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 .
DETAILED ACTION
Applicant has amended claims 1, 14,16, 18, 20 in the filed amendment on 7/9/2026. Claims 1-20 are pending in this office action.
Response to Arguments
Applicant’s arguments with respect to claim(s) 14-20 have been considered but are moot in new ground of rejection.
For 101 rejection:
Applicant argued that these limitations "select a second-tier bucket from the second plurality of stratified data buckets" and "serve a batch of data samples of the variable length sequenced data from the selected second-tier bucket to one or more graphics processing units (GPUs) via transmission over one or more communication links" necessarily require interaction with a specific buffer structure and one or more GPUs in a manner that cannot be practically performed in the human mind or using pen and paper.
Applicant submits that independent claim 1, as a whole, integrates any purported judicial exception into a practical application and thus is not directed to an abstract idea. Independent claim 1 is directed to a practical application that improves machine learning model training. See, for example, paragraphs [0019] and [0030] of the present application. Several technical challenges arise when training machine learning models using variable length data (e.g., audio data, video data, words, characters, tokens, etc.). See, for example, paragraph [0015] of the present application. Further, previous approaches such as buffer padding and bucketing may result in a waste of computation resources and under- utilization of GPU memory and cores. See, for example, paragraphs [0016]-[0018] of the present application. The features of independent claim 1 addresses the technological problem related to wasted computation resources and under-utilization of GPU memory and cores when training machine learning models using variable length data. In particular, by using the tiered, two- dimensional bucketing and serving batches of data samples of the variable length sequence data to the GPU(s) via transmission over one or more communication links as claimed, improved efficiency may be achieved by reducing the waste of computation resources applied to padding data and under-utilization of GPU memory and cores that occur when using conventional approaches. See, for example, paragraphs [0019] and [0030] of the present application.
For at least the above reasons, independent claim 1, as a whole, recites a technological solution to a technological problem and integrates any purported judicial exception into a practical application. Therefore, independent claim 1 is not directed to the judicial exception and is directed to patent eligible subject matter under 35 U.S.C. § 101. For at least the above reasons, Applicant respectfully submits that claim 1 does not recite a mental process as asserted in the Office Action and thus is not directed to a judicial exception.
Examiner respectfully disagrees.
The limitations “training machine learning models using variable length data” are not recited in claims.
However, Claim 1 recite abstract idea of
obtain variable length sequenced data comprising a first sequence length dimension and a second sequence length dimension;
select a second-tier bucket from the second plurality of stratified data buckets;
serve a batch of data samples of the variable length sequenced data from the selected second-tier bucket to one or more graphics processing units (GPUs)) as drafted, is a process or system or medium that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. The human mind can perform step of obtaining, selecting and serving. Accordingly, the claims recite an abstract idea.
Claims 14, 20 recite abstract idea of
(serve a batch of data samples of the variable length sequenced data from a selected second-tier bucket of the buffer structure to one or more graphics processing units (GPUs) or
serving a batch of data samples of variable length sequenced data from a buffer structure to one or more graphics processing units (GPUs)) as drafted, is a process or system or medium that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. The human mind can perform step serving. Accordingly, the claims recite an abstract idea.
step 2A:
Claim does not recite any additional elements that integrate the judicial exception into a practical application because additional elements of one or more processors comprising processing circuitry (in claim 1); one or more processors (in claim 14) that are recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer component for obtaining that are well understood routine and conventional activities.
The additional limitations of
((instantiate a buffer structure comprising at least a first bucketing tier that includes a first plurality of stratified data buckets; store the variable length sequenced data to the buffer structure based at least on correlating the first sequence length dimension with the first bucketing tier and correlating the second sequence length dimension with the second bucketing tier) in claim 1; (store variable length sequenced data to a buffer structure based at least on correlating a first sequence length dimension with a first bucketing tier of the buffer structure that includes a first plurality of data buckets and correlating a second sequence length dimension with a second bucketing tier of the buffer structure that includes a second plurality of data buckets) in claim 14) that would be insignificant post-solution data outputting, and are insignificant extra solution activities which are well understood routine and conventional activities, see (Presenting offers and gathering statistics, OIP Techs and Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec) and See (MPEP 2106.05(g) or 2106.05(d) for Receiving or transmitting data over a network, e.g. see Intellectual Ventures v. Symantec; Storing and retrieving information in memory: Versata; Analyzing data: Genetic Techs; Determining: OIP Techs; Electronic recordkeeping: Alice Corp); and
The additional limitations of
(where individual stratified data buckets of the first plurality of stratified data buckets comprise a second bucketing tier that includes a second plurality of stratified data buckets) in claim 1 that just indicates definition of data buckets;
(the batch of data samples having a data sample batch size determined based at least on a memory utilization of the one or more GPUs and a two-dimensional bucket size of the second-tier bucket) in claim 1, that just indicates definition of the batch of data samples;
(the batch of data samples having a data sample batch size determined based at least on a memory utilization of the one or more GPUs and a two-dimensional bucket size of the selected second-tier bucket) in claim 14 that just indicates definition of the batch of data samples;
(where individual data buckets of the first plurality of data buckets comprise the second bucketing tier) in claim 14 that just indicates buckets comprising bucking tier;
(the buffer structure comprising a first bucketing tier that includes a first plurality of data buckets corresponding to a first sequence length dimension of the variable length sequenced data and a second bucketing tier that includes a second plurality of data buckets corresponding to a second sequence length dimension of the variable length sequenced data) in claim 20 that just indicates definition of buffer structure;
(individual data buckets of the first plurality of data buckets comprising the second bucketing tier) that just indicate buckets comprising bucking tier;
(wherein the batch of data samples is served from a selected second-tier bucket of the buffer structure and comprises a data sample batch size determined based at least on a memory utilization of the one or more GPUs and a two-dimensional bucket size of the selected second-tier bucket) in claim 20 just indicates the batch that is served has size based on memory utilization of GPU and bucket size.
Merely applying the mental process (i.e. serve a batch of data samples) by using a computer or computer environment as a tool (i.e. via transmission over one or more communication links) is not an improvement and does not negate the identified mental process. Rather this appears to be nothing more than mere instructions to apply the abstract idea on a computer or in a computer environment as per MPEP 2106.05(f). More specifically, the amendment essentially recites transmission over one or more communication links as a tool to perform the abstract idea. Generic parallel processing is not any improvement to the function of the computer or technology, there is no recitation of any new or improved way or parallelizing operations. Instead it is generic computer implementation, consistent with the Federal Circuit decision for instance in SAP Am., Inc. v. InvestPic, LLC, 898 F.3d 1161, 1170 (Fed. Cir. 2018).
Accordingly, these additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claim is directed to an abstract idea.
step 2B:
Claims do not recite any additional elements that amount to significantly more than the judicial exception because additional elements of one or more processors comprising processing circuitry (in claim 1); one or more processors (in claim 14) that are recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer component for obtaining that are well understood routine and conventional activities.
