Kulkarni (paras. [0051] and [0053]), and the past lifespans of Jennings (p. 1431), reads on the recited limitation. The rationales for combining the teachings of the cited references, from the rejection of claim 1, also apply to this rejection of claim 34.
Claim 26 is rejected under 35 U.S.C. 103 as being unpatentable over Kulkarni, in view of Iliadis, further in view of Lin, further in view of Jennings, and further in view of Bettilyon, Tyler Elliot. “What Is an API and Why Should I Use One?” Medium online (2018) (last accessed on 09 June 2025 at https://medium.com/@TebbaVonMathenstien/what-is-an-api-and-why-should-i-use-one-863c3365726b). (hereinafter referred to as “Bettilyon”).
Regarding claim 26, the combination of Kulkarni, Iliadis, Lin, Jennings, and Bettilyon (hereinafter referred to as “Kulkarni/Iliadis/Jennings/Bettilyon”) teaches limitations below that do not appear to be taught in their entirety by Kulkarni/Iliadis/Lin/Jennings:
“The method of claim 1, wherein the information regarding the product is received from the another computing device via an application programming interface (API), and wherein the first and second predictions are sent to the another computing device via the API.” - See the aspects of Kulkarni that have been cited above. The communications between user devices and the prediction system, in Kulkarni, reads on the recited “wherein the information regarding the product is received from the another computing device” limitation. Kulkarni does not appear to provide additional technical details about how the communicating is performed. Bettilyon, however, discloses, “the term API is most often used to describe a particular kind of web interface. These Web APIs are a set of rules for interacting with a webserver (such as a Salesforce server), with the most common use case being data retrieval. API’s provide mechanisms for CRM customers to access and manipulate data stored by the API provider (Salesforce in this example). The user makes a ‘request’ to a Salesforce webserver, that webserver accesses a Salesforce database (with the customers data), and returns it to the requester in a ‘response’” (pp. 2 and 3). Using the API of Bettilyon, in the context of the communications between systems and devices, of Kulkarni, reads on the recited limitation.
Bettilyon discloses using APIs for communicating between networked systems and devices (see pp. 2 and 3), similar to the claimed invention, and to Kulkarni/Iliadis/Lin/Jennings. It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the means for network communications, of Kulkarni/Iliadis/Lin/Jennings, to include the APIs and use thereof, of Bettilyon, because “they are built around the HTTP protocol, nearly any programming language can be used to access them,” per Bettilyon (p. 7).
Claim 35 is rejected under 35 U.S.C. 103 as being unpatentable over Kulkarni, in view of Iliadis, further in view of Lin, further in view of Jennings, and further in view of U.S. Pat. App. Pub. No. 2022/0383092 A1 to Zhao (hereinafter referred to as “Zhao”).
“The method of claim 1, wherein the multi-output DNN is trained by passing training data through the multi-output DNN a specified number of times and determining whether a loss value calculated by a loss function has been reduced to a specific number, further comprising, when the loss value has not been reduced to the specific number, conducting one or more of hyperparameter tuning to change one or more of the loss function or an optimizer algorithm, adding additional hidden layers to at least one the first network branch or the second network branch, or passing the training data through the multi-output DNN an additional number of times.” - See the aspects of Kulkarni and Iliadis that have been cited above. The two-branch neural network being trained by passing training data through the neural network, per Kulkarni, and having multiple outputs, per Iliadis, reads on the recited “wherein the multi-output DNN is trained by passing training data through the multi-output DNN” limitation. Zhao discloses, “The control processor(s) 102 and/or the AI processor(s) 104 determine, at 208, whether training during the first training phase satisfies one or more first criteria. The one or more first criteria may include a criterion related to convergence, such as a criterion specifying a threshold for training loss convergence. More specifically, the criterion may specify that performance of the model 110 has converged if the training loss remains within a defined threshold (e.g., ±1%) of a training loss value or error for a defined number of samples” (para. [0033]), “As a result of determining in 208 that the one or more first criteria are not satisfied, the method 200 proceeds back to 204, where the model or the processors may be further configured based on the training parameters. For instance, the learning rate for the training, the precision, the sparsity, or hyperparameters may be adjusted before the model 110 is subjected to further training in the