DOCUMENT SAMPLING USING PREFETCHING AND PRECOMPUTING

DETAILED ACTION This office action is responsive to the request for reconsideration-after Non-Final filed 8/3/2026. The application contains claims 1-21, all examined and rejected. Notice of Pre-AIA or AIA Status The present application, filed on or after March 16, 2013, is being examined 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. 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 1-4, 12-14 are rejected under 35 U.S.C. 103 as being unpatentable over HUGHES et al . [US 2020/0202171 A1, hereinafter Hughes] in view of “Cold-start Active Learning through Self-supervised Language Modeling“ published on 2020 [hereinafter D1] in view of Heng Zhang [WO2021135810, hereinafter Zhang] With regard to Claim 1, Hughes teach a system to facilitate document sampling through precomputing documents, comprising: a sampling service engine, coupled to a document data store that contains a set of unlabeled documents (Fig. 3, ¶145, “computer architecture 200 includes an annotation server 202 … in communication with a database 204 that is configured to store the information stack 100 therein. … The annotation server 202 may receive the unannotated data 102 over a network 208 from an annotation client 206 for storage in the database 204”, ¶147, “At 302, the unannotated data 102 is imported to the annotation server 202 for storage on the database 204”, ¶142, “information stack 100 includes unannotated data 102 … text data may come from plain text files, such as from electronic communications through email or chat, flat files, or other types of document files (e.g., .pdf, .doc, etc.) ... The unannotated data 102 may be provided from pre-existing data stores”, ¶169), the sampling service engine including: local storage, (¶145, ¶228, “In the distributed computing environment, the program modules, applications, and other data may be stored on local and/or remote computer storage media”, ¶189, “the unannotated data 102 stored in the database 204 is pre-processed and a subset of the pre-processed data is selected for producing set of training candidates”, ¶191, “the priority queues 608 each provide a limited length priority queue for each type of sampler. Each of the priority queues 608 may store only the top 10, 100, 1000, or 10000 predictions … The priority queues 608 can be persisted to disk on the client computer 206 or the annotation server 202. In some embodiments, the priority queues are stored using distributed database technologies, such as by being stored on the database 204”), and download a subset of the documents from the document data store (Fig. 6, 602, ¶189, “In the computational architecture 600 … At 602, a sample is retrieved from the unannotated data 102 or the set of training candidates and pre-processed”, ¶190, “At 604, the pre-processed data (e.g., text) is streamed through the model 406 … At 606, a tuple of … streamed through a sampling storage writer to select one or more priority queues 608 in which to write the prediction”), a precomputing component to: upon detecting a trigger comprising a completion of the ML model training process (Fig. 6, 608, 610, 614, 616, 604, ¶190, “At 604, the pre-processed data (e.g., text) is streamed through the model 406 … to produce a prediction of the prediction set”, “At 606, a tuple of (sample identifier, vector score) (or other prediction output by the model) are streamed through a sampling storage writer to select one or more priority queues 608 in which to write the prediction. At 610, sampling selection logic selects a priority queue from which samples and predictions are presented to a user for annotation at 612 “, “Upon receiving the annotations from the user, the model is trained at 614 to produce an updated model 616 for continuing the process to make predictions from the pre-processed data at 604”, ¶213, “Once the user has annotated a number of examples, an initial model can be trained and additional sampling methods become available”) or a completion of the download of the subset of the documents, execute a sampling algorithm to select viable documents from the stored subset of the documents for user-provided labels (¶62, “requesting annotations of samples can comprise selecting a sample from the set of unannotated data based on user input or an automated sampler selection”, ¶48, “ the method can further comprise storing the predicted one or more annotations in a priority queue based on a sampling score”, ¶56, “requesting annotations of samples of the unannotated data can comprise selecting the priority queue from among a plurality of priority queues”, Fig. 6, ¶190, “At 610, sampling selection logic selects a priority queue from which samples and predictions are presented to a user for annotation at 612”) the sampling algorithm applies random sampling (Fig. 6, ¶170, “The sampling algorithms include … random sampling”, ¶182, “The sampling algorithms include, … random samples … or a linear progression through the data”, ¶66, “sampling algorithm can be selected from … random sampling algorithm”); and the viable documents are predicted to improve an accuracy score of an ML model trained using the ML model training process as compared to non-selected documents in the subset of documents (¶12, respective sampling algorithm can be selected from a density sampling algorithm; entropy sampling algorithm; estimated error reduction sampling algorithm; exhaustive sampling algorithm; flagged predictions algorithm; hard negative mining sampling algorithm; high confidence sampling algorithm; linear sampling algorithm; map visualization sampling algorithm; metadata search sampling algorithm; minimum margin sampling algorithm; query by committee sampling algorithm; random sampling algorithm; review sampling algorithm; search sampling algorithm; similarity sampling algorithm; sampling of samples for which the input was to skip the sample type algorithm; stratified sampling algorithm; most confident samples algorithm; or most uncertain samples algorithm”, ¶13, “progression can comprise successively changing between samplers of the plurality of the samplers”, system move between samplers that are that are associated with multiple algorithms without identifying a specific algorithms (randomly selection of sub-sampling type). For example system could select seed sampler algorithms that include sub-sampling type as random sampling, stratified sampling; without selecting the specific algorithm to be applied as clarified in ¶¶64-65, “automated sampler selection can be from one of a plurality of samplers in a progression”, ¶176, ¶181, “uncertainty sampler 508 uses one or more of a maximum entropy algorithm, a minimum margin algorithm, a query by committee algorithm, or other such uncertainty sampling algorithm”, ¶183, “human annotator agrees with the model's prediction, the progression 500 may automatically shift to samplers that provide samples for which there is more uncertainty”). Hughes does not explicitly teach sampling algorithm is determined based upon whether the ML model training process is completed, wherein: upon determining that the ML model training process is not completed; upon determining that the ML model training process is completed. D1 teach a precomputing component to: upon detecting a trigger comprising a completion of the ML model training process or a completion of the download of the subset of the documents (Algorithm 1 AL for Sentence Classification steps 11, “θt ← Fine-tune f(x;θ0) on D”, 6 “Mt(x) = f(x;θt−1)”, 7, “Qt ←Apply A