The additional limitations of
((instantiate a buffer structure comprising at least a first bucketing tier that includes a first plurality of stratified data buckets; store the variable length sequenced data to the buffer structure based at least on correlating the first sequence length dimension with the first bucketing tier and correlating the second sequence length dimension with the second bucketing tier) in claim 1; (store variable length sequenced data to a buffer structure based at least on correlating a first sequence length dimension with a first bucketing tier of the buffer structure that includes a first plurality of data buckets and correlating a second sequence length dimension with a second bucketing tier of the buffer structure that includes a second plurality of data buckets) in claim 14) that would be insignificant post-solution data outputting, and are insignificant extra solution activities which are well understood routine and conventional activities, see (Presenting offers and gathering statistics, OIP Techs and Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec) and See (MPEP 2106.05(g) or 2106.05(d) for Receiving or transmitting data over a network, e.g. see Intellectual Ventures v. Symantec; Storing and retrieving information in memory: Versata; Analyzing data: Genetic Techs; Determining: OIP Techs; Electronic recordkeeping: Alice Corp); and
The additional limitations of
(where individual stratified data buckets of the first plurality of stratified data buckets comprise a second bucketing tier that includes a second plurality of stratified data buckets) in claim 1 that just indicates definition of data buckets;
(the batch of data samples having a data sample batch size determined based at least on a memory utilization of the one or more GPUs and a two-dimensional bucket size of the second-tier bucket) in claim 1, that just indicates definition of the batch of data samples;
(the batch of data samples having a data sample batch size determined based at least on a memory utilization of the one or more GPUs and a two-dimensional bucket size of the selected second-tier bucket) in claim 14 that just indicates definition of the batch of data samples;
(where individual data buckets of the first plurality of data buckets comprise the second bucketing tier) in claim 14 that just indicates buckets comprising bucking tier;
(the buffer structure comprising a first bucketing tier that includes a first plurality of data buckets corresponding to a first sequence length dimension of the variable length sequenced data and a second bucketing tier that includes a second plurality of data buckets corresponding to a second sequence length dimension of the variable length sequenced data) in claim 20 that just indicates definition of buffer structure;
(individual data buckets of the first plurality of data buckets comprising the second bucketing tier) that just indicate buckets comprising bucking tier;
(wherein the batch of data samples is served from a selected second-tier bucket of the buffer structure and comprises a data sample batch size determined based at least on a memory utilization of the one or more GPUs and a two-dimensional bucket size of the selected second-tier bucket) in claim 20 just indicates the batch that is served has size based on memory utilization of GPU and bucket size.
Merely applying the mental process (i.e. serve a batch of data samples) by using a computer or computer environment as a tool (i.e. via transmission over one or more communication links) is not an improvement and does not negate the identified mental process. Rather this appears to be nothing more than mere instructions to apply the abstract idea on a computer or in a computer environment as per MPEP 2106.05(f). More specifically, the amendment essentially recites transmission over one or more communication links as a tool to perform the abstract idea. Generic parallel processing is not any improvement to the function of the computer or technology, there is no recitation of any new or improved way or parallelizing operations. Instead it is generic computer implementation, consistent with the Federal Circuit decision for instance in SAP Am., Inc. v. InvestPic, LLC, 898 F.3d 1161, 1170 (Fed. Cir. 2018).
Accordingly, these additional elements do not amount to significantly more than the judicial exception. The claims are not patent eligible.
As discussed above 101 rejection for claims is still maintained in this office action.
For 103 rejection:
Applicant respectfully submits that nothing in the asserted combination of Rengasamy, Al- Kateb, and Heck teaches or suggests "store variable length sequenced data to a buffer structure based at least on correlating a first sequence length dimension with a first bucketing tier of the buffer structure that includes a first plurality of data buckets and correlating a second sequence length dimension with a second bucketing tier of the buffer structure that includes a second plurality of data buckets, where individual data buckets of the first plurality of data buckets comprise the second bucketing tier" as recited in amended claim 14.
In response to Applicant’s argument claims are rejected under the new ground.
In addition, “store ……sequenced data to a buffer structure based at least on correlating ……with a first bucketing tier of the buffer structure and correlating ……with a second bucketing tier of the buffer structure” as bucket as store training samples of training dataset to a bucket e.g., bucket 1 in a structure based on associating, at a first level, sample sizes with a first bucketing subset, which includes a number of buckets 1-n, is represented as a first bucketing tier of a structure and associating, at a second level, sample sizes with a second bucketing subset, which includes a number of buckets 1-n, is represented as a second bucketing tier of a structure (fig. 2, paragraphs 65-68). The structure, which includes subsets having buckets 1-n, is represented as a buffer structure (fig. 2, paragraphs 71-73). The training dataset, which includes data samples 102a-n containing a sequence of different length e.g., (1±5%) n, is represented as sequenced data (paragraphs 61, 72);
In particularly: within each subset 122a-n, training samples may be bucketized into a number of buckets according to their sample sizes, e.g., the number of sequences a data sample contains, the total length of a data sample, and/or the like. In at least one embodiment, as shown in FIG. 2 for example, for each subset 122a-n containing a total number of M data samples, M data samples may be bucketized into bucket #1 of a first sample size range, bucket #2 of a second sample size range . . . , bucket #m of a mth sample size range. In at least one embodiment, a sample size may be measured by a number of sequences a data sample contains, a total length of tokens a data sample contains, and/or the like (paragraph 65);
“serve a batch of data samples of ……the sequenced data from a selected second-tier bucket of the buffer structure to one or more graphics processing units (GPUs) ……” as assign as serve a subset of data samples as a batch of data samples of the training dataset from a defined second-level bucket e.g., bucket 1 of the second subset of the structure to one GPU (fig. 2, paragraphs 64, 68-70), the training dataset, which includes data samples 102a-n containing a sequence of different length e.g., (1±5%) n, is represented as sequenced data (paragraphs 61, 72-73).
In particularly: if data samples are sampled for GPU 120a from bucket #1, bucket #4, bucket #2, . . . bucket #(m−1) of subset 122a in consecutive training steps, data samples may be sampled for other GPUs in buckets of the same order in their corresponding subsets. In at least one embodiment, for example, at a time step, GPUs 120a-n may all be processing data samples 102a-n sampled from bucket #2 of subsets 122a-n. In at least one embodiment, in this way, at any given time step, GPUs 120a-n may be processing data samples 102a-n with similar sample sizes. In at least one embodiment, for example, at each training step, GPUs 120a-n may be processing data samples 102a-n that each contain a similar or same number of sequences such that each GPU 120a-n may process a data sample of a similar structure (paragraph 68);
“the batch of data samples having a data sample batch size determined based at least on ……of the one or more GPUs and a ……bucket size of the selected second-tier bucket” as the subset of data samples as the batch of data samples has a data sample subset size defined based on a number buckets of the GPU and second sample size range as a bucket size of the defined second-level bucket e.g., bucket 1 of the second subset (fig. 2, paragraphs 64-68).
