DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is made final.
Claims 1-30 are pending. Claims 1, 10, 16, and 25 are independent claims.
Response to Arguments
Applicant’s arguments, dated 10/29/2025, regarding the 35 U.S.C. 101 rejections of the previous office action have been fully considered but are not persuasive. Due to the claim amendments, the scope of the claims has changed and new grounds of rejection are applied – see the updated rejection below.
Applicant argues that the claims embody a technical solution of providing tensor densification by squeezing tensors, which can reduce a storage and memory footprint, and be used efficiently in convolution. The examiner argues that the possible storage/memory savings and possible more efficient computation are not present in the claims as there is no mention of sizes of densified weight tensor and sparsity map (in figures 2A and 2B of the instant application, it seems that the densified structure and sparsity map can possibly result in a combined size equivalent to that of the original matrix) or how convolution operations are performed using the densified weight tensor (instead, Claim 1 recites: “performing a convolution operation using a machine learning model”). The applicant does not appear to specify any particular additional elements within the claims that can integrate the abstract ideas into a practical application.
Applicant’s arguments, dated 10/29/2025, regarding the 35 U.S.C. 103 rejections of the previous office action have been fully considered but are not persuasive. Due to the claim amendments, the scope of the claims has changed and new grounds of rejection are applied – see the updated rejection below.
Claim Objections
Claim 25 is objected to because of the following informalities: the limitation “wherein the densified weight tensor was generated by directionally squeezing a weight tensor, having unstructured sparsity, to remove zero-value elements, wherein directionally squeezing the weight tensor comprises maintaining non-zero elements of the weight tensor” is included twice in the claim. Appropriate correction is required.
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 1-30 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1:
Step 1: This part of the eligibility analysis evaluates whether the claim falls within any statutory category. See MPEP 2106.03. Claim 1 is directed to a method (Step 1: YES).
Step 2A prong 1: Does the claim recite a judicial exception? Claim 1 recites: A processor-implemented method, comprising… generating a densified weight tensor based on the weight tensor, comprising: directionally squeezing the weight tensor to remove zero-value elements; wherein directionally squeezing the weight tensor comprises maintaining non-zero elements of the weight tensor (directionally squeezing a tensor can be performed mentally with the aid of pen and paper, i.e., removing a row of zero values or removing zeroes from each row or column); and generating a sparsity map based on the directional squeezing (generating a sparsity map can be performed mentally with the aid of pen and paper, i.e., recording indices where zeros removed in the previous step) and performing a convolution operation… to generate an output tensor (a convolution operation is a mathematical calculation). These steps can be performed mentally, or are mathematical calculations (Step 2A prong 1: YES).
Step 2A prong 2: Does the claim recite additional elements? Do those additional elements, considered individually and in combination, integrate the judicial exception into a practical application? Claim 1 recites: accessing a weight tensor having unstructured sparsity… using a machine learning model, the densified weight tensor, and the sparsity map… and generating an inference using the machine learning model based on the output tensor. Accessing a weight tensor having unstructured sparsity and generating an inference using a machine learning model are insignificant extra-solution activities of data gathering and data outputting, respectively. Performing the convolution operation using a machine learning model is mere instructions to implement the abstract idea on a generic computer, recited at a high level of generality, and using the densified weight tensor and sparsity map in the convolution is insignificant extra-solution activity of selecting a particular data source (MPEP 2106.05(g)). (Step 2A prong 2: NO).
Step 2B: These elements are recited at such a high level of generality that they fail to integrate the abstract idea into a practical application, since they since they only amount to data gathering or outputting without significantly more (MPEP 2106.05(g)), selecting data to be manipulated (MPEP 2106.05(g)) without particularly more, or provide nothing more than mere instructions to implement an abstract idea on a generic computer (MPEP 2106.05(f)). These limitations, taken either alone or in combination, fail to provide an inventive concept (Step 2B: NO). Thus, the claim is not patent eligible.