on model Mt(x), data U”, P. 6, 5.2, “We fine-tune on a batch size of thirty-two for three epochs”, θt exist only once the run ends) execute a sampling algorithm to select viable documents from the stored subset of the documents for user-provided labels (P. 2, Col. 2, ¶3, “The goal of AL is to sample a subset D ⊂ U efficiently so that fine-tuning the classifier f on subset D improves test accuracy. On each iteration t, the learner uses strategy A to acquire k sentences from dataset U and queries for their labels (Algorithm 1)”, Algorithm 1, Steps 7-8), wherein the sampling algorithm is determined based upon whether the ML model training process is completed (P. 2, Col. 2, Algorithm 1, “ PNG media_image1.png 182 609 media_image1.png Greyscale ”, P. 2, Col. 2, ¶3, “Strategy A usually depends on an acquisition model Mt (Lowell and Lipton, 2019). If the strategy depends on model warm-starting, then the acquisition model Mt is f with parameters θt−1 from the previous iteration. Otherwise, we assume that Mt is the pre-trained model with parameters θ0”) wherein: upon determining that the ML model training process is not completed, the sampling algorithm applies random sampling (P. 2, Col. 1, ¶2, “While many AL methods randomly choose an initial sample”, P. 1, Col. 2, ¶3, “In reality, it is impractical to use because the AL strategy depends on warm-starting the model with information about the task (Ash and Adams, 2019). Thus, a fitting solution to AL for deep classifiers is a cold-start approach”, P. 3, Col. 1, ¶1, “Uncertainty sampling requires model warm starting because it depends on model predictions, whereas diversity sampling can be a cold-start approach”, this show that before a warm started trained model is available, the initial stage uses random or cold start compatible sampling sampling); upon determining that the ML model training process is completed, the sampling algorithm randomly selects and utilizes a [second] types (P. 2, Col. 2, Algorithm 1, “ PNG media_image1.png 182 609 media_image1.png Greyscale ”, P. 2, Col. 2, ¶3, “Strategy A usually depends on an acquisition model Mt (Lowell and Lipton, 2019). If the strategy depends on model warm-starting, then the acquisition model Mt is f with parameters θt−1 from the previous iteration. Otherwise, we assume that Mt is the pre-trained model with parameters θ0.”, P. 3, Col. 1, ¶1, “Uncertainty sampling requires model warm starting because it depends on model predictions, whereas diversity sampling can be a cold-start approach”, completed training process is read as completion of a prior training phase sufficient to produce an acquisition model and not a final training endpoint (that is based on the specification that clarify that ML training is complete if a trained model exist and is ready to use See specification ¶37. Under this interpretation, D1 teaches the limitation as when a trained model exists (ϴt-1), model based sampling strategies (e.g. uncertainty) are used, whereas before a trained model is available (ϴo), initial sampling is performed without reliance on model prediction (e.g. random or cold start approach)), the viable documents are predicted to improve an accuracy score of an ML model trained using the ML model training process as compared to non-selected documents in the subset of documents (P. 2, Col. 2, ¶3, “The goal of AL is to sample a subset D ⊂ U efficiently so that fine-tuning the classifier f on subset D improves test accuracy”). Hughes and D1 are analogous art to the claimed invention because they are from a similar field of endeavor of sampling data for machine learning model training within active learning. Thus, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Hughes to resulting in resolutions as disclosed by D1 with a reasonable expectation of success. One of ordinary skill in the art would be motivated to modify Hughes as described above to improve sample selection for labeling so as to reach higher accuracy within fewer sampling iterations and computation time (D1, Abstract, “minimizes labeling costs for text classification“, “our approach reaches higher accuracy within less sampling iterations and computation time.”). Hughes-D1 does not explicitly teach document sampling through prefetching, a prefetching component to: initiate a download of a subset of the documents from the document data store in response to a start of a Machine Learning (“ML”) model training process, and downloaded documents of store the subset of the documents in the local storage. Zhang teach a system to facilitate document sampling through prefetching documents, comprising: a sampling service engine, coupled to a document data store that contains a set of unlabeled documents (¶128, system is coupled to database that include unlabeled documents), a prefetching component to: initiate a download of a subset of the documents from the document data store in response to a start of a Machine Learning (“ML”) model training process (¶43, “Each main process corresponds to a pre-sampled data queue, a plurality of sample data is stored in the pre-sampled data queue”, ¶44, “the main process starts a new iterative training, and first reads a set of sample data from a pre-sampled data queue corresponding to the main process; and the set of sample data is deleted from the pre-sampled data queue after being read, so as to leave a storage position for the new sample data”, ¶129, “ the main process can determine the number of the sample data currently stored in the pre-sampled data queue … When the quantity is less than the target prefetch quantity, the new sample data is directly read from the sample database, and the new sample data is stored in the pre-sampled data queue”, ¶47, ¶129, “When the quantity is less than the target prefetch quantity, the new sample data is directly read from the sample database, and the new sample data is stored in the pre-sampled data queue”, ¶8, ¶131) download a subset of the documents from the document data store, (¶128, “main process can directly read new sample data from the sample database (i.e. document data store) when the number of sample data currently included in the pre-fetched sample data queue does not reach the target pre-fetched number”, the downloading is occurring through a database, ¶74, “when the pre-fetching quantity needs to be increased, … avoid delays in the model training process due to too small a pre-fetching quantity”) before completion of an executing Machine Learning (“ML”) model training process, and store the subset of the documents in the local storage (¶¶7-8, ¶74, “when the pre-fetching quantity needs to be increased, the pre-fetching quantity can be increased faster to ensure that the sample data stored in the pre-fetched sample data queue reaches a larger quantity faster, so as to meet the usage requirements of subsequent training iteration cycles and avoid delays in the model training process due to too small a pre-fetching quantity”). Hughes-D1and Zhang are analogous art to the claimed invention because they are from a similar field of endeavor of sampling data for machine learning model training. Thus, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Hughes-D1 to include a prefetching component resulting in resolutions as disclosed by Zhang with a reasonable expectation of success. One of ordinary skill in the art would be motivated to modify Hughes-D1 as described above to improve the convergence speed of machine learning models during training and increase training efficiency (Zhang, ¶4, ¶74, “to meet the usage