In particularly: within each subset 122a-n, training samples may be bucketized into a number of buckets according to their sample sizes, e.g., the number of sequences a data sample contains, the total length of a data sample, and/or the like. In at least one embodiment, as shown in FIG. 2 for example, for each subset 122a-n containing a total number of M data samples, M data samples may be bucketized into bucket #1 of a first sample size range, bucket #2 of a second sample size range . . . , bucket #m of a mth sample size range. In at least one embodiment, a sample size may be measured by a number of sequences a data sample contains, a total length of tokens a data sample contains, and/or the like (paragraph 65).
A sample size ranges for buckets 1-m may be determined by dividing a difference between a maximum sample size and a minimum sample size in each subset 122a-n into m intervals and each interval may correspond to a size range for a bucket (paragraph 66).
Subsets 122a-n have similar distributions of different sample sizes, each subset 122a-n may be similarly bucketized into buckets 1-m based on similar size ranges. In at least one embodiment, for each subset 122a-n, buckets 1-m may be ranked in a same order according to their respective sample size ranges (paragraph 67).
AI teaches limitations
“a memory utilization” as a memory is utilized to store sequentially arriving tuples (paragraphs 11-13);
“a first sequence length dimension” as x-axis of a first sequence length 9000-10000 as a first sequence length dimension (figs. 4-5, paragraphs 55-57);
“a second sequence length dimension” as y-axis of a first sequence length 99.97-100 as a first sequence length dimension (figs. 4-5, paragraphs 55-57);
“a two-dimensional” as graph that includes two dimensions e.g., x-axis and y-axis (figs. 4-5, paragraphs 55-56)
“variable length; the variable length” as size r that can be increased or decreased is represented as variable length: paragraphs 35, 39, 41).
AI further teaches limitations
“variable length sequenced data; the variable length sequenced data; the variable length sequenced data” as size r of a sampling reservoir that can be increased or decreased is represented as variable length: paragraphs 35, 39, 41). The size r of a sampling reservoir which stores sequentially arriving tuples (paragraphs 18-19) is represented as variable length sequenced data;
“store sequenced data to a buffer structure based at least on correlating ……with a first bucketing tier of the buffer structure and correlating ……with a second bucketing tier of the buffer structure” as store sequentially arriving tuples as sequenced data to a machine memory as a buffer structure in sampling reservoir based on establishing as correlating a first size with a first sampling reservoir as a first bucking tier of the machine memory and establishing as correlating a second size with a second sampling reservoir as a second bucketing tier of the machine memory (figs. 11-13, paragraphs 17-18, 83-84);
“……determined based at least on a memory utilization of ……and……” as a tuple added as determined based on a memory utilization of join buffer and size of the reservoir (paragraphs 94-95);
“a two-dimensional bucket size” as graph that includes two dimensions e.g., x-axis and y-axis (figs. 4-5, paragraphs 55-56).
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Per Step 1, claim 1 is directed to one or more processors, claim 14 is directed to a system, claim 20 is directed to method, which are statutory categories of invention per Step 1. However, the claims are rejected under 35 U.S.C. 101 because they are directed to an abstract idea, a judicial exception, without reciting additional elements that integrate the judicial exception into a practical application or are significantly more.
Step 2:
a) In analyzing under step 2A Prong One, Does the claim recite an abstract idea law of nature or natural phenomenon? Yes.
Claim 1 recite abstract idea of
obtain variable length sequenced data comprising a first sequence length dimension and a second sequence length dimension;
select a second-tier bucket from the second plurality of stratified data buckets;
serve a batch of data samples of the variable length sequenced data from the selected second-tier bucket to one or more graphics processing units (GPUs)) as drafted, is a process or system or medium that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. The human mind can perform step of obtaining, selecting and serving. Accordingly, the claims recite an abstract idea.
Claims 14, 20 recite abstract idea of
(serve a batch of data samples of the variable length sequenced data from a selected second-tier bucket of the buffer structure to one or more graphics processing units (GPUs) or
serving a batch of data samples of variable length sequenced data from a buffer structure to one or more graphics processing units (GPUs)) as drafted, is a process or system or medium that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. The human mind can perform step serving. Accordingly, the claims recite an abstract idea.
b) In analyzing under step 2A Prong Two, Does the claim recite additional elements that integrate the judicial exception into a practical application? NO.
Claim does not recite any additional elements that integrate the judicial exception into a practical application because additional elements of (one or more processors comprising processing circuitry (in claim 1); one or more processors (in claim 14); “via transmission over one or more communication links” in claims 1, 14, 20) that are recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer component for obtaining that are well understood routine and conventional activities.
The additional limitations of
((instantiate a buffer structure comprising at least a first bucketing tier that includes a first plurality of stratified data buckets; store the variable length sequenced data to the buffer structure based at least on correlating the first sequence length dimension with the first bucketing tier and correlating the second sequence length dimension with the second bucketing tier) in claim 1; (store variable length sequenced data to a buffer structure based at least on correlating a first sequence length dimension with a first bucketing tier of the buffer structure that includes a first plurality of data buckets and correlating a second sequence length dimension with a second bucketing tier of the buffer structure that includes a second plurality of data buckets) in claim 14 that would be insignificant post-solution data outputting, and are insignificant extra solution activities which are well understood routine and conventional activities, see (Presenting offers and gathering statistics, OIP Techs and Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec) and See (MPEP 2106.05(g) or 2106.05(d) for Receiving or transmitting data over a network, e.g. see Intellectual Ventures v. Symantec; Storing and retrieving information in memory: Versata; Analyzing data: Genetic Techs; Determining: OIP Techs; Electronic recordkeeping: Alice Corp); and
The additional limitations of
(where individual stratified data buckets of the first plurality of stratified data buckets comprise a second bucketing tier that includes a second plurality of stratified data buckets) in claim 1 that just indicates definition of data buckets;
(the batch of data samples having a data sample batch size determined based at least on a memory utilization of the one or more GPUs and a two-dimensional bucket size of the second-tier bucket) in claim 1, that just indicates definition of the batch of data samples;
(where individual data buckets of the first plurality of data buckets comprise the second bucketing tier) in claim 14 that just indicates buckets comprising bucking tier;
(the batch of data samples having a data sample batch size determined based at least on a memory utilization of the one or more GPUs and a two-dimensional bucket size of the selected second-tier bucket) in claim 14 that just indicates definition of the batch of data samples; and
(the buffer structure comprising a first bucketing tier that includes a first plurality of data buckets corresponding to a first sequence length dimension of the variable length sequenced data and a second bucketing tier that includes a second plurality of data buckets corresponding to a second sequence length dimension of the variable length sequenced data) in claim 20 that just indicates definition of buffer structure;
(individual data buckets of the first plurality of data buckets comprising the second bucketing tier) in claim 20 that just indicate buckets comprising bucking tier;
(wherein the batch of data samples is served from a selected second-tier bucket of the buffer structure and comprises a data sample batch size determined based at least on a memory utilization of the one or more GPUs and a two-dimensional bucket size of the selected second-tier bucket) in claim 20 just indicates the batch that is served has size based on memory utilization of GPU and bucket size.
