DETAILED ACTION
Notice of AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendment
Applicant’s Amendment and remarks dated 9/17/2026 have been considered. Claims 1-20 are pending.
Response to Arguments
On page 6 of Applicant’s 9/17/2026 Amendment and remarks, Applicant asserts that no new matter has been added via the amendments to the claims.
The examiner agrees that at least paras. 0058-0062 and 0080-0086 provide sufficient written description support for the claim amendments.
On pages 7-11 of Applicant’s 9/17/2026 Amendment and remarks, with respect to the rejections to all claims under 35 U.S.C. 101, Applicant makes several arguments.
Applicant’s arguments on pages 9-10, with respect to Step 2A, Prong 2, which explain why the present claims (as amended) improve “GPU-based neural-network processing systems” are persuasive. All rejections under 35 U.S.C. 101 are hereby withdrawn. The examiner respectfully submits that Applicant’s remaining arguments under 35 U.S.C. 101 are moot in view of the withdrawal of all such rejections.
On pages 12-13 of Applicant’s 9/17/2026 Amendment and remarks, with respect to the rejections to the independent claims under 35 U.S.C. 102 as anticipated by the LIN reference , Applicant argues that the claim amendments overcome such rejection.
The examiner agrees. All rejections under 35 U.S.C. 102 are hereby withdrawn. However new grounds of rejection under 35 U.S.C. 103 are provided herein, where such new grounds of rejection are necessitated by Applicant’s claim amendments to the independent claims.
On page 13 of Applicant’s 9/17/2026 Amendment and remarks, with respect to the rejections to the dependent claims under 35 U.S.C. 103, Applicant argues that the amendments to the independent claims should make the dependent claims allowable.
The examiner respectfully disagrees for the same reasons explained with respect to the independent claims. New grounds of rejection for each of the dependent claims are set forth herein, where such new grounds of rejection are necessitated by Applicant’s claim amendments to the independent claims.
Claim Rejections - 35 USC § 103
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1, 3, 6, 8, 10, 13, 15, 17, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over US 20250209336 A1, hereinafter referenced as LEE, in view of US 20220398747 A1, hereinafter referenced as LIN, hereinafter referenced as LIN, and further in view of US 20250065916 A1, hereinafter referenced as GUPTA.
Regarding Claim 1
LEE teaches:
One or more processors, comprising: circuitry to: (LEE, para. 0011: “The device comprises a processor and a memory coupled with the processor. The processor executes computer-readable instructions stored in the memory to perform operations.”)
identify one or more first portions of first information by portioning the first information into the one or more first portions sized to fit within an amount of available shared memory of a graphics processing unit (GPU) … to generate second information (LEE, para. 0048: “In this stage, we aim to encode the cells in each chunk and represent the collective phenotype of the cells using latent network embedding. The present disclosure combines the immunophenotype representation learning and the classifier training in a discriminative deep network using a GMM probabilistic distribution assumption for the hematologic malignancy type classification tasks. … As a result, the supervised deep network can optimize the latent GMM weights for the targeted hematologic malignancy labels. After training, we can then extract the latent embedding in the network (e.g., the chunk-level pooling network 250) to represent each chunk 241 of the sample 211.”;
LEE, para. 0127: “The chunking process (e.g., the operation of chunking 240 or operation 305) acts as the key step to enable efficient GPU memory usage and the improvement of accuracy. The chunk size can be simply determined by the available GPU memory size, and we explored the performance change with different chunk sizes. The FC data from the UPMC and hema.to datasets mostly include a standard number of collected cells. Therefore, the change in chunk size will change the chunk quantity. That is, a small chunk size causes only a few cells to be used as a training sample for the deep network, while the total number of training samples become larger.”
Examiner’s Note: the chunks with respect to cell data correspond to the recited “first portions of first information” and the output malignancy classification corresponds to the recited “second information”)
However, LEE fails to explicitly teach:
cause one or more neural networks to: … to be used by the one or more neural networks …
comprising an expected relevance between the one or more first portions of the first information and one or more second portions of the first information.