requirements of the subsequent training iteration period, and avoid the model training process caused by an excessively small prefetching number being delayed; At the same time, when the prefetching number needs to be reduced, the pre-fetching amount is reduced and the pre-sampling number is reduced, so that the length variation of the pre-sampling data queue can be ensured to be more stable, and the oscillation of the training process is avoided due to an excessively fast decrease in the number of pre-fetched sample data”). With regard to Claim 2, Hughes-D1-Zhang teach the system of claim 1, wherein the plurality of sub-sampling types includes at least one of: (i) predicted positive (Hughes, ¶50, “prior to storing the predicted one or more annotations in the priority queue, it can be determined, according to the method, whether the sampling score is greater than a threshold sampling score”, ¶169, “sampling score may be equal to the confidence score or may be derived from a prediction vector to represent how well a prediction fits in the sampling algorithm”, ¶196, “sampling score may be the confidence score or a value otherwise derived by the prediction vector”, ¶197, ¶107), (ii) predicted negative (¶12, “hard negative mining sampling algorithm”, ¶196, “sampling score may be the confidence score or a value otherwise derived by the prediction vector”, ¶197, “If the sampling score is below a threshold value 716 for a given priority queue 608, then the priority queue 608 may discard 720 the prediction”, ¶107), or (iii) uncertainty (Hughes, ¶12, “most uncertain samples algorithm”, ¶22, ¶170, ¶181, “uncertainty sampler 508 uses one or more of a maximum entropy algorithm, a minimum margin algorithm, a query by committee algorithm, or other such uncertainty sampling algorithm”, ¶180, “stratified sampling algorithm identifies samples where the score for a given outcome is between two floats [A, B]”, ¶66, “sampling algorithm can be selected from a density sampling algorithm; entropy sampling algorithm; estimated error reduction sampling algorithm; exhaustive sampling algorithm; flagged predictions algorithm; hard negative mining sampling algorithm; high confidence sampling algorithm; linear sampling algorithm; map visualization sampling algorithm; metadata search sampling algorithm; minimum margin sampling algorithm; query by committee sampling algorithm; random sampling algorithm; review sampling algorithm; search sampling algorithm; similarity sampling algorithm; sampling of samples for which the input was to skip the sample type algorithm; stratified sampling algorithm; most confident samples algorithm; or most uncertain samples algorithm”, ¶170, ¶183, “human annotator agrees with the model's prediction, the progression 500 may automatically shift to samplers that provide samples for which there is more uncertainty”), and wherein the predicted positive sub-sampling type selects samples with prediction scores above a decision boundary (¶50, “prior to storing the predicted one or more annotations in the priority queue, it can be determined, according to the method, whether the sampling score is greater than a threshold sampling score”, ¶169, “sampling score may be equal to the confidence score or may be derived from a prediction vector to represent how well a prediction fits in the sampling algorithm”, ¶196, “sampling score may be the confidence score or a value otherwise derived by the prediction vector”, ¶197, ¶107), the predicted negative sub-sampling type selects samples with prediction scores below the decision boundary (¶12, “hard negative mining sampling algorithm” ¶196, “sampling score may be the confidence score or a value otherwise derived by the prediction vector”, ¶197, “If the sampling score is below a threshold value 716 for a given priority queue 608, then the priority queue 608 may discard 720 the prediction”, ¶107), and the uncertainty sub-sampling type selects samples with prediction scores within an uncertainty range (¶180, “stratified sampling algorithm identifies samples where the score for a given outcome is between two floats [A, B]”, ¶66, “sampling algorithm can be selected from a density sampling algorithm; entropy sampling algorithm; estimated error reduction sampling algorithm; exhaustive sampling algorithm; flagged predictions algorithm; hard negative mining sampling algorithm; high confidence sampling algorithm; linear sampling algorithm; map visualization sampling algorithm; metadata search sampling algorithm; minimum margin sampling algorithm; query by committee sampling algorithm; random sampling algorithm; review sampling algorithm; search sampling algorithm; similarity sampling algorithm; sampling of samples for which the input was to skip the sample type algorithm; stratified sampling algorithm; most confident samples algorithm; or most uncertain samples algorithm”, ¶170). The same motivation to combine for claim 1 equally applies for current claim With regard to Claim 3, Hughes-D1-Zhang teach the system of claim 1, wherein the document data store containing the set of unlabeled documents is accessed over a computer network (Hughes, ¶145, “annotation server 202 may receive the unannotated data 102 over a network 208 from an annotation client 206 for storage in the database 204”, ¶229). The same motivation to combine for claim 1 equally applies for current claim With regard to Claim 4, Hughes-D1-Zhang teach the system of claim 3, wherein the document data store and the sampling service engine execute in a cloud computing environment (Hughes ¶229, “the functionality disclosed above may be provided by executing the application and/or applications in a cloud computing environment. Cloud computing may comprise providing computing services”). The same motivation to combine for claim 3 equally applies for current claim With regard to Claim 12, Hughes-D1-Zhang teach the system of claim 1, wherein the sampling service engine is associated with at least one of: (i) applications (Hughes, ¶148, “distribution of document types (email, chat, Portable Document Format documents, hypertext markup language and so so), … For example, the user may elect to annotate Spanish chat messages for the week spanning Apr. 2, 2018 through Apr. 7, 2018”, (ii) bots, and (iii) Internet of Things (“IoT”) devices. The same motivation to combine for claim 1 equally applies for current claim With regard to Claim 13, Claim 13 is similar in scope to claim 1; therefore it is rejected under similar rationale. Hughes Further teach selecting, by the precomputing component, the viable documents based on the sampling algorithm (¶190, “a tuple of (sample identifier, vector score) (or other prediction output by the model) are streamed through a sampling storage writer to select one or more priority queues 608 in which to write the prediction”, ¶191, “Each of the priority queues 608 may store only the top 10, 100, 1000, or 10000 predictions for samples according to the sampling algorithm used by the sampler”) and ML model prediction scores provided by an ML model (¶¶190-191, “At 604, the pre-processed data (e.g., text) is streamed through the model 406 that converts the pre-processed data (e.g., text) into a vector of scores“) trained using the ML model training process (¶¶190-191, “the model is trained at 614 to produce an updated model 616 for continuing the process to make predictions from the pre-processed data at 604”, ¶193, “sampling storage writer to select one or more priority queues 608 in which to write the prediction. At 610, sampling selection logic