Merely applying the mental process (i.e. serve a batch of data samples) by using a computer or computer environment as a tool (i.e. via transmission over one or more communication links) is not an improvement and does not negate the identified mental process. Rather this appears to be nothing more than mere instructions to apply the abstract idea on a computer or in a computer environment as per MPEP 2106.05(f). More specifically, the amendment essentially recites transmission over one or more communication links as a tool to perform the abstract idea. Generic parallel processing is not any improvement to the function of the computer or technology, there is no recitation of any new or improved way or parallelizing operations. Instead it is generic computer implementation, consistent with the Federal Circuit decision for instance in SAP Am., Inc. v. InvestPic, LLC, 898 F.3d 1161, 1170 (Fed. Cir. 2018).
Accordingly, these additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claim is directed to an abstract idea.
c) In analyzing under step 2B, does the claim recite additional elements that amount to significantly more than the judicial exception? NO
Claims do not recite any additional elements that amount to significantly more than the judicial exception because additional elements (one or more processors comprising processing circuitry (in claim 1); one or more processors (in claim 14); “via transmission over one or more communication links” in claims 1, 14, 20) that are recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer component for obtaining that are well understood routine and conventional activities.
The additional limitations of
((instantiate a buffer structure comprising at least a first bucketing tier that includes a first plurality of stratified data buckets; store the variable length sequenced data to the buffer structure based at least on correlating the first sequence length dimension with the first bucketing tier and correlating the second sequence length dimension with the second bucketing tier) in claim 1; (store variable length sequenced data to a buffer structure based at least on correlating a first sequence length dimension with a first bucketing tier of the buffer structure that includes a first plurality of data buckets and correlating a second sequence length dimension with a second bucketing tier of the buffer structure that includes a second plurality of data buckets) in claim 14 that would be insignificant post-solution data outputting, and are insignificant extra solution activities which are well understood routine and conventional activities, see (Presenting offers and gathering statistics, OIP Techs and Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec) and See (MPEP 2106.05(g) or 2106.05(d) for Receiving or transmitting data over a network, e.g. see Intellectual Ventures v. Symantec; Storing and retrieving information in memory: Versata; Analyzing data: Genetic Techs; Determining: OIP Techs; Electronic recordkeeping: Alice Corp); and
The additional limitations of
(where individual stratified data buckets of the first plurality of stratified data buckets comprise a second bucketing tier that includes a second plurality of stratified data buckets) in claim 1 that just indicates definition of data buckets;
(the batch of data samples having a data sample batch size determined based at least on a memory utilization of the one or more GPUs and a two-dimensional bucket size of the second-tier bucket) in claim 1, that just indicates definition of the batch of data samples;
(where individual data buckets of the first plurality of data buckets comprise the second bucketing tier) in claim 14 that just indicates buckets comprising bucking tier;
(the batch of data samples having a data sample batch size determined based at least on a memory utilization of the one or more GPUs and a two-dimensional bucket size of the selected second-tier bucket) in claim 14 that just indicates definition of the batch of data samples; and
(the buffer structure comprising a first bucketing tier that includes a first plurality of data buckets corresponding to a first sequence length dimension of the variable length sequenced data and a second bucketing tier that includes a second plurality of data buckets corresponding to a second sequence length dimension of the variable length sequenced data) in claim 20 that just indicates definition of buffer structure;
(individual data buckets of the first plurality of data buckets comprising the second bucketing tier) in claim 20 that just indicate buckets comprising bucking tier;
(wherein the batch of data samples is served from a selected second-tier bucket of the buffer structure and comprises a data sample batch size determined based at least on a memory utilization of the one or more GPUs and a two-dimensional bucket size of the selected second-tier bucket) in claim 20 just indicates the batch that is served has size based on memory utilization of GPU and bucket size.
Merely applying the mental process (i.e. serve a batch of data samples) by using a computer or computer environment as a tool (i.e. via transmission over one or more communication links) is not an improvement and does not negate the identified mental process. Rather this appears to be nothing more than mere instructions to apply the abstract idea on a computer or in a computer environment as per MPEP 2106.05(f). More specifically, the amendment essentially recites transmission over one or more communication links as a tool to perform the abstract idea. Generic parallel processing is not any improvement to the function of the computer or technology, there is no recitation of any new or improved way or parallelizing operations. Instead it is generic computer implementation, consistent with the Federal Circuit decision for instance in SAP Am., Inc. v. InvestPic, LLC, 898 F.3d 1161, 1170 (Fed. Cir. 2018).
Accordingly, these additional elements do not amount to significantly more than the judicial exception. The claims are not patent eligible.
Dependent claims 2-13, 15-19 include all the limitations of claim 1. Therefore, claims 2-13 recite the same abstract idea of processing and generating practically being performed in the mind, and the analysis must therefore proceed to Step 2A Prong Two.
In particularly:
Claims 2, 15 recite limitation of (determine the data sample batch size based at least on iterative execution of a training step of a machine learning model on the one or more GPUs to converge on a valid batch size that does not generate a memory error) as drafted, is a process or system or medium that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. The human mind can perform step of determining. Accordingly, the claims recite an abstract idea.
Claims 3, 16 recite limitation of (execute the training step of the machine learning model on the one or more GPUs using pseudo data; or execute a first training step of a machine learning model on the one or more GPUs based on the batch of data samples of the variable length sequenced data) as drafted, is a process or system or medium that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. The human mind can perform step of executing. Accordingly, the claims recite an abstract idea.
Claim 4 recites limitation of ( store a data sample of the variable length sequenced data to an arbitrary buffer position within the second-tier bucket when the variable length sequenced data is stored to the buffer structure) that would be insignificant post-solution data outputting, and are insignificant extra solution activities which are well understood routine and conventional activities, see (Presenting offers and gathering statistics, OIP Techs and Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec) and See (MPEP 2106.05(g) or 2106.05(d) for Receiving or transmitting data over a network, e.g. see Intellectual Ventures v. Symantec; Storing and retrieving information in memory: Versata; Analyzing data: Genetic Techs; Determining: OIP Techs; Electronic recordkeeping: Alice Corp).
Claim 5 recites limitation of (serve the batch of data samples from the selected second-tier bucket based on an interleaved selection scheme) that would be insignificant post-solution data outputting, and are insignificant extra solution activities which are well understood routine and conventional activities, see (Presenting offers and gathering statistics, OIP Techs and Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec) and See (MPEP 2106.05(g) or 2106.05(d) for Receiving or transmitting data over a network, e.g. see Intellectual Ventures v. Symantec; Storing and retrieving information in memory: Versata; Analyzing data: Genetic Techs; Determining: OIP Techs; Electronic recordkeeping: Alice Corp).