However, in a related field of endeavor (deep neural networks, see paras. 0048-0049), LIN teaches and makes obvious:
cause one or more neural networks to: (LIN, para. 0066: “As described above, the flow search area can be sized according to characteristics of the dataset being analyzed by the optical flow estimation system. ... The size of the flow search area selected by search area engine 104 can determine the computational and memory requirements for performing optical flow estimation. Accordingly, the search area engine 104 can also determine the size of the flow search are based at least in part on the available computing and memory resources (e.g., the amount of TCM) and the processing speed requirements for the optical flow estimation (e.g., whether the optical flow estimation needs to occur in real-time). In some cases, a neural network based machine learning system and/or algorithm (e.g., a deep neural network) can be used to determine the size of the flow search area.”;
Examiner’s Note: LIN discloses specifically using a neural network to select or identify data sizing; the LEE-LIN combination now modifies LEE so that instead of a computer determining the chunk size, now the deep neural network of LIN determines the chunk size; the examiner notes that LEE discloses using deep networks (see para. 0048), and it would therefore have been straightforward to modify LEE to have deep networks performing the chunking instead of a generic computer) … to be used by the one or more neural networks … (LIN, para. 0066: “As described above, the flow search area can be sized according to characteristics of the dataset being analyzed by the optical flow estimation system. ... The size of the flow search area selected by search area engine 104 can determine the computational and memory requirements for performing optical flow estimation. Accordingly, the search area engine 104 can also determine the size of the flow search are based at least in part on the available computing and memory resources (e.g., the amount of TCM) and the processing speed requirements for the optical flow estimation (e.g., whether the optical flow estimation needs to occur in real-time). In some cases, a neural network based machine learning system and/or algorithm (e.g., a deep neural network) can be used to determine the size of the flow search area.”;
Examiner’s Note: LIN discloses specifically using a neural network to select or identify data sizing; the LEE-LIN combination now modifies LEE so that the deep neural networks of LIN are used together with the deep networks of LEE (see para. 0048) to generate the malignancy classification)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of LEE with LIN as explained above. One of ordinary skill would have been motivated to do so specifically have a neural network (instead of a generic computer or processor) to perform operations, as one of ordinary skill would understand that a trained neural network provides advantages over rules-based approaches carried out by a generic computer or processor. As disclosed by LIN, one of ordinary skill would be motivated to do so in order to take advantage of the techniques of LIN to “improve the efficiency” of determining the output of a deep neural network. (para. 0091).
However, LEE and LIN fail to explicitly teach:
comprising an expected relevance between the one or more first portions of the first information and one or more second portions of the first information.
However, in a related field of endeavor (neural networks used with respect to image data, see para. 0027), GUPTA teaches and makes obvious:
comprising an expected relevance between the one or more first portions of the first information and one or more second portions of the first information (GUPTA, para. 0036: “The trajectory model 212 may include a recurrent neural network (RNN) operably encoding the sequential kinematic data and output spatiotemporal output data, which include, without limitations, the kinematic data, road and lane data, and their relevance. ... By doing so, the input data and their relevance may be obtained and compressed to be output as the output data, such as spatiotemporal output data. The output data may include all the information in the input data and the relevance between the input data.”;
Examiner’s Note: the LEE-LIN-GUPTA combination now modifies the neural networks of LIN so that such neural networks also output the relevance between different portions of the input data of LEE)
Before the effective filing date, it would have been obvious to one of ordinary skill in the art to combine the teachings of LEE with LIN and GUPTA as explained above. As disclosed by GUPTA, one of ordinary skill would have been motivated so that “each layer learns the relevance of input data and selects input data based on the relevance to pass to the next layer such that the next layer retains the data of input data are more relevant” and therefore a neural network utilizes higher relevancy data when making a prediction. (para. 0036).
Regarding Claim 3
LEE, LIN, and GUPTA teach the one or more processors of claim 1 as explained above. LEE further teaches:
adjust the one or more first portions based at least on a change in an amount of available shared memory of one or more GPUs. (LEE, para. 0127: “The chunking process (e.g., the operation of chunking 240 or operation 305) acts as the key step to enable efficient GPU memory usage and the improvement of accuracy. The chunk size can be simply determined by the available GPU memory size, and we explored the performance change with different chunk sizes. The FC data from the UPMC and hema.to datasets mostly include a standard number of collected cells. Therefore, the change in chunk size will change the chunk quantity. That is, a small chunk size causes only a few cells to be used as a training sample for the deep network, while the total number of training samples become larger.”