Regarding claims 2-9, they recite limitations which further narrow the abstract idea by specifying more details of the mental and mathematical process that occurs (Claim 2, compiling at least part of a machine learning model based on a densified weight tensor is data outputting without significantly more; Claim 3, setting an indicator that a weight tensor is densified in the compiled model is data outputting without significantly more; Claim 4, squeezing the weight tensor as described is still a mental process; Claim 5, selecting a direction is a mental process; Claim 6, having the sparsity map match dimensions of the densified weight tensor can be done mentally, i.e., the map may have the same width or height – and the sparsity map could be created mentally, with the aid of pen and paper, to associate densified weight elements with activation output values in subsequent tensor computations; Claim 7, the mentally created sparsity map could use an absolute indication that corresponds to the activation output, i.e., define a row or column that densified weight data corresponds to; Claim 8, the mentally created sparsity map could use a relative spacing indication that corresponds to the output, i.e., define how many elements in the output to skip based a count of zero valued weights in a row or column; Claim 9, evaluating the sparsity of the unstructured sparsity data and comparing it to a threshold is a mental process or set of mathematical calculations).
Regarding claim 10,
Step 1: This part of the eligibility analysis evaluates whether the claim falls within any statutory category. See MPEP 2106.03. Claim 10 is directed to a method (Step 1: YES).
Step 2A prong 1: Does the claim recite a judicial exception? Claim 10 recites:
A processor-implemented method of machine learning, comprising… performing the convolution operation using a machine learning model (performing a convolution operation is a mathematical calculation), comprising… wherein the densified weight tensor was generated by directionally squeezing a weight tensor, having unstructured sparsity, to remove zero-value elements, wherein directionally squeezing the weight tensor comprises maintaining non-zero elements of the weight tensor (squeezing a sparse tensor to remove zero values is a mental process, i.e., it can be accomplished in the human mind with the aid of pen and paper); generating a set of intermediate tensors by multiplying the densified weight tensor and the activation tensor (multiplying tensors is a mathematical calculation); and generating an output tensor for the convolution operation by accumulating the set of intermediate tensors based on the sparsity map (accumulating tensors is a mathematical calculation). These steps are mathematical calculations or mental processes (Step 2A prong 1: YES).
Step 2A prong 2: Does the claim recite additional elements? Do those additional elements, considered individually and in combination, integrate the judicial exception into a practical application? Claim 10 recites: accessing an activation tensor for processing using a convolution operation… retrieving a densified weight tensor for the convolution operation… retrieving a sparsity map associated with the densified weight tensor… and generating an inference using the machine learning model based on the output tensor. Accessing an activation tensor, retrieving a densified weight tensor and retrieving an associated sparsity map are extra-solution activities of data gathering that do not add a meaningful limitation to the tensor densification method. Generating an inference using the machine learning model is insignificant extra-solution activity of data outputting (Step 2A prong 2: NO).
Step 2B: These elements are recited at such a high level of generality that they fail to integrate the abstract idea into a practical application, since they since they only amount to data gathering or data outputting without significantly more (MPEP 2106.05(g)). These limitations, taken either alone or in combination, fail to provide an inventive concept (Step 2B: NO). Thus, the claim is not patent eligible.
Regarding claims 11-15, they recite limitations which further narrow the abstract idea by specifying more details of the mental and mathematical process that occurs (Claim 11, creating a sparsity map that indicates where the processed tensor weights correspond to can be performed mentally; Claim 12, the mentally created sparsity map could use an absolute indication that corresponds to the output, i.e., define a row or column that densified weight data corresponds to; Claim 13, the mentally created sparsity map could use a relative spacing indication that corresponds to the output, i.e., define how many elements in the output to skip based a count of zero valued weights in a row or column; Claim 14, identifying relevant elements of intermediate tensors using the sparsity map is a mental process, and accumulating those elements is a mathematical calculation; Claim 15, indicating that the weight tensor is densified is a mental process).
Regarding claim 16, it is a system implementing the method of claim 1 and is rejected on the same grounds – see above.
Regarding claims 17-24, they recite similar limitations as claims 2-19 and are rejected on the same grounds – see above.
Regarding claim 25, it is a system implementing the method of claim 10 and is rejected on the same grounds – see above.