Claim 6 recites limitation of (execute a first training step of a machine learning model on the one or more GPUs based on the batch of data samples of the variable length sequenced data) as drafted, is a process or system or medium that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. The human mind can perform step of executing. Accordingly, the claims recite an abstract idea.
Claim 7 recites limitation of (select a second second-tier bucket from the second plurality of stratified data buckets; and serve a second batch of data samples of the variable length sequenced data from the selected second-tier bucket to the one or more GPUs) as drafted, is a process or system or medium that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. The human mind can perform step of selecting and serving. Accordingly, the claim recites an abstract idea.
The addition limitation of (the second batch of data samples having a second data sample batch size determined based at least on the memory utilization of the one or more GPUs and a two-dimensional bucket size of the second second-tier bucket) just indicates the batch that is served has size based on memory utilization of GPU and bucket size.
The additional limitation “via transmission over one or more communication links” is a tool to perform abstract idea of selecting and serving.
Claims 8, 17 recite abstract idea of (determine bucket bin boundaries for at least one of the first bucketing tier and the second bucketing tier based on computing one or more sequence length histograms) as drafted, is a process or system or medium that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. The human mind can perform step of determining. Accordingly, the claims recite an abstract idea.
Claim 9 recites limitation of (wherein the buffer structure is instantiated by a pre-processor executed on a processing unit distinct from the one or more GPUs) that just indicated buffer structured instantiated by a processor.
Claim 10 recites limitation of (the buffer structure is instantiated by a pre-processor accessible as a network service hosted on a cloud-computing platform) that just indicated buffer structured instantiated by a pre- processor.
Claims 11, 18 recite limitation of (dynamically store the variable length sequenced data to the buffer structure while the batch of data samples is concurrently served to the one or more GPUs; or dynamically store the sequenced data to the buffer structure while the batch of data samples is concurrently served to the one or more GPUs) that would be insignificant post-solution data outputting, and are insignificant extra solution activities which are well understood routine and conventional activities, see (Presenting offers and gathering statistics, OIP Techs and Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec) and See (MPEP 2106.05(g) or 2106.05(d) for Receiving or transmitting data over a network, e.g. see Intellectual Ventures v. Symantec; Storing and retrieving information in memory: Versata; Analyzing data: Genetic Techs; Determining: OIP Techs; Electronic recordkeeping: Alice Corp).
Claim 12 recites abstract idea of (measure a two-dimensional length of a data sample of the variable length sequenced data to obtain a first sequence length dimension vector and a second sequence length dimension vector; and apply a search algorithm to match the first sequence length dimension vector to a first corresponding bucket of the first bucketing tier and to match the second sequence length dimension vector to a second corresponding sub-bucket of the second bucketing tier) as drafted, is a process or system or medium that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. The human mind can perform step of measuring, applying. Accordingly, the claim recites an abstract idea.
Claim 13 recites limitation of (wherein the processing circuitry is comprised in at least one of: a control system for an autonomous or semi-autonomous machine; a perception system for an autonomous or semi-autonomous machine; a system for performing simulation operations; a system for performing digital twin operations; a system for performing light transport simulation; a system for performing collaborative content creation for three-dimensional assets; a system for performing deep learning operations; a system for performing remote operations; a system for performing real-time streaming; a system for generating or presenting one or more of augmented reality content, virtual reality content, or mixed reality content; a system implemented using an edge device; a system implemented using a robot; a system for performing conversational AI operations; a system implementing one or more language models; a system implementing one or more large language models (LLMs); a system implementing one or more vision language models (VLMs); a system implementing one or more multimodal language models (MMLMs); a system implemented using one or more cloud-hosted microservices; a system for generating synthetic data; a system for generating synthetic data using AI; a system incorporating one or more virtual machines (VMs); a system using or deploying one or more inference microservices; a system that incorporates one or more machine learning models deployed in a service or microservice along with an OS-level virtualization package; a system implemented at least partially in a data center; or a system implemented at least partially using cloud computing resources) that just provides definition of circuitry.
Claim 19 recites limitation of (the system is comprised in at least one of: a control system for an autonomous or semi-autonomous machine; a perception system for an autonomous or semi-autonomous machine; a system for performing simulation operations; a system for performing digital twin operations; a system for performing light transport simulation; a system for performing collaborative content creation for three-dimensional assets; a system for performing deep learning operations; a system for performing remote operations; a system for performing real-time streaming; a system for generating or presenting one or more of augmented reality content, virtual reality content, or mixed reality content; a system implemented using an edge device; a system implemented using a robot; a system for performing conversational AI operations; a system implementing one or more language models; a system implementing one or more large language models (LLMs);a system implementing one or more vision language models (VLMs);a system implementing one or more multimodal language models (MMLMs);a system implemented using one or more cloud-hosted microservices; a system for generating synthetic data; a system for generating synthetic data using AI; a system incorporating one or more virtual machines (VMs);a system using or deploying one or more inference microservices; a system that incorporates one or more machine learning models deployed in a service or microservice along with an OS-level virtualization package; a system implemented at least partially in a data center; or a system implemented at least partially using cloud computing resources) that just provides definition of a system.
Accordingly, these additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claims are directed to an abstract idea.
Accordingly, these additional elements do not amount to significantly more than the judicial exception. The claims are not patent eligible.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 14, 16, 18-20 are rejected under 35 U.S.C. 103 as being unpatentable over Rengasamy et al (or hereinafter “Re”) (US 20250045589) in view of Al-Kateb et al (or hereinafter “AI”) (US 20110313977) and Freeberg et al (or hereinafter “Free”) (US 20050203917).
As to claim 14, Re teaches limitations
“one or more processors to” as one or more processors to (paragraphs 86, 405):
“store ……sequenced data to a buffer structure based at least on correlating ……with a first bucketing tier of the buffer structure…… and correlating ……with a second bucketing tier of the buffer structure……” as bucket as store training samples of training dataset to a bucket e.g., bucket 1 in a structure based on associating, at a first level, sample sizes with a first bucketing subset, which includes a number of buckets 1-n, is represented as a first bucketing tier of a structure and associating, at a second level, sample sizes with a second bucketing subset, which includes a number of buckets 1-n, is represented as a second bucketing tier of a structure (fig. 2, paragraphs 65-68). The structure, which includes subsets having buckets 1-n, is represented as a buffer structure (fig. 2, paragraphs 71-73). The training dataset, which includes data samples 102a-n containing a sequence of different length e.g., (1±5%) n, is represented as sequenced data (paragraphs 61, 72);
In particularly: within each subset 122a-n, training samples may be bucketized into a number of buckets according to their sample sizes, e.g., the number of sequences a data sample contains, the total length of a data sample, and/or the like. In at least one embodiment, as shown in FIG. 2 for example, for each subset 122a-n containing a total number of M data samples, M data samples may be bucketized into bucket #1 of a first sample size range, bucket #2 of a second sample size range . . . , bucket #m of a mth sample size range. In at least one embodiment, a sample size may be measured by a number of sequences a data sample contains, a total length of tokens a data sample contains, and/or the like (paragraph 65);
“serve a batch of data samples of ……the sequenced data from a selected second-tier bucket of the buffer structure to one or more graphics processing units (GPUs) ……” as assign as serve a subset of data samples as a batch of data samples of the training dataset from a defined second-level bucket e.g., bucket 1 of the second subset of the structure to one GPU (fig. 2, paragraphs 64, 68-70), the training dataset, which includes data samples 102a-n containing a sequence of different length e.g., (1±5%) n, is represented as sequenced data (paragraphs 61, 72-73).