Examiner’s Note: the chunks with respect to cell data correspond to the recited “first portions of first information” which is adapted specifically to the available GPU memory size)
Regarding Claim 6
LEE, LIN, and GUPTA teach the one or more processors of claim 1 as explained above. LEE further teaches:
wherein the amount of available shared memory is based, at least in part, on one or more allocations to a GPU shared memory (LEE, para. 0127: “The chunking process (e.g., the operation of chunking 240 or operation 305) acts as the key step to enable efficient GPU memory usage and the improvement of accuracy. The chunk size can be simply determined by the available GPU memory size, and we explored the performance change with different chunk sizes. The FC data from the UPMC and hema.to datasets mostly include a standard number of collected cells. Therefore, the change in chunk size will change the chunk quantity. That is, a small chunk size causes only a few cells to be used as a training sample for the deep network, while the total number of training samples become larger.”
Examiner’s Note: one of ordinary skill in the art would understand that the available GPU memory size is impacted by the amount of GPU memory already allocated to other processes)
Regarding Claim 8
LEE teaches:
A system comprising: one or more processors to cause (LEE, para. 0009: “In accordance with various embodiments, exemplary systems, methods, devices and computer program products are disclosed herein.”;
LEE, para. 0011: “The device comprises a processor and a memory coupled with the processor. The processor executes computer-readable instructions stored in the memory to perform operations.”)
The remaining limitations correspond to the processor of claim 1, and are therefore rejected for the same reasons explained above with respect to claim 1.
Claim 10 depends from claim 8 and recites a system that corresponds to the processor of claim 3, and is therefore rejected for the same reasons explained with respect to claims 3 and 8.
Claim 13 depends from claim 8 and recites a system that corresponds to the processor of claim 6, and is therefore rejected for the same reasons explained with respect to claims 6 and 8.
Claim 15 recites a method that corresponds to the processor of claim 1, and is therefore rejected for the same reasons explained above with respect to claim 1.
Claim 17 depends from claim 15 and recites a method that corresponds to the processor of claim 3, and is therefore rejected for the same reasons explained with respect to claims 3 and 15.
Claim 20 depends from claim 15 and recites a method that corresponds to the processor of claim 6, and is therefore rejected for the same reasons explained with respect to claims 6 and 15.
Claims 2, 9, and 16 are rejected under 35 U.S.C. 103 as being unpatentable over LEE, LIN, and GUPTA and further in view of US 20150081319 A1, hereinafter referenced as KEMP.
Regarding Claim 2
LEE, LIN, and GUPTA teach the one or more processors of claim 1 as explained above. However, LEE, LIN, and GUPTA fail to explicitly teach:
wherein the second information further comprises data format information
However, in a related field of endeavor to at least LEE (medical records with respect to the medical field, see para. 0002), KEMP teaches and makes obvious:
wherein the second information further comprises data format information (KEMP, para. 0030: “A data-point is an extraction of particular data from a medical record in accordance with a specification. The format for data-points may be controlled for consistency, readability and relevancy to medical importance.”;
Examiner’s Note: the LEE-LIN-GUPTA-KEMP combination now modifies the neural networks of LIN so that such neural networks also output the relevance (as in GUPTA) between different portions of the input data of LEE and additionally outputs information concerning how consistent and readable the data format is as in KEMP)
Before the effective filing date, it would have been obvious to one of ordinary skill in the art to combine the teachings of LEE with LIN, GUPTA, and KEMP as explained above. As disclosed by KEMP, one of ordinary skill would have been motivated to do so because determining the consistency, readability, and relevancy of data forms is of “medical importance.” (para. 0030). One of ordinary skill would further have been motivated to confirm whether the data types are consistent with one other to determine if additional processing is required for downstream applications.