first training phase. On the other hand, if it is determined in 208 that one or more of the first criteria are satisfied, the method 200 proceeds to 210” (para. [0034]), “FIG. 7 illustrates a training environment in which a neural network model 702 is trained during a first training phase 700A and a second training phase 700B according to one or more embodiments. In the first training phase 700A, a partial model training process is implemented in which a subset of layers 703 of the model 702 are trained, the subset 703 being fewer in number than a total number of layers of the model 702 (i.e., a proper subset). For instance, in the first training phase 700A, an input layer 704, a hidden layer 706, and a hidden layer 708 are trained whereas hidden layers 710 through 716 and an output layer 718 are not trained during the first training phase 700A. Partial model training according to the present disclosure includes training a contiguous proper subset of layers of a neural network model. In some embodiments, the contiguous subset of layers trained in partial model training includes training an input layer of a neural network” (para. [0060]), “In the second training phase 700B, which is the final training phase for training a neural network 702 shown, all layers 720 of the neural network model 702 are trained and a trained neural network model is produced thereby. In some embodiments, a set of intermediate training phases 700N may be implemented to train the neural network model 702, the set of intermediate training phases 700N being implemented between the first training phase 700A and the second training phase 700B. In one or more embodiments, the set of intermediate training phases 700N are partial model training phases in which a subset of the layers 720 are trained. In one or more embodiments, the set of intermediate training phases 700N are full model training phases in which all layers 720 are trained. In some embodiments, one subset of the intermediate training phases 700N are partial model training phases and a remaining subset of the intermediate training phases 700N are full model training phases” (para. [0061]). The training iterations including calculation of training loss relative to thresholds for iterations, adjusting parameters, adding layers, and the like, in Zhao, reads on the recited “a specified number of times and determining whether a loss value calculated by a loss function has been reduced to a specific number, further comprising, when the loss value has not been reduced to the specific number, conducting one or more of hyperparameter tuning to change one or more of the loss function or an optimizer algorithm, adding additional hidden layers to at least one the first network branch or the second network branch, or passing the training data through the multi-output DNN an additional number of times” limitation.
Zhao discloses “training a neural network model” (Abstract), similar to the claimed invention and to Kulkarni/Iliadis/Lin/Jennings. It would have been obvious to a person having ordinary skill in the art, before the effective fling date of the claimed invention, to have modified the neural network training, of Kulkarni/Iliadis/Lin/Jennings, to include the training aspects, of Zhao, at least for the reasons set forth in para. [0022] of Zhao (e.g., improved training performance, reducing overall computational resources, reducing training time, lower validation losses, improved accuracy, and the like).
Response to Arguments
On pp. 9-18 of the Amendment, the applicant requests reconsideration and withdrawal of the claim rejection under 35 USC 101. The examiner has considered the applicant’s arguments, and in view of the amendments and arguments, the ineligibility rejection has been withdrawn.
On pp. 18-22 of the Amendment, the applicant requests reconsideration and withdrawal of the claim rejections under 35 USC 103. While the examiner indicated that the amendments appear to overcome the claim rejections, during the most recent interview, upon further consideration of the text of the cited references, and of the text of Iliadis in particular, the claim rejections have been maintained. The applicant’s arguments appear to be focused on deficiencies of one embodiment described and depicted in Iliadis. Iliadis, however, discloses another embodiment in which, “Datasets with multiple classes (multi-class multi-task learning) could be tacked by replacing the single output node with a number of nodes that is equal to the number of classes” (p. 674). For at least this reason, the rejections have been maintained.
Conclusion
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 THOMAS Y. HO, whose telephone number is (571)270-7918. The examiner can normally be reached Monday through Friday, 9:30 AM to 5:30 PM Eastern.
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, Jerry O'Connor, can be reached at 571-272-6787. 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.
/THOMAS YIH HO/Primary Examiner, Art Unit 3624
Prosecution Insights