In particularly: if data samples are sampled for GPU 120a from bucket #1, bucket #4, bucket #2, . . . bucket #(m−1) of subset 122a in consecutive training steps, data samples may be sampled for other GPUs in buckets of the same order in their corresponding subsets. In at least one embodiment, for example, at a time step, GPUs 120a-n may all be processing data samples 102a-n sampled from bucket #2 of subsets 122a-n. In at least one embodiment, in this way, at any given time step, GPUs 120a-n may be processing data samples 102a-n with similar sample sizes. In at least one embodiment, for example, at each training step, GPUs 120a-n may be processing data samples 102a-n that each contain a similar or same number of sequences such that each GPU 120a-n may process a data sample of a similar structure (paragraph 68);
“the batch of data samples having a data sample batch size determined based at least on ……of the one or more GPUs and a ……bucket size of the selected second-tier bucket” as the subset of data samples as the batch of data samples has a data sample subset size defined based on a number buckets of the GPU and second sample size range as a bucket size of the defined second-level bucket e.g., bucket 1 of the second subset (fig. 2, paragraphs 64-68).
In particularly: within each subset 122a-n, training samples may be bucketized into a number of buckets according to their sample sizes, e.g., the number of sequences a data sample contains, the total length of a data sample, and/or the like. In at least one embodiment, as shown in FIG. 2 for example, for each subset 122a-n containing a total number of M data samples, M data samples may be bucketized into bucket #1 of a first sample size range, bucket #2 of a second sample size range . . . , bucket #m of a mth sample size range. In at least one embodiment, a sample size may be measured by a number of sequences a data sample contains, a total length of tokens a data sample contains, and/or the like (paragraph 65).
A sample size ranges for buckets 1-m may be determined by dividing a difference between a maximum sample size and a minimum sample size in each subset 122a-n into m intervals and each interval may correspond to a size range for a bucket (paragraph 66).
Subsets 122a-n have similar distributions of different sample sizes, each subset 122a-n may be similarly bucketized into buckets 1-m based on similar size ranges. In at least one embodiment, for each subset 122a-n, buckets 1-m may be ranked in a same order according to their respective sample size ranges (paragraph 67).
Re does not explicitly teach limitations
a memory utilization;
a first sequence length dimension; a second sequence length dimension; a two-dimensional;
variable length; the variable length
that includes a first plurality of data buckets; that includes a second plurality of data buckets, where individual data buckets of the first plurality of data buckets comprise the second bucketing tier;
via transmission over one or more communication links.
AI teaches limitations
“a memory utilization” as a memory is utilized to store sequentially arriving tuples (paragraphs 11-13);
“a first sequence length dimension” as x-axis of a first sequence length 9000-10000 as a first sequence length dimension (figs. 4-5, paragraphs 55-57);
“a second sequence length dimension” as y-axis of a first sequence length 99.97-100 as a first sequence length dimension (figs. 4-5, paragraphs 55-57);
“a two-dimensional” as graph that includes two dimensions e.g., x-axis and y-axis (figs. 4-5, paragraphs 55-56)
“variable length; the variable length” as size r that can be increased or decreased is represented as variable length: paragraphs 35, 39, 41).
AI further teaches limitations
“variable length sequenced data; the variable length sequenced data” as size r of a sampling reservoir that can be increased or decreased is represented as variable length: paragraphs 35, 39, 41). The size r of a sampling reservoir which stores sequentially arriving tuples (paragraphs 18-19) is represented as variable length sequenced data;
“store variable length sequenced data to a buffer structure based at least on correlating ……with a first bucketing tier of the buffer structure and correlating ……with a second bucketing tier of the buffer structure” as store sequentially arriving tuples of size r that can be increased or decreased (paragraphs 35, 39) as sequenced data to a machine memory as a buffer structure in sampling reservoir based on establishing as correlating a first size with a first sampling reservoir as a first bucking tier of the machine memory and establishing as correlating a second size with a second sampling reservoir as a second bucketing tier of the machine memory (figs. 11-13, paragraphs 17-18, 83-84);
“……determined based at least on a memory utilization of ……and……” as a tuple added as determined based on a memory utilization of join buffer and size of the reservoir (paragraphs 94-95);
“a two-dimensional bucket size” as graph that includes two dimensions e.g., x-axis and y-axis (figs. 4-5, paragraphs 55-56).
AI and Re disclose a method of receiving and storing data into buckets. These references are same field with application’s field. Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply AI’s teaching to Re’s system to allow one or more sample reservoirs to change size during sampling without loss data, and further to reduce network communications and further to increase lifetime of the network.
Free teaches limitations
“that includes a first plurality of data buckets” as a first level as a first bucketing tier of a hierarchy as the buffer structure includes a plurality of data folders 712 as data buckets (figs. 7, 11, paragraphs 150, 161);
“that includes a second plurality of data buckets, where individual data buckets of the first plurality of data buckets comprise the second bucketing tier” as a second level as a second bucketing tier of the hierarchy include a plurality of second data folders 714 as data buckets (figs. 7, 11, paragraphs 150, 161), a folder GRIB 712 and weather charts 712 as individual data buckets of the plurality of data folders 712 as data buckets includes the second level as the second bucketing tier e.g., subfolders 716 (figs. 7, 11, paragraphs 150-151, 161), a folder GRIB 712 includes subfolders 716 and weather charts 712 includes subfolders 716 when these folders are expanded (figs. 7, 11, paragraphs 150-151, 161);
“via transmission over one or more communication links” as via transfer of desired information over communication links (paragraphs 36-37, 82);
“a first sequence length dimension” as longitude of a first sequence length from -95 to -80 as a first sequence length dimension (figs. 15-16, paragraphs 53, 168-169);
“a second sequence length dimension” as latitude of second sequence length from 18-20 as a second sequence length dimension (figs. 15-16, paragraphs 53, 168-169);
“a two-dimensional” as an image of a file of latitude and longitude as a two-dimensional (figs. 15-16, paragraphs 53, 168-169);
variable length; the variable length (variable size as variable length: fig. 11, paragraphs 79, 154-155).