Claim 9 depends from claim 8 and recites a system that corresponds to the processor of claim 2, and is therefore rejected for the same reasons explained with respect to claims 2 and 8.
Claim 16 depends from claim 15 and recites a method that corresponds to the processor of claim 2, and is therefore rejected for the same reasons explained with respect to claims 2 and 15.
Claims 4, 11, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over LEE, LIN, and GUPTA and further in view of US 20220200669 A1, hereinafter referenced as GOWDA.
Regarding Claim 4
LEE, LIN, and GUPTA teach the one or more processors of claim 1 as explained above. However, LEE, LIN, and GUPTA fail to explicitly teach:
wherein the circuitry is to use the one or more neural networks to perform one or more matrix multiplication operations using information of the one or more first portions.
However, in a related field of endeavor (neural networks, see para. 0092), GOWDA teaches and makes obvious:
wherein the one or more circuits are to use the one or more neural networks to perform one or more matrix multiplication operations using information of the one or more first portions. (GOWDA, para. 0075: “In at least one embodiment, precoding technique 400 includes generating precoding weights as values in a coefficient matrix W.sup.RZF=M.sup.H(D.sup.−1L.sup.−1), where D=diag(U). In at least one embodiment, augmented matrix 408 is referred to as a first augmented matrix, intermediate matrix 412 is referred to as a second augmented matrix, and generating precoding weights includes generating a conjugate transpose matrix M.sup.H of a first portion (e.g., matrix M) of second augmented matrix, and performing a matrix multiplication operation (e.g., multiplying M.sup.H by (D.sup.−1L.sup.−1)) based, at least in part, on generated conjugate transpose matrix M.sup.H.”
GOWDA, para. 0429: “In at least one embodiment, tensor cores are configured to perform deep learning matrix arithmetic, such as convolution operations for neural network training and inferencing. In at least one embodiment, each tensor core operates on a 4×4 matrix and performs a matrix multiply and accumulate operation D=A×B+C, where A, B, C, and D are 4×4 matrices.”;
Examiner’s Note: the LEE-LIN-GUPTA-GOWDA combination now performs neural network training using matrix multiplication as disclosed by GOWDA)
Before the effective filing date, it would have been obvious to one of ordinary skill in the art to combine the teachings of LEE with LIN, GUPTA, and GOWDA as explained above. As disclosed by GOWDA, one of ordinary skill would have been motivated to utilize a GPU for parallel processing. (para. 0074). As further disclosed by GOWDA, one of ordinary skill would have been motivated to do so in order to “accelerate machine learning application frameworks.” (para. 0276).
Claim 11 depends from claim 8 and recites a system that corresponds to the processor of claim 4, and is therefore rejected for the same reasons explained with respect to claims 4 and 8.
Claim 18 depends from claim 15 and recites a method that corresponds to the processor of claim 4, and is therefore rejected for the same reasons explained with respect to claims 4 and 15.
Claims 5, 12, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over LEE, LIN, and GUPTA and further in view of US 20060182357 A1, hereinafter referenced as LIU.
Regarding Claim 5
LEE, LIN, and GUPTA teach the one or more processors of claim 1 as explained above. However, LEE, LIN, and GUPTA fail to explicitly teach:
wherein the one or more first portions comprise are represented using a data format dynamically selected based, at least in part, on the amount of available storage.