selects a priority queue from which samples and predictions are presented to a user for annotation at 612”, ¶169, Fig. 7). With regard to Claim 14, Claim 14 is similar in scope to claim 4; therefore it is rejected under similar rationale. Claims 5 are rejected under 35 U.S.C. 103 as being unpatentable over HUGHES et al . [US 2020/0202171 A1, hereinafter Hughes] n view of “Cold-start Active Learning through Self-supervised Language Modeling“ published on 2020 [hereinafter D1] in view of Heng Zhang [WO2021135810, hereinafter Zhang] in view of GUIM BERNAT et al . [US 2020/0004685 A1, hereinafter GUIM]. With regard to Claim 5, Hughes-D1-Zhang teach the system of claim 1. Hughes-D1-Zhang does not explicitly teach prefetching component is associated with at least one of: (1) a maximum number of documents, and (ii) a maximum download timeout. GUIM teach prefetching component is associated with at least one of: (1) a maximum number of documents, and (ii) a maximum download timeout (¶151, “determination can be made of a resources needed to perform the prefetch request within the time limit (e.g., copy of specified content in a remote memory to a local memory tier or memory of a network interface or to a L1 or L2 cache within the time limit)”, ¶221, “prefetcher communicatively coupled to the interface and to receive a command to perform a prefetch of content from a remote memory with associated information, wherein the associated information comprises a time limit”). Hughes-D1-Zhang and GUIM are analogous art to the claimed invention because they are from a similar field of endeavor of prefetching data. Thus, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Hughes-D1-Zhang to include associate the prefetching component with a maximum download timeout resulting in resolutions as disclosed by GUIM with a reasonable expectation of success. One of ordinary skill in the art would be motivated to modify Hughes-D1-Zhang as described above to better manage available system resources by prevent overwhelming the system resources by limiting the prefetching process to a specific time limit (GUIM, ¶151). The same motivation to combine for claim 1 equally applies for current claim Claims 6-9, 15-18, 20 are rejected under 35 U.S.C. 103 as being unpatentable over HUGHES et al . [US 2020/0202171 A1, hereinafter Hughes] n view of “Cold-start Active Learning through Self-supervised Language Modeling“ published on 2020 [hereinafter D1] in view of Heng Zhang [WO2021135810, hereinafter Zhang] in view of https://fauna.com/blog/relational-database [hereinafter D2]. With regard to Claim 6, Hughes-D1-Zhang teach the system of claim 1 , wherein the precomputing component stores information about the viable documents in a precomputed samples [storage] (Hughes, ¶102 “functions performed by the system can further comprise storing the predicted one or more annotations in a priority queue based on a sampling score”, ¶190, ¶194). The same motivation to combine for claim 1 equally applies for current claim Hughes-D1-Zhang does not explicitly teach stores information in a table. D2 teach stores information in a table (P. 1-2, “relational databases have been around since the 1970s, are used in a broad range of use cases, and remain a popular choice even today because they offer a mature, stable, and proven technology You can expect to find them managing data for e-commerce applications, inventory management, payroll, customer data, and more”). Hughes-D1-Zhang and D2 are analogous art to the claimed invention because they are from a similar field of endeavor of storing data. Thus, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Hughes-D1-Zhang to use relational database resulting in resolutions as disclosed by D2 with a reasonable expectation of success. One of ordinary skill in the art would be motivated to modify Hughes-Zhang-D1 as described above to efficiently retrieve and combine data from different tables through the use of relational operations (D2, P.2, “efficiently retrieve and combine data from different tables through the use of relational operations”). With regard to Claim 7, Hughes-D1-Zhang-D2 teach the system of claim 6, wherein the sampling algorithm is associated with at least one of: (i) diversity sampling, (ii) active learning (Hughes, ¶12, “query by committee sampling algorithm “, “most uncertain samples algorithm”), (iii) cold start random sampling, and (iv) cold start text search sampling. The same motivation to combine for claim 6 equally applies for current claim With regard to Claim 8, Hughes-D1-Zhang-D2 teach the system of claim 6, further comprising: a sampling component to consume documents for labeling by a user based on information in the precomputed samples table (Hughes, Fig. 6, ¶190, “sampling selection logic selects a priority queue from which samples and predictions are presented to a user for annotation at 612. Upon receiving the annotations from the user, the model is trained at 614 to produce an updated model 616”, Fig. 21, ¶141, “user can connect appropriate data sources, set up an annotation process, annotate data”, D1, P. 1-2, “relational databases have been around since the 1970s, are used in a broad range of use cases, and remain a popular choice even today because they offer a mature, stable, and proven technology”). The same motivation to combine for claim 6 equally applies for current claim With regard to Claim 9, Hughes-D1-Zhang-D2 teach the system of claim 8, wherein the prefetching component triggers additional downloads of documents from the document data store responsive to the consumption of documents (Zhang, ¶12, “prefetch quantity is dynamically updated based on the total memory space currently occupied by the prefetch sample data queue, the upper limit threshold of memory usage, and the data throughput for training the deep learning model. When the data throughput increases, the amount of data in the prefetch sample data queue can keep up with the consumption of sample data”, ¶52, ¶67, ¶69, ¶¶76-77, as the documents in the prefetched samples are consumed the memory usage will decrease triggering prefetching for more sample document). The same motivation to combine for claim 8 equally applies for current claim With regard to Claim 15, Claim 15 is similar in scope to claim 6; therefore it is rejected under similar rationale. With regard to Claim 16, Claim 16 is similar in scope to claim 7; therefore it is rejected under similar rationale. With regard to Claim 17, Claim 17 is similar in scope to claim 8; therefore it is rejected under similar rationale. With regard to Claim 18, Claim 18 is similar in scope to claim 9; therefore it is rejected under similar rationale. With regards claim 20, HUGHES teach a system to facilitate creation of a Machine Learning (“ML”) model, comprising: a document labeling device associated with a user, including a computer processor, and a computer memory, coupled to the computer processor, storing instructions that when executed by the computer processor cause the document labeling device (Fig. 2, Fig. 6, ¶¶225-226) to: (i) receive, from a ML model creation server side, information about a subset of documents (Fig. 3, 306, Fig. 6, 602, 612), wherein upon detecting a trigger comprising a completion of the ML model training process (Fig. 608, 610, 614, 616, 604, ¶190, “At 604, the pre-processed data (e.g., text) is streamed through the model 406 … to produce a prediction of the prediction set”, “At 606, a tuple of (sample identifier, vector score) (or other prediction output by the model) are