Regarding claims 26-30, they recite similar limitations as claims 11-15 and are rejected on the same grounds – see above.
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.
Claim(s) 1, 4, 5, 6, 7, 10, 11, 12, 13, 14, 15, 16, 19, 20, 21, 22, and 25-30 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chinya et al. (US 20210042617 A1), herein Chinya, in view of Shin et al. (WO 2022241168 A1), herein Shin.
Regarding claim 1, Chinya teaches: A processor-implemented method (¶55, the system 158 includes a host processor 160 (e.g., CPU with one or more processor cores)), comprising: accessing a weight tensor having unstructured sparsity (¶16, The sparsity aware compression scheme may operate on unstructured sparsity data); generating a densified weight tensor based on the weight tensor, comprising: directionally squeezing the weight tensor to remove zero-value elements… and generating a sparsity map based on the directional squeezing (¶18, Zero values may be removed from the workload to compress the data of the workload 104 – see Fig. 3 for sparsity bitmap 360 and compressed data 354).
Chinya fails to explicitly teach: wherein directionally squeezing the weight tensor comprises maintaining non-zero elements of the weight tensor… and performing a convolution operation using a machine learning model, the densified weight tensor, and the sparsity map to generate an output tensor; and generating an inference using the machine learning model based on the output tensor.
However, in the same field of endeavor, Shin teaches: wherein directionally squeezing the weight tensor comprises maintaining non-zero elements of the weight tensor (Fig. 2, squeezing the weight tensor results in non-zero values being maintained –see, e.g., highlighted portion of original matrix 202, where non-zero elements 1-32 are maintained in the compressed matrix 206. Also note the sparsity map produced 208)… and performing a convolution operation using a machine learning model, the densified weight tensor, and the sparsity map to generate an output tensor; and generating an inference using the machine learning model based on the output tensor. (¶47, In at least one embodiment, sparse matrix multiplication operations are performed as part of training or performing a neural network, convolution, or a machine learning operation – and – ¶79, by implementing techniques described herein, an increase of processing speed also results in faster end-to-end training times and inferencing times).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to maintain non-zero elements of the weight tensor, perform a convolution operation and generate an inference as disclosed by Shin in the method disclosed by Chinya to improve reduce unnecessary computation and save on memory (¶79, with a sparse matrix, less computational operations occur while also using less memory).
Regarding claim 4, Chinya further teaches: The processor-implemented method of Claim 1, wherein directionally squeezing the weight tensor comprises: removing the zero-value elements from the weight tensor; and compressing the non-zero elements of the weight tensor along one dimension in the weight tensor (Fig. 3, compressed data 354 has fewer zeroes and has been vertically compressed).
Regarding claim 5, Chinya further teaches: The processor-implemented method of Claim 1, wherein directionally squeezing the weight tensor further comprises: selecting either a vertical direction… and squeezing the weight tensor in the selected direction (Fig. 3, compressed data 354 has fewer zeroes and has been vertically compressed).
Chinya fails to teach: or a horizontal direction.
However, in the same field of endeavor, Shin teaches: or a horizontal direction (Fig. 2, matrix 202 is compressed horizontally).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to horizontally compress a weight tensor as disclosed by Shin in the method disclosed by Chinya to improve reduce unnecessary computation and save on memory (¶79, with a sparse matrix, less computational operations occur while also using less memory).
Regarding claim 6, Chinya fails to teach: The processor-implemented method of Claim 1, wherein: dimensions of the sparsity map match dimensions of the densified weight tensor; and the sparsity map indicates associations between elements in the densified weight tensor and corresponding output elements for the densified weight tensor during convolution.