However, in a related field of endeavor (handling graphics and video content, see para. 0002), LIU teaches and makes obvious:
wherein the one or more first portions comprise are represented using a data format dynamically selected based, at least in part, on the amount of available storage. (LIU, para. 0008: “The present invention, termed herein an Intelligent, dynamic, long-term digital Surveillance Media Storage System (ISMS), is a long-term digital media storage system with intelligent decision making based on event and content semantics to dynamically select the encoding method for new recording based on application defined priority and accuracy requirements. At the same time, it also, dynamically selects and converts previously recorded media to a less accurate recording format based on time and relative priority between a set of candidate media files and thus, forms a multistage media data compression mechanism that can best reduce the total storage size iteratively.”;
LIU, para. 0024: “Thus, the system can dynamically and intelligently select a proper encoding mechanism to record or transcode the video to long-term surveillance media storages. As a result, this flexible profile architecture allows extending the system to fit any application needs.”;
LIU, claim 3: “wherein said encoding optimizer processes media streams in multiple stages over time, each stage including dynamic selection of an encoding format based on an amount of available storage space”;
Examiner’s Note: the LEE-LIN-GUPTA-LIU combination now modifies LEE to select the available memory resources (as in LEE) to be formatted in a certain way based on the dynamic selection of an encoding format based on an amount of available storage space as in LIU)
Before the effective filing date, it would have been obvious to one of ordinary skill in the art to combine the teachings of LEE with LIN, GUPTA, and LIU as explained above. As disclosed by LIU, one of ordinary skill would have been motivated to select a particular encoding format in order to meet certain priorities, including at least available storage space. (para. 0008). As further disclosed by LIU, one of ordinary skill would have been motivated to do so in order to “reduce the storage demands.” (para. 0021).
Claim 12 depends from claim 8 and recites a system that corresponds to the processor of claim 5, and is therefore rejected for the same reasons explained with respect to claims 5 and 8.
Claim 19 depends from claim 15 and recites a method that corresponds to the processor of claim 5, and is therefore rejected for the same reasons explained with respect to claims 5 and 15.
Claims 7 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over LEE, LIN, and GUPTA in view of US 20220301228 A1, hereinafter referenced as JUNKINS.
Regarding Claim 7
LEE, LIN, and GUPTA teach the one or more processors of claim 1 as explained above. However, LEE, LIN, and GUPTA fail to explicitly teach:
wherein the first information is represented using a tile data format of one or more buffers to be stored in a shared memory
However, in a related field of endeavor (graphics processing, see para. 0066), JUNKINS teaches and makes obvious:
wherein the first information is represented using a tile data format of one or more buffers to be stored in a shared memory. (JUNKINS, para. 0270: “Writing Native Tiling from Shared Local Memory Buffer—In some embodiments, an additional method refinement would allow compute shader kernels such as the Huffman decoder to directly write native tiled GPU format for block compressed textures.”;
Examiner’s Note: The LEE-LIN-GUPTA-JUNKINS combination now modifies LEE to represent the available memory capacity as tiled GPU format as in JUNKINS).
Before the effective filing date, it would have been obvious to one of ordinary skill in the art to combine the teachings of LEE with LIN, GUPTA, and JUNKINS as explained above. As disclosed by JUNKINS, one of ordinary skill would have been motivated to do so in order to provide a “capability [that] would alleviate the GPU from having to run a secondary copy engine postprocessing pass to store the decoded texture in the native tiled GPU format.” (para. 0270).
Claim 14 depends from claim 8 and recites a system that corresponds to the processor of claim 7, and is therefore rejected for the same reasons explained with respect to claims 7 and 8.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
US 20220270357 A1 (Goyal). “Described herein is the use of a U-Net based neural network architecture to predict an initial segmentation mask as is depicted at FIG. 1A (element 105). Further described is a trained 3D nn-UNet as it provides a robust and self-adapting framework with the ability to dynamically adapt to the details of the datasets (median patient size, input patch size, batch size, etc.) and amount of available GPU memory.” (para. 0053).
US 20220172039 A1 (Liu). “Batch size depends on several factors, including the available memory on the computing device or GPU.” (para. 0045).
US 20200175678 A1 (Abraham). “This network requires significant graphics processing unit (GPU) resources and is also limited by the amount of the available GPU memory. The algorithm therefore implements a tiling strategy with configurable overlapping. Tile sizes are configurable but can also be chosen automatically depending on the available GPU memory.” (para. 0105).
US 20220351112 A1 (Mochty). “In such an example, the predefined set of statistical rules are employed to validate the input data according to at least one of: data type, data value, data structure, data consistency, data format, data accuracy, and data relevance.” (para. 0090).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MICHAEL C LEE whose telephone number is (571)272-4933. The examiner can normally be reached M-F 12:00 pm - 8:00 pm ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Omar Fernandez Rivas can be reached at 571-272-2589. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/MICHAEL C. LEE/Examiner, Art Unit 2128