streamed through a sampling storage writer to select one or more priority queues 608 in which to write the prediction. At 610, sampling selection logic selects a priority queue from which samples and predictions are presented to a user for annotation at 612 “, “Upon receiving the annotations from the user, the model is trained at 614 to produce an updated model 616 for continuing the process to make predictions from the pre-processed data at 604”, ¶213, “Once the user has annotated a number of examples, an initial model can be trained and additional sampling methods become available”) or a completion of the download of the subset of the documents, a precomputing component on the ML model creation server side ran a sampling algorithm on the stored downloaded documents and stored (Fig. 6, Fig. 3, ¶8, ¶20 ¶¶11-12, “each of the plurality of samplers can use a different sampling algorithm”, ¶145, “annotation server 202 may receive the unannotated data 102 over a network 208 from an annotation client 206 for storage in the database 204”, ¶147, “At 302, the unannotated data 102 is imported to the annotation server 202 for storage on the database 204”), information about viable documents for user-provided labels based on ML model prediction scores provided by the ML model trained using the training process (¶48, “storing the predicted one or more annotations in a priority queue based on a sampling score”, ¶49-51) the sampling algorithm applies random sampling (Fig. 6, ¶170, “The sampling algorithms include … random sampling”, ¶182, “The sampling algorithms include, … random samples … or a linear progression through the data”, ¶66, “sampling algorithm can be selected from … random sampling algorithm”); the sampling algorithm alternates among a plurality of sub-sampling types, including at least the sampling algorithm randomly selects and utilizes a sub-sampling type from a plurality of sub-sampling types (¶180, “stratified sampling algorithm identifies samples where the score for a given outcome is between two floats [A, B]”, ¶66, “sampling algorithm can be selected from a density sampling algorithm; entropy sampling algorithm; estimated error reduction sampling algorithm; exhaustive sampling algorithm; flagged predictions algorithm; hard negative mining sampling algorithm; high confidence sampling algorithm; linear sampling algorithm; map visualization sampling algorithm; metadata search sampling algorithm; minimum margin sampling algorithm; query by committee sampling algorithm; random sampling algorithm; review sampling algorithm; search sampling algorithm; similarity sampling algorithm; sampling of samples for which the input was to skip the sample type algorithm; stratified sampling algorithm; most confident samples algorithm; or most uncertain samples algorithm”, ¶170); and wherein the viable documents are predicted to improve an accuracy score of an ML model trained using the ML model training process as compared to non-selected documents in the subset of documents (¶174, ¶176, ¶169, ¶197, ¶183, 183, “human annotator agrees with the model's prediction, the progression 500 may automatically shift to samplers that provide samples for which there is more uncertainty”, no criteria for a specific selected sampler of the samplers that could provide samples for which there is more uncertainty), display information about the subset of documents to the user (Fig. 13, Fig. 17, Fig. 20) receive, from the user, label information about the subset of documents (Fig. 20-21, ¶¶218-219, ¶172, “sample is annotated based on the user feedback. At 416, the model is updated based on the annotated samples”, ¶148, “user may elect to annotate Spanish chat messages for the week spanning Apr. 2, 2018 through Apr. 7, 2018”), and transmit the label information to the ML model creation server side (Fig. 6, 614, 616, ¶172, “sample is annotated based on the user feedback. At 416, the model is updated based on the annotated samples”). Hughes does not explicitly teach sampling algorithm is determined based upon whether the ML model training process is completed, wherein: upon determining that the ML model training process is not completed; upon determining that the ML model training process is completed. D1 teach upon detecting a trigger comprising a completion of the ML model training process or a completion of the download of the subset of the documents (Algorithm 1 AL for Sentence Classification steps 11, “θt ← Fine-tune f(x;θ0) on D”, 6 “Mt(x) = f(x;θt−1)”, 7, “Qt ←Apply A on model Mt(x), data U”, P. 6, 5.2, “We fine-tune on a batch size of thirty-two for three epochs”, θt exist only once the run ends) a precomputing component to: wherein the sampling algorithm is determined based upon whether the ML model training process is completed (P. 2, Col. 2, Algorithm 1, “ PNG media_image1.png 182 609 media_image1.png Greyscale ”, P. 2, Col. 2, ¶3, “Strategy A usually depends on an acquisition model Mt (Lowell and Lipton, 2019). If the strategy depends on model warm-starting, then the acquisition model Mt is f with parameters θt−1 from the previous iteration. Otherwise, we assume that Mt is the pre-trained model with parameters θ0”) wherein: upon determining that the ML model training process is not completed, the sampling algorithm applies random sampling (P. 2, Col. 1, ¶2, “While many AL methods randomly choose an initial sample”, P. 1, Col. 2, ¶3, “In reality, it is impractical to use because the AL strategy depends on warm-starting the model with information about the task (Ash and Adams, 2019). Thus, a fitting solution to AL for deep classifiers is a cold-start approach”, P. 3, Col. 1, ¶1, “Uncertainty sampling requires model warm starting because it depends on model predictions, whereas diversity sampling can be a cold-start approach”, this show that before a warm started trained model is available, the initial stage uses random or cold start compatible sampling sampling); upon determining that the ML model training process is completed, the sampling algorithm selects and utilizes a [second] type (P. 2, Col. 2, Algorithm 1, “ PNG media_image1.png 182 609 media_image1.png Greyscale ”, P. 2, Col. 2, ¶3, “Strategy A usually depends on an acquisition model Mt (Lowell and Lipton, 2019). If the strategy depends on model warm-starting, then the acquisition model Mt is f with parameters θt−1 from the previous iteration. Otherwise, we assume that Mt is the pre-trained model with parameters θ0.”, P. 3, Col. 1, ¶1, “Uncertainty sampling requires model warm starting because it depends on model predictions, whereas diversity sampling can be a cold-start approach”, completed training process is read as completion of a prior training phase sufficient to produce an acquisition model and not a final training endpoint (that is based on the specification that clarify that ML training is complete if a trained model exist and is ready to use See specification ¶37. Under this interpretation, D1 teaches the limitation as when a trained model exists (ϴt-1), model based sampling strategies (e.g. uncertainty) are used, whereas before a trained model is available (ϴo), initial sampling is performed without reliance on model prediction (e.g. random or cold start approach)), the viable documents are predicted to improve an accuracy score of an ML model trained using the ML model training process as compared to non-selected documents in the subset of documents (P. 2, Col. 2, ¶3, “The goal of AL is to sample a subset D ⊂ U efficiently so that fine-tuning the classifier f on subset D improves test