However, in the same field of endeavor, Shin teaches: wherein: dimensions of the sparsity map match dimensions of the densified weight tensor; and the sparsity map indicates associations between elements in the densified weight tensor and corresponding output elements for the densified weight tensor during convolution (Fig. 2, compressed matrix 206 and sparsity map 208 have dimensions that match – ¶96, after a matrix multiplication of a sparse matrix in process 455, a processor receives an instruction to generate an expanded matrix… one or more circuits receive index values for non-zero values of a matrix multiplication result – the index values may be in a matrix like in Fig. 2 – and – ¶246, tensor cores are configured to perform deep learning matrix arithmetic, such as convolution operations for neural network training and inferencing).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to match dimensions of the sparsity map and densified tensor, and use the sparsity map to inform calculations including convolution as disclosed by Shin in the method disclosed by Chinya to improve computation speed (¶79, by implementing techniques described herein, an increase of processing speed also results in faster end-to-end training times and inferencing times).
Regarding claim 7, Chinya fails to explicitly teach: The processor-implemented method of Claim 6, wherein the sparsity map indicates the associations using, for each element in the densified weight tensor, an absolute indication of the corresponding output element.
However, in the same field of endeavor, Shin teaches: wherein the sparsity map indicates the associations using, for each element in the densified weight tensor, an absolute indication of the corresponding output element (¶73, In at least one embodiment, indices within metadata 208 are pointers to locations within original sparse matrix 202).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use the absolute indication of Shin in the method disclosed by Chinya to improve computation speed (¶79, by implementing techniques described herein, an increase of processing speed also results in faster end-to-end training times and inferencing times).
Regarding claim 10, Chinya teaches: A processor-implemented method of machine learning, comprising (¶55, the system 158 includes a host processor 160 (e.g., CPU with one or more processor cores)): accessing an activation tensor (¶41, The activation data and weight register file 454 may provide outputs to the multiplier block 466 and the summation block 468 to be multiplied, summed and/or accumulated)… comprising: retrieving a densified weight tensor for the… operation, wherein the densified weight tensor was generated by directionally squeezing a weight tensor, having unstructured sparsity, to remove zero-value elements (¶18, Zero values may be removed from the workload to compress the data of the workload 104 – see Fig. 3 for sparsity bitmap 360 and compressed data 354 – also, ¶16, The sparsity aware compression scheme may operate on unstructured sparsity data)… retrieving a sparsity map associated with the densified weight tensor (see Fig. 3 for sparsity bitmap 360 and compressed data 354); generating a set of intermediate tensors by multiplying the densified weight tensor and the activation tensor (Fig. 5, Multiplier Block 466 receives weights and activations); and generating an output tensor for the… operation by accumulating the set of intermediate tensors based on the sparsity map (Fig. 5, Summation Block 468 accumulates values from the Multiplier Block 466 – and – ¶41, The activation data and weight register file 454 may provide outputs to the multiplier block 466 and the summation block 468 to be multiplied, summed and/or accumulated. In some embodiments, a multiply and accumulate or a MAC may be a computation element of the PE 452. The summed value may be stored in the partial sum registers 458 for further processing. In some embodiments, the weight Sparsity Bitmap pointer may be identical in dimensions and functionality to the activation sparsity bitmap pointer counterpart).
Chinya fails to explicitly teach: for processing using a convolution operation; and performing the convolution operation using a machine learning model… convolution… wherein directionally squeezing the weight tensor comprises maintaining non-zero elements of the weight tensor… convolution… and generating an inference using the machine learning model based on the output tensor.
However, in the same field of endeavor, Shin teaches: for processing using a convolution operation; and performing the convolution operation using a machine learning model (¶47, In at least one embodiment, sparse matrix multiplication operations are performed as part of training or performing a neural network, convolution, or a machine learning operation)… convolution… wherein directionally squeezing the weight tensor comprises maintaining non-zero elements of the weight tensor (Fig. 2, squeezing the weight tensor results in non-zero values being maintained –see, e.g., highlighted portion of original matrix 202, where non-zero elements 1-32 are maintained in the compressed matrix 206. Also note the sparsity map produced 208)… convolution… and generating an inference using the machine learning model based on the output tensor (¶79, by implementing techniques described herein, an increase of processing speed also results in faster end-to-end training times and inferencing times).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to maintain non-zero elements of the weight tensor, perform a convolution operation and generate an inference as disclosed by Shin in the method disclosed by Chinya to improve reduce unnecessary computation and save on memory (¶79, with a sparse matrix, less computational operations occur while also using less memory).