Free further teaches limitations
“a first bucketing tier of the buffer structure that includes a first plurality of data buckets” as a first level as a first bucketing tier of a hierarchy as the buffer structure includes a plurality of data folders 712 as data buckets (figs. 7, 11, paragraphs 150, 161);
“a second bucketing tier of the buffer structure that includes a second plurality of data buckets” as a second level as a second bucketing tier of the hierarchy include a plurality of second data folders 714 as data buckets (figs. 7, 11, paragraphs 150, 161);
“a two-dimensional bucket size of the selected second-tier bucket” as size of a file of latitude and longitude as a two-dimensional as a two-dimensional bucket size (figs. 15-16, paragraphs 53, 168-169) of a hierarchy include a second level as the second bucketing tier e.g., subfolders 716 (figs. 7, 11, paragraphs 150-151, 161).
Free and Re disclose a method of receiving and storing data into folders. These references are same field as application’s field. Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Free’s teaching to Re’s system in order to minimize connection times and costs of content when data is requested, to minimize time required to retrieve requested data, and further to provide faster effective transfer rates.
As to claim 16, Re, AI, and Free teach limitations
“wherein the one or more processors are further to execute a first training step of a machine learning model on the one or more GPUs based on the batch of data samples of the variable length sequenced data” as the processor is further to execute (Re: paragraph 80) a first training process of a machine learning model on the GPU based on (Re: fig. 39A, paragraphs 53, 126, 230) the subset of samples of the training data (Re: paragraphs 64-69) which includes data samples 102a-n containing a sequence as sequenced data (Re: paragraphs 61, 72) of variable length (AI: fig. 11, paragraphs 79, 154-155) or variable size as variable length (Free: fig. 11, paragraphs 79, 154-155).
As to claim 18, Re, AI, and Free teach limitations
“wherein the one or more processors are further to dynamically store the variable length sequenced data to the buffer structure while the batch of data samples is concurrently served to the one or more GPUs” as the processor is further to (Re: paragraph 80; AI: paragraph 17) store sequenced data of variable length (AI: fig. 11, paragraphs 79, 154-155) or variable size as variable length (Free: fig. 11, paragraphs 79, 154-155) to a machine memory as a buffer structure while (AI: paragraphs 11, 17) the batch of data samples is concurrently assigned to the one or more GPUs (Re: paragraphs 59-62, 72-73).
As to claim 19, Re, AI, and Free teach limitations
“wherein the system is comprised in at least one of:
a control system for an autonomous or semi-autonomous machine;
a perception system for an autonomous or semi-autonomous machine;
a system for performing simulation operations;
a system for performing digital twin operations;
a system for performing light transport simulation;
a system for performing collaborative content creation for three-dimensional assets;
a system for performing deep learning operations;
a system for performing remote operations;
a system for performing real-time streaming;
a system for generating or presenting one or more of augmented reality content, virtual reality content, or mixed reality content;
a system implemented using an edge device;
a system implemented using a robot;
a system for performing conversational AI operations;
a system implementing one or more language models;
a system implementing one or more large language models (LLMs);
a system implementing one or more vision language models (VLMs);
a system implementing one or more multimodal language models (MMLMs);
a system implemented using one or more cloud-hosted microservices;
a system for generating synthetic data;
a system for generating synthetic data using AI;
a system incorporating one or more virtual machines (VMs);
a system using or deploying one or more inference microservices;
a system that incorporates one or more machine learning models deployed in a service or microservice along with an OS-level virtualization package;
a system implemented at least partially in a data center; or
a system implemented at least partially using cloud computing resources” as the system is included in form of a system implemented in a data center 800 (Re: fig.8, paragraphs 118-122) or the computing device as the system is included in form of a computer system 1800 including (AI: paragraph 123; Free: figs. 1-3) a virtual machine (Re: paragraph 80).
Claim 20 has the same limitation subject matter as discussed in claim 14; thus, claim 20 is rejected under the same reason as discussed in claim 14. In addition, Re teaches limitations
“serving a batch of data samples of variable length sequenced data from a buffer structure to one or more graphics processing units (GPUs)” as assigning as serving a subset of data samples of the training dataset from a defined second-level bucket e.g., bucket 1 of the structure to one GPU (fig. 2, paragraphs 64, 68-70). The training dataset, which includes data samples 102a-n containing a sequence of different length e.g., (1±5%) n as variable length, is represented as variable length sequenced data (paragraphs 61, 72-73).
In particularly: if data samples are sampled for GPU 120a from bucket #1, bucket #4, bucket #2, . . . bucket #(m−1) of subset 122a in consecutive training steps, data samples may be sampled for other GPUs in buckets of the same order in their corresponding subsets. In at least one embodiment, for example, at a time step, GPUs 120a-n may all be processing data samples 102a-n sampled from bucket #2 of subsets 122a-n. In at least one embodiment, in this way, at any given time step, GPUs 120a-n may be processing data samples 102a-n with similar sample sizes. In at least one embodiment, for example, at each training step, GPUs 120a-n may be processing data samples 102a-n that each contain a similar or same number of sequences such that each GPU 120a-n may process a data sample of a similar structure (paragraph 68);
“ the buffer structure comprising a first bucketing tier…… corresponding to ……of ……sequenced data and a second bucketing tier corresponding to ……of the variable length sequenced data” as the structure includes, at a first level, a first bucketing subset as a first bucketing tier corresponding to a sample sizes of the training dataset 102, which comprises data samples 102a-n having a sequence length of (1±5%) n, is represented as the variable length sequenced data and at a second level a second bucketing subset as a second bucketing tier corresponding to a sample sizes of the training dataset 102, which comprises data samples 102a-n having a sequence length of (1±5%) n, is represented as the variable length sequenced data (fig. 2, paragraphs 59-63, 72);
“wherein the batch of data samples is served from a selected second-tier bucket of the buffer structure and comprises a data sample batch size determined based at least on ……of the one or more GPUs and ……of the selected second-tier bucket” as the subset of data samples is assigned as served from an assigned second-subset bucket e.g., bucket 1 at a first level of the structure ((fig. 2, paragraphs 68-69), and the subset of data samples as the batch of data samples has a data sample subset size defined based on a number buckets of the GPU and second sample size range as a bucket size of the defined second-level bucket e.g., bucket 1 of the second subset (fig. 2, paragraphs 64-68).
In particularly: within each subset 122a-n, training samples may be bucketized into a number of buckets according to their sample sizes, e.g., the number of sequences a data sample contains, the total length of a data sample, and/or the like. In at least one embodiment, as shown in FIG. 2 for example, for each subset 122a-n containing a total number of M data samples, M data samples may be bucketized into bucket #1 of a first sample size range, bucket #2 of a second sample size range . . . , bucket #m of a mth sample size range. In at least one embodiment, a sample size may be measured by a number of sequences a data sample contains, a total length of tokens a data sample contains, and/or the like (paragraph 65).
A sample size ranges for buckets 1-m may be determined by dividing a difference between a maximum sample size and a minimum sample size in each subset 122a-n into m intervals and each interval may correspond to a size range for a bucket (paragraph 66).