accuracy”). Hughes and D1 are analogous art to the claimed invention because they are from a similar field of endeavor of sampling data for machine learning model training within active learning. Thus, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Hughes to resulting in resolutions as disclosed by D1 with a reasonable expectation of success. One of ordinary skill in the art would be motivated to modify Hughes as described above to improve sample selection for labeling so as to reach higher accuracy within fewer sampling iterations and computation time (D1, Abstract, “minimizes labeling costs for text classification“, “our approach reaches higher accuracy within less sampling iterations and computation time.”). Hughes-D1 does not explicitly teach the subset of documents having been downloaded, by a prefetching component, from a document data store in response to a start of a training process for a ML model and stored in local storage. Zhang teach the subset of documents having been downloaded, by a prefetching component, from a document data store in response to a start of a training process for a ML model and stored in local storage ((¶43, “Each main process corresponds to a pre-sampled data queue, a plurality of sample data is stored in the pre-sampled data queue”, ¶44, “the main process starts a new iterative training, and first reads a set of sample data from a pre-sampled data queue corresponding to the main process; and the set of sample data is deleted from the pre-sampled data queue after being read, so as to leave a storage position for the new sample data”, ¶129, “ the main process can determine the number of the sample data currently stored in the pre-sampled data queue … When the quantity is less than the target prefetch quantity, the new sample data is directly read from the sample database, and the new sample data is stored in the pre-sampled data queue”, ¶47, ¶129, “When the quantity is less than the target prefetch quantity, the new sample data is directly read from the sample database, and the new sample data is stored in the pre-sampled data queue”, ¶8, ¶131 (¶128, “main process can directly read new sample data from the sample database (i.e. document data store) when the number of sample data currently included in the pre-fetched sample data queue does not reach the target pre-fetched number”, the downloading is occurring through a database, ¶74, “when the pre-fetching quantity needs to be increased, … avoid delays in the model training process due to too small a pre-fetching quantity” ¶¶7-8). Hughes-D1 and Zhang are analogous art to the claimed invention because they are from a similar field of endeavor of sampling data for machine learning model training. Thus, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Hughes-D1 to include a prefetching component resulting in resolutions as disclosed by Zhang with a reasonable expectation of success. One of ordinary skill in the art would be motivated to modify Hughes-D1 as described above to improve the convergence speed of machine learning models during training and increase training efficiency (Zhang, ¶4, ¶74, “to meet the usage requirements of the subsequent training iteration period, and avoid the model training process caused by an excessively small prefetching number being delayed; At the same time, when the prefetching number needs to be reduced, the pre-fetching amount is reduced and the pre-sampling number is reduced, so that the length variation of the pre-sampling data queue can be ensured to be more stable, and the oscillation of the training process is avoided due to an excessively fast decrease in the number of pre-fetched sample data”). Hughes-D1-Zhang does not explicitly teach stores information in a table. D2 teach stores information in a table (P. 1-2, “relational databases have been around since the 1970s, are used in a broad range of use cases, and remain a popular choice even today because they offer a mature, stable, and proven technology You can expect to find them managing data for e-commerce applications, inventory management, payroll, customer data, and more”). Hughes-D1-Zhang and D2 are analogous art to the claimed invention because they are from a similar field of endeavor of storing data. Thus, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Hughes-D1-Zhang to use relational database resulting in resolutions as disclosed by D2 with a reasonable expectation of success. One of ordinary skill in the art would be motivated to modify Hughes-D1-Zhang as described above to efficiently retrieve and combine data from different tables through the use of relational operations (D2, P.2, “efficiently retrieve and combine data from different tables through the use of relational operations”). Claims 10 and 21 are rejected under 35 U.S.C. 103 as being unpatentable over HUGHES et al . [US 2020/0202171 A1, hereinafter Hughes] n view of “Cold-start Active Learning through Self-supervised Language Modeling“ published on 2020 [hereinafter D1] in view of Heng Zhang [WO2021135810, hereinafter Zhang] in view of https://fauna.com/blog/relational-database [hereinafter D2] in view Madhavan et al. [ US 11,178,091 B1, hereinafter Madhavan]. With regard to Claim 10, Hughes-D1-Zhang-D2 teach the system of claim 9, wherein the additional downloads are triggered (Zhang, ¶12, “prefetch quantity is dynamically updated based on the total memory space currently occupied by the prefetch sample data queue, the upper limit threshold of memory usage, and the data throughput for training the deep learning model. When the data throughput increases, the amount of data in the prefetch sample data queue can keep up with the consumption of sample data”, ¶52, ¶67, ¶69, ¶¶76-77), and unconsumed documents in the precomputed samples table (P. 1-2, “relational databases have been around since the 1970s, are used in a broad range of use cases, and remain a popular choice even today because they offer a mature, stable, and proven technology You can expect to find them managing data for e-commerce applications, inventory management, payroll, customer data, and more”). The same motivation to combine for claim 9 equally applies for current claim Hughes-D1-Zhang-D2 does explicitly teach additional downloads are triggered when a number of documents falls below a threshold value. Madhavan teach additional downloads are triggered when a number of documents falls below a threshold value (Col. 17, lines, 25-45, “a prefetch operation forces a predefined number of messages to be pushed to the consumer before being requested … assume the system is processing messages in batches of 10 and prefetch is set to 15 messages. Depending on the logic for replenishment of the prefetch queue—the consumer will always receive more message either when the prefetch queue falls below a threshold). Hughes-D1-Zhang-D2 and Madhavan are analogous art to the claimed invention because they are from a similar field of endeavor of downloading data based on available resources. Thus, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Hughes-D1-Zhang-D2 to trigger additional downloads when a number of documents falls below a threshold value as disclosed by Madhavan with a reasonable expectation of success. One of ordinary skill in the art would be motivated to modify Hughes-D1-Zhang-D2 as described above to better manage available system resources by prevent overwhelming the system resources by limiting the prefetching process to a specific number. With regards claim 21, Hughes-D1-Zhang-D2 teach the system of claim 20, [data