Regarding claim 11, Chinya further teaches: The processor-implemented method of Claim 10, wherein the sparsity map indicates associations between elements in the densified weight tensor and corresponding output elements in the output tensor (¶43, processing block 482 identifies a decode operation 482. Illustrated processing block 484 identifies a lookahead window for a sparsity bitmap decode operation based on a current position in the bitmap. Illustrated processing block 486 determines if any of the sparsity bitmap values from the sparsity bitmap in the lookahead window are associated with a non-zero number. If not, illustrated processing block 488 simultaneously processes and loads activation values (e.g., weights) associated with the lookahead window and the current position. Illustrated processing block 494 determines if any values remain in the bitmap after the lookahead window. If so, processing block 496 sets the current position to a next position after lookahead window – i.e., a process for reversing sparsification, applied to activations).
Regarding claim 12, Chinya further teaches: The processor-implemented method of Claim 11, wherein the sparsity map indicates the associations using, for each element in the densified weight tensor, an absolute indication of the corresponding output element (Fig. 3, sparsity map 360 contains the explicit location of each non-zero and zero element).
Regarding claim 13, Chinya further teaches: The processor-implemented method of Claim 11, wherein the sparsity map indicates the associations using, for each respective element in the densified weight tensor, a relative spacing indicating a number of output elements to skip between the respective element and a subsequent element in the output tensor (¶33, As an example, if PE.sub.0 holds 16 weight points in 8-bit uncompressed hex format represented as [00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 2a, 00, 04, 0a], a compressed equivalent sparsity representation (which is referred to as the sparsity bitmap) would be [00001011] and [00000000] for byte 0 358 and byte 1 – in byte zero, having a pattern like 1011 indicates skipping the second element).
Regarding claim 14, Chinya further teaches: The processor-implemented method of Claim 10, wherein generating the output tensor comprises, for each respective output element in the output tensor: identifying a respective set of elements, from the set of intermediate tensors, that correspond to the respective output element based on the sparsity map (¶54, The activation data within a PE may include a sparsity bitmap register 814 and activation register 812 to hold activation data bytes – and – ¶54, When the activate skip signal is equal to 1 (a zero value detected), the sparsity activation pointer may be incremented… when an activation weight identifier is “high” (illustrated in FIG. 5), activation data (illustrated in FIG. 5) may be written into an activation register file – i.e., skipping over zero activations identified by the bitmap and recording non-zero activations); and accumulating the respective set of elements (¶2, Zero valued weights may not contribute towards partial operations during the training (e.g., sum accumulation during multiply-and-accumulate operation in convolution)).
Regarding claim 15, Chinya further teaches: The processor-implemented method of Claim 10, wherein accumulating the set of intermediate tensors based on the sparsity map is performed in response to determining that a sparsity indicator associated with the convolution operation indicates that at least one densified weight tensor is used in the convolution operation (¶36, the MUX control signal denotes a sparsity bitmap byte. When the count of the byte counter 406 is equal to or above the maximum value… the MUX control may denote a weight data byte – the MUX control signal is used to differentiate between sparsity bitmap bytes and weight values).
Regarding claim 16, it is a system implementing the method of claim 1 and is rejected on the same grounds – see above.
Regarding claims 19, 20, 21 and 22, they recite similar limitations as claims 4, 5, 6 and 7 and are rejected on the same grounds – see above.
Regarding claim 25, it is a system implementing the method of claim 10 and is rejected on the same grounds – see above.
Regarding claims 26-30, they recite similar limitations as claims 11-15 and are rejected on the same grounds – see above.
Claim(s) 2, 3, 17, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Chinya in view of Shin as applied to claims 1 and 16 above, and further in view of Tan et al. (US 20230100930 A1), herein Tan.
Regarding claim 2, Chinya in view of Shin fails to explicitly teach: The processor-implemented method of Claim 1, further comprising compiling at least a portion of the machine learning model based at least in part on the densified weight tensor.