Subsets 122a-n have similar distributions of different sample sizes, each subset 122a-n may be similarly bucketized into buckets 1-m based on similar size ranges. In at least one embodiment, for each subset 122a-n, buckets 1-m may be ranked in a same order according to their respective sample size ranges (paragraph 67).
AI further teaches limitations
“serving a batch of data samples of variable length sequenced data” as receiving or maintaining as serving a data stream as a batch of samples of sequentially arriving tuples as sequenced data (paragraph 11) of a size e.g., K is number as variable length of the tuples that are arriving tuples as sequenced data (paragraph 6);
“the buffer structure comprising a first bucketing tier corresponding to ……of the variable length sequenced data and a second bucketing tier corresponding to ……of the variable length sequenced data” as a memory as the buffer structure includes a first reservoir corresponding to a size e.g., K is number as variable length of the tuples that are arriving tuples as sequenced data (paragraphs 6 (table 2), 11-13) and a second reservoir corresponding to a size e.g., K is number as variable length of the tuples that are arriving tuples as sequenced data (paragraphs 6 (table 2, paragraphs 11-13).
In particularly:
The algorithm places all tuples in the reservoir until the reservoir (of the size of r tuples) becomes full. After that, each k.sup.th tuple is sampled with the probability r/k. A sampled tuple replaces a randomly selected tuple in the reservoir. It is easy to verify that the reservoir always holds a uniform random sample of the k tuples seen so far. Conventional reservoir sampling assumes a fixed size reservoir (i.e., the size of a sample is fixed) (paragraph 6).
The machine memory has a limited size and the method further includes: establishing in the machine memory a plurality of sampling reservoirs each having a size; receiving a plurality of data streams each containing a plurality of sequentially arriving tuples, the plurality of data streams corresponding respectively to the plurality of sampling reservoirs; checking whether the size of any one or more of the plurality of sampling reservoirs should be changed; and for each of the plurality of sampling for which the size should be changed, adjusting the size of that one of the plurality of sampling reservoirs as a function of the limited size of the machine memory (paragraph 12);
“……determined based at least on a memory utilization of……and ……” as the stream of samples is received (paragraphs 11-13) and a tuple added as determined based on a memory utilization of join buffer and size of the reservoir (paragraphs 94-95);
“the variable length sequenced data” as size r of a sampling reservoir that can be increased or decreased is represented as variable length: paragraphs 35, 39, 41). The size r of a sampling reservoir which stores sequentially arriving tuples (paragraphs 18-19) is represented as variable length sequenced data.
AI and Re disclose a method of receiving and storing data into buckets. These references are same field as application’s field. Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply AI’s teaching to Re’s system to allow one or more sample reservoirs to change size during sampling without loss data, and further to reduce network communications and further to increase lifetime of the network.
Free further teaches limitations
“a buffer structure” as hierarchy of folders (figs. 15-16, paragraphs 53, 168-169);
the buffer structure comprising a first bucketing tier that includes a first plurality of data buckets and a second bucketing tier that includes a second plurality of data buckets as a hierarchy of folders as the buffer structure includes a first level as a first bucketing tier of a hierarchy as the buffer structure includes a plurality of data folders 712 as data buckets (figs. 7, 11, paragraphs 150, 161); and a second level as a second bucketing tier of the hierarchy include a plurality of second data folders 714 as data buckets (figs. 7, 11, paragraphs 150, 161).
Free and Re disclose a method of receiving and storing data into folders. These references are same field as application’s field. Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Free’s teaching to Re’s system in order to minimize connection times and costs of content when data is requested, to minimize time required to retrieve requested data, and further to provide faster effective transfer rates.
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Re in view of AI and Free and further in view of Choi et al (US 20240338789).
As to claim 15, Re, AI and Free teach limitation
“the one or more processors are further to determine the data sample batch size based at least on iterative execution of a training step of a machine learning model on the one or more GPUs to……” as the processor is further to (Re: paragraph 86) define S.sub.[r] (sample S of size r) as the data sample batch size based on (AI: paragraph 41), the training step of a neural network to update a portion of neural network e.g., machine learning model (Re: 230) on GPU (Re: fig. 4, paragraph 75-77), to maintain the sample in a reservoir (AI: paragraph 41).
Re, AI and Free do not explicitly teach limitation
converge on a valid batch size that does not generate a memory error.
Choi teaches limitation
“converge on a valid batch size that does not generate a memory error” as reduce or adjust data batch size of training dataset as a valid batch size to make sure a GPU model circuitry 155 do not return an out-of-memory (OOM) error (fig. 1, paragraphs 38, 58).
In particularly: [0038] When the current estimated batch size consumes more than the available system memory, the GPU model circuitry 155 returns an out-of-memory (OOM) error, and the batch size adaptation circuitry 115 reduces its estimated batch size to increase the memory margin. On the other hand, if the current estimated batch size shows low memory usage, then the batch size adaptation circuitry 115 can increase the estimated batch size based on this information. When estimating the batch size, a bigger batch size can be more informative because each update to the model's parameters is based on more data points.
Choi further teaches limitation
“determine the data sample batch size …… to converge on a valid batch size that does not generate a memory error” as determine the data batch size of the training dataset consumes more than the available system memory, to reduce the data batch size of the training dataset so that the circuitry does not return memory error (fig. 1, paragraphs 20, 38, 58).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Choi’s teaching to Re’s system to modify a batch size to allow more datasets to be processed simultaneously during each training operation, which improves the efficiency of the training pipeline.
Claim 17 is rejected under 35 U.S.C. 103 as being unpatentable over Re in view of AI and Free and further in view of Slavitch et al (or hereinafter “Sla”) (US 20210365458)
As to claim 17, Re, AI, and Free teach limitations
“wherein the one or more processors are further to determine ……for at least one of the first bucketing tier and the second bucketing tier based on computing……” as the processor is further to (Re: paragraph 80) adjust as determine sizes for a first reservoir as the first bucketing tier based on sampling as computing rates of sensors 104 (AI: paragraph 42; Re: paragraphs 59-64, figs. 1-2).
Re, AI, and Free do not explicitly teach limitation
bucket bin boundaries; one or more sequence length histograms.
Sla teaches limitation
“bucket bin boundaries” as define bucket bin boundaries (paragraphs 47, 51);
“one or more sequence length histograms” as a sequence length histogram (fig. 1, paragraph 42-43).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Sla’s teaching to Re’s system in order to provide improved performance for different types of workloads and further to provide a significant reduction in the number of cache lines being read.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Noel et al (US 20150040052) discloses stores 103 may be distributed across multiple indexers, each responsible for storing and searching a subset, or buckets, of the events generated by the system (paragraph 189).
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.
Contact Information
Any inquiry concerning this communication or earlier communications from the examiner should be directed to CAM-Y T TRUONG whose telephone number is (571)272-4042. The examiner can normally be reached (571) 272 4042.
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, SHERIEF BADAWI can be reached at (571) 272-9782. 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.
/CAM Y T TRUONG/ Primary Examiner, Art Unit 2169