consumption] triggers prefetch and precompute processing on the ML model creation server side (Zhang, ¶12, “prefetch quantity is dynamically updated based on the total memory space currently occupied by the prefetch sample data queue, the upper limit threshold of memory usage, and the data throughput for training the deep learning model. When the data throughput increases, the amount of data in the prefetch sample data queue can keep up with the consumption of sample data”, ¶52, ¶67, ¶69, ¶¶76-77, as the documents in the prefetched samples are consumed the memory usage will decrease triggering prefetching for more sample document, HUGHES, ¶172, “model may be updated in a batched manner, such as after a predetermined number of annotations, or the like”, ¶172, “sample is annotated based on the user feedback. At 416, the model is updated based on the annotated samples”). The same motivation to combine for claim 20 equally applies for current claim Hughes-D1-Zhang-D2 does explicitly teach wherein the transmission of label information by the document labeling device. Madhavan teach wherein the [consumption of messages] (i.e. transmission of label information by the document labeling device) triggers prefetch (Col. 17, lines, 25-45, “a prefetch operation forces a predefined number of messages to be pushed to the consumer before being requested … assume the system is processing messages in batches of 10 and prefetch is set to 15 messages. Depending on the logic for replenishment of the prefetch queue—the consumer will always receive more message either when the prefetch queue falls below a threshold or in the best case when it gets to 0”). Hughes-D1-Zhang-D2 and Madhavan are analogous art to the claimed invention because they are from a similar field of endeavor of downloading data based on available resources. Thus, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Hughes-D1-Zhang-D2 to trigger prefetching data when all the sampled data are consumed as disclosed by Madhavan with a reasonable expectation of success. One of ordinary skill in the art would be motivated to modify Hughes-D1-Zhang-D2 as described above to better manage available system resources by prevent overwhelming the system resources by triggering the prefetching process when the previous data are consumed. This is a simple combination of prior art elements according to known methods to yield predictable results, Use of known technique to improve similar devices (methods, or products) in the same way, and applying a known technique to a known device (method, or product) ready for improvement to yield predictable results (MPEP 2143). In other words, Zhang teach the ability to trigger prefetch data based on the available memory and Hughes teach the ability to process labeling trigger precompute processing for unlabeled documents upon receiving as shown in Fig. 6. However, Zhang does not disclose that the triggering condition is the consumption of all the data, examiner rely on Madhavan to modify Hughes-D1-Zhang-D2 triggering condition to be based on the consumption of all the data in the memory (labeling all the data will consume all the samples and as the model could be updated in a streaming manner See ¶172, then the last document labeling will be associated with the transmission of the last label information by the document labeling device and the consumption of the last sample) as taught by Madhavan. Claims 11 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over HUGHES et al . [US 2020/0202171 A1, hereinafter Hughes] n view of “Cold-start Active Learning through Self-supervised Language Modeling“ published on 2020 [hereinafter D1] in view of Heng Zhang [WO2021135810, hereinafter Zhang] in view of Rahman et al. [US 11,443,232 B1, hereinafter RAHMAN] . With regard to Claim 11, Hughes-D1-Zhang teach the system of claim 1, set of unlabeled documents in the document data store (Hughes, Fig. 3, ¶145, “annotation server 202 may receive the unannotated data 102 over a network 208 from an annotation client 206 for storage in the database 204”, ¶147, “At 302, the unannotated data 102 is imported to the annotation server 202 for storage on the database 204”). The same motivation to combine for claim 1 equally applies for current claim Hughes-D1-Zhang does not teach a manifest data structure is created containing a list of the set of documents. RAHMAN teach a manifest data structure is created containing a list of the set of unlabeled documents in the document data store (Col. 8, lines 28-37, “the dataset may include a manifest file which describes dataset properties and records . A record may include named attributes , including metadata such as image size , or labels such as “ dog " or " cat ” . Other attributes may include raw data which needs labeling , such as image or sentences in natural language processing ( NLP )”, col. 10, lines 8-14, “dataset service 116 can identify which are labeled and which are not labeled based on the manifest during the initial intake . The dataset service can then process only the unlabeled data”). Hughes-D1-Zhang and RAHMAN are analogous art to the claimed invention because they are from a similar field of endeavor of storing and labeling documents. Thus, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Hughes-D1-Zhang to include a manifest data structure containing a list of the set of unlabeled documents in the document data store as disclosed by RAHMAN with a reasonable expectation of success. One of ordinary skill in the art would be motivated to modify Hughes-D1-Zhang as described above to facilitate identifying the available unlabeled documents (RAHMAN, col. 10, lines 8-14, “dataset service 116 can identify which are labeled and which are not labeled based on the manifest”). With regard to Claim 19, Claim 19 is similar in scope to claim 11; therefore it is rejected under similar rationale. Response to Arguments Examiner respectfully withdraw the 35 USC 101 rejection based on the applicant argument related to the current amended claims are directed to an improvement in the field of machine learning (Remarks P. 11-12) and clarified in the specification submitted on 1/22/2021 ¶64 and reflected in the claims. Applicant argue that Hughes, Yuan, and Zhang fail to teach or render obvious "initiating a download of a subset of documents from a document data store in response to a start of a Machine Learning ("ML") model training process" as recited in amended claim 1. In its rejection of claim 1, the Office relies on FIG. 6 of Hughes and its accompanying description to disclose "download a subset of the documents from the document data store." (Office action, p. 14). FIG. 6 of Hughes depicts a process flow diagram for annotating data. (Hughes, Par. [0189]). The Office relies on "step 602," which shows the retrieval of a sample from a database of unannotated data. However, there is no discussion regarding how the download operation is initiated in response to the start of a ML training process. Instead, "step 602" is merely depicted as the start of a process in which the retrieved sample is annotated (step 612) for use in the model training process (step 614). As such, Applicant submits that the cited portions of Hughes fail to disclose the initiation of a download process "in response to a start of a ML model training process" as recited in amended claim 1. Examiner respectfully disagrees, the rejection does not rely on Hughes for the download timing. As the applicant disclose Hughes loop already download samples. The only missing element