However, in the same field of endeavor, Tan teaches: further comprising compiling at least a portion of the machine learning model based at least in part on the densified weight tensor (¶33, Accordingly, compressed weight tensor 152 can be used to implement a sparse neural network model having at least 75% zero weight values – and – ¶84, any of processes 1000, 1100, 1200, and 1300, or a portion thereof, can be performed, for example, by a compiler that interprets programming code describing a neural network model, and translates the programming code into machine instructions for execution on hardware).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to compile the model based on a densified tensor as disclosed by Tan in the method disclosed by Chinya in view of Shin to enable the model to be run (¶84, translates the programming code into machine instructions for execution on hardware).
Regarding claim 3, Chinya further teaches: The processor-implemented method of Claim 2, wherein compiling the at least the portion of the machine learning model comprises setting a sparsity indicator to indicate that the at least the compiled portion of the machine learning model includes the densified weight tensor (¶36, the MUX control signal denotes a sparsity bitmap byte).
Regarding claims 17 and 18, they recite similar limitations to claims 2 and 3, respectively, and are rejected on the same grounds.
Claim(s) 8 and 23 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chinya in view of Shin as applied to claims 6 and 21 above, and further in view of Erdeljan et al. (“IP Core for Efficient Zero-Run Length Compression of CNN Feature Maps”, 2017), herein Erdeljan.
Regarding claim 8, Chinya in view of Shin fails to teach: The processor-implemented method of Claim 6, wherein the sparsity map indicates the associations using, for each respective element in the densified weight tensor, a relative spacing indicating a number of output elements to skip between the respective element and a subsequent element in the densified weight tensor.
However, in the same field of endeavor, Erdeljan teaches: wherein the sparsity map indicates the associations using, for each respective element in the densified weight tensor, a relative spacing indicating a number of output elements to skip between the respective element and a subsequent element in the densified weight tensor (pg. 2, Section II, B. Zero interval, The ZI vector, on the other hand, is based on encoding relative positions of non-zero values… Each non-zero value has a corresponding zero-interval value. The ZI values are determined by the number of zero values seen before the non-zero one… counts the distance between the two non-zero values – also see the example in Fig. 2).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use a sparsity map that encodes zero-value skip lengths between non-zero values as disclosed by Erdeljan in the method disclosed by Chinya in view of Shin to provide an alternative compression method that performs better under certain sparsity conditions (pg. 2, Section II, C. Criteria for choosing an algorithm, In case when compression ratio is small, NZIV offers a better solution, while ZI is superior in case when compression ratio is high).
Regarding claim 23, it recites similar limitations to claim 8 and is rejected on the same grounds – see above.
Claim(s) 9 and 24 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chinya in view of Shin as applied to claim 1 and 16 above, and further in view of Abuhatzera et al. (US 20200320375 A1), herein Abuhatzera.
Regarding claim 9, Chinya fails to teach: The processor-implemented method of Claim 1, wherein generating the densified weight tensor is performed in response to determining that the unstructured sparsity in the weight tensor satisfies one or more defined thresholds.
However, in the same field of endeavor, Abuhatzera teaches: wherein generating the densified weight tensor is performed in response to determining that the unstructured sparsity in the weight tensor satisfies one or more defined thresholds (¶45, In some implementations, a combination of the sparsity component 150 and/or the compression circuit 155 may detect sparsity involved in multiply-accumulate operations and/or apply conditional masked-based block-compression techniques… if an expected or actual sparsity satisfies (e.g., is above) a threshold).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to generate a compressed weight tensor when sparsity exceeds a threshold as disclosed by Abuhatzera in the method disclosed by Chinya in view of Shin to reduce storage and calculation requirements (¶45, for data storage and movement savings).
Regarding claim 24, it recites similar limitations to claim 9 and is rejected on the same grounds – see above.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HARRISON CHAN YOUNG KIM whose telephone number is (571)272-0713. The examiner can normally be reached Monday - Thursday 9:00 am - 5:00 pm.
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, Cesar Paula can be reached at (571) 272-4128. 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.
/HARRISON C KIM/ Examiner, Art Unit 2145
/CESAR B PAULA/ Supervisory Patent Examiner, Art Unit 2145