in the claim is the timing of the download. Zhang disclose the argument limitation See at least ¶43, “Each main process corresponds to a pre-sampled data queue, a plurality of sample data is stored in the pre-sampled data queue”, ¶44, “the main process starts a new iterative training, and first reads a set of sample data from a pre-sampled data queue corresponding to the main process; and the set of sample data is deleted from the pre-sampled data queue after being read, so as to leave a storage position for the new sample data”, ¶129, “ the main process can determine the number of the sample data currently stored in the pre-sampled data queue … When the quantity is less than the target prefetch quantity, the new sample data is directly read from the sample database, and the new sample data is stored in the pre-sampled data queue”, ¶47, ¶129, “When the quantity is less than the target prefetch quantity, the new sample data is directly read from the sample database, and the new sample data is stored in the pre-sampled data queue”, ¶8, ¶131. In response to applicant's arguments against the references individually, one cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986). Applicant argue that Second Applicant respectfully submits that Hughes, Yuan, and Zhang fail to teach or render obvious "upon detecting a trigger comprising a completion of the ML model training process or a completion of the download of the subset of the documents, executing a sampling algorithm to select viable documents" as recited in amended claim 1. Although the cited portions of Hughes describe an application of sampling selection logic, there is no discussion on when said sampling logic is applied relative to a trigger as recited in amended claim 1. More specifically, the application of the sampling selection logic in "step 610" of FIG. 6 of Hughes does not temporally depend on whether the ML model training process is completed or whether the download of the subset of the documents is completed. In FIG. 6 of Hughes, the application of the sampling selection logic in "step 610" is depicted and described as being performed after scored samples are stored in queues (at step 608). Examiner respectfully disagrees, Hughes disclose the argued limitation through the sampling algorithm shown in Fig. 6, 604-61610. Trained model score each sample (604), sampling storage writer (606) write each prediction to the priority queues, and each queue keep only the top predictions See at least ¶191, ¶169, ¶197. Selection logic finally select from those queues. As step 610 follow 608 that hold only scores produced at 604 by updated model 616. That model exists only because “the model is trained at 614 to produce an updated model 616 for continuing the process to make predictions from the pre-processed data at 604” ¶190. The sampling algorithm therefore cannot execute on model 616 until training (614) is complete. Applicant argue that In Hughes, samples are annotated in a loop that involves downloading samples, scoring the downloaded samples using a model (or updated model), sampling from the scored samples, annotating the sampled samples, and training the model using the annotated samples. In contrast, claim 1 recites a system in which an initial training process is performed to produce a model capable of providing useful prediction scores. Documents are downloaded, or prefetched, concurrently with the initial training process to increase efficiency. Upon either the completion of the initial training process or the completion of the download of the documents, the sampling algorithm can then be applied to sample from the downloaded documents. Examiner respectfully disagrees, claim 1 does not require “initial”, “concurrently”, and “providing useful prediction scores”. In response to applicant's argument that the references fail to show certain features of the invention, it is noted that the features upon which applicant relies (i.e., “initial”, “concurrently”, and “providing useful prediction scores”) are not recited in the rejected claim(s). Although the claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns, 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993). Assuming arguendo that the claims require the disclosed argued limitation, Hughes train an initial model before additional samplers become available See at least ¶213, and Zhang prefetch while training occurs See at least ¶4, ¶8. Applicant’s arguments with respect to independent claims 13 and 20 have been considered are not persuasive for the same reasons disclosed for claim 1. As to the remaining dependent claims, applicant argue that they are allowable due to their respective direct and indirect dependencies upon one of the aforementioned Independent claims. The examiner respectfully disagrees, Independent claims were not allowable as stated in the paragraph above in this “Response to Arguments” section in this office action. Conclusion The prior art made of record and not relied upon is considered pertinent to the applicant’s disclosure. US Patent Application Publication No. 20220253594 filed by Potts et al. that disclose labeling data for active learning See at least Abstract US Patent Application Publication No. 20160371262 filed by Cormack et al. that disclose sampling data See at least Abstract, “Technology-Assisted Review (“TAR”) process. In certain embodiments, the TAR process is a Scalable Continuous Active Learning (“S-CAL”) approach. In certain embodiments, S-CAL selects an initial sample from a document collection, trains a classifier by using a default classification for a portion of the initial sample, scores the initial sample, selects a sub-sample from the initial sample for review, removes the reviewed sub-sample from the initial sample, and repeats the process by re-training the classifier until the initial sample is exhausted”. Examiner has pointed out particular references contained in the prior arts of record in the body of this action for the convenience of the applicant. Although the specified citations are representative of the teachings in the art and are applied to the specific limitations within the individual claim, other passages and Figures may apply as well. It is respectfully requested from the applicant, in preparing the response, to consider fully the entire references as potentially teaching all or part of the claimed invention, as well as the context of the passage as taught by the prior arts or disclosed by the examiner. It is noted that any citation to specific pages, columns, figures, or lines in the prior art references any interpretation of the references should not be considered to be limiting in any way. A reference is relevant for all it contains and may be relied upon for all that it would have reasonably suggested to one having ordinary skill in the art. In re Heck, 699 F.2d 1331-33, 216 USPQ 1038-39 (Fed. Cir. 1983) (quoting In re Lemelson, 397 F.2d 1006, 1009, 158 USPQ 275, 277 (CCPA 1968)). THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a). A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action. Any inquiry concerning this communication or earlier communications from the examiner should be directed to MOHAMED ABOU EL SEOUD whose telephone number is (303)297-4285. The examiner can normally be reached Monday-Thursday 9:00am-6:00pm MT. 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, Michelle Bechtold can be reached at (571) 431-0762. 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. /MOHAMED ABOU EL SEOUD/Primary Examiner, Art Unit 2148