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 office action is in response to submission of application on 2/28/2024
Claims 1-20 are presented for examination.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 10 and 20 are objected to because of the following informalities:
Claim 10 and 20 recite the limitation of “compare each outcome metric to a depth value for the potential outcome.” There is insufficient antecedent basis for this limitation in the claims because there are multiple “potential outcomes” such that it’s unclear which is “the potential outcome”
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 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1: Is the claim to a process, machine, manufacture, or composition of matter?
Claims 1-10 are directed to a machine and claims 11-20 are directed to a method; therefore, all claims are directed to one of the four statutory categories.
Step 2A Prong One: Does the claim recite an abstract idea, law of nature, or natural
phenomenon?
Claim 1 recites limitations of:
control the bank of processing elements to accumulate the result vector with an outcome vector for each potential outcome of the decision tree to obtain a respective outcome metric for each potential outcome; - mathematical concept (relationships, formulas or equations, calculations) of accumulating the result vector with an outcome vector
and select one potential outcome as the determined outcome of the decision tree for the input vector based on the respective outcome metrics for each potential outcome. – mental process (observation, evaluation, judgement) as a human mind can select a potential outcome
Step 2A Prong Two: Does the claim recite additional elements that integrate the judicial
exception into a practical application?
Claim 1 recites additional elements of:
A computing device comprising: a bank of processing elements; a controller interconnected with the bank of processing elements, the controller configured to: – components recited at a high level are construed as generic computer components used to implement the abstract idea. See MPEP 2106.05(f)(2).
obtain an input vector having a plurality of input attributes, the input vector to be processed by a decision tree to identify a determined outcome for the input vector; - obtaining an input vector merely amounts to data gathering which is insignificant extra-solution activity. See MPEP § 2106.05(g), item (3), which identifies necessary data gathering and outputting as an example of extra-solution activity.
control the bank of processing elements to process the input vector to obtain a result vector, wherein each input attribute is processed by one of the processing elements in the bank to obtain a result, and wherein the result vector comprises a combination of the results; - “control the bank of processing elements to process the input vector”: components recited at a high level are construed as generic computer components used to implement the abstract idea. See MPEP 2106.05(f)(2). And “to obtain a result vector, wherein each input attribute is processed by one of the processing elements in the bank to obtain a result, and wherein the result vector comprises a combination of the results;”: obtaining a result vector merely amounts to data gathering which is insignificant extra-solution activity. See MPEP § 2106.05(g), item (3), which identifies necessary data gathering and outputting as an example of extra-solution activity.
The additional elements do not integrate the abstract idea into a practical application.
Step 2B: Does the claim recite additional elements that amount to significantly more than the
judicial exception?
The additional elements are:
A computing device comprising: a bank of processing elements; a controller interconnected with the bank of processing elements, the controller configured to: – components recited at a high level are construed as generic computer components used to implement the abstract idea. See MPEP 2106.05(f)(2).
obtain an input vector having a plurality of input attributes, the input vector to be processed by a decision tree to identify a determined outcome for the input vector; - obtaining an input vector merely amounts to data gathering which is insignificant extra-solution activity. See MPEP § 2106.05(g). Data gathering is well-understood, routine, and conventional. See MPEP 2106.05(d)(II)(iv).
control the bank of processing elements to process the input vector to obtain a result vector, wherein each input attribute is processed by one of the processing elements in the bank to obtain a result, and wherein the result vector comprises a combination of the results; - “control the bank of processing elements to process the input vector”: components recited at a high level are construed as generic computer components used to implement the abstract idea. See MPEP 2106.05(f)(2). And “to obtain a result vector, wherein each input attribute is processed by one of the processing elements in the bank to obtain a result, and wherein the result vector comprises a combination of the results;”: obtaining a result vector merely amounts to data gathering which is insignificant extra-solution activity. See MPEP § 2106.05(g). Data gathering is well-understood, routine, and conventional. See MPEP 2106.05(d)(II)(iv).
The additional elements do not amount to significantly more than the abstract idea. Therefore,
the claim is not patent eligible.
Independent claim 11 recites the same relevant limitations and a similar analysis applies. Claim 11 does not recite any additional elements as it is a method claim. Therefore, it does not integrate the abstract idea into a practical application. Nor do they amount to significantly more. Therefore, the independent claims are not patent eligible.
The above analysis similarly applies to the dependent claims.
Dependent claim 2 and 12 recites, wherein the decision tree comprises a plurality of nodes, each node configured to process a given input attribute of the input vector. - this element constitutes
“mere instructions to apply an exception.” (MPEP § 2106.05(f)).
Dependent claim 3 and 13 recites, assign each node of the decision tree to one of the processing elements to process the given input attribute to obtain the result. – this element constitutes “mere instructions to apply an exception.” (MPEP § 2106.05(f)).
Dependent claim 4 and 14 recites, wherein to process the input attribute, the processing element is configured to compare the input attribute to a predefined threshold for the input attribute. – mental process (observation, evaluation, judgement) as a human mind can compare the input attributes to a threshold
Dependent claim 5 and 15 recites, initialize the bank of processing elements to store the predefined threshold for each input attribute in a respective corresponding memory cell of the processing element. – storing the predefined threshold merely amounts to storing and retrieving data in memory which is insignificant extra-solution activity. See MPEP § 2106.05(g). Storing and retrieving data in memory is well-understood, routine, and conventional. See MPEP 2106.05(d)(II)(iv).
Dependent claim 6 and 16 recites, assign each respective outcome metric to be accumulated by one of the processing elements in the bank. - this element constitutes “mere instructions to apply an exception.” (MPEP § 2106.05(f)).
Dependent claim 7 and 17 recites, wherein the respective outcome metric comprises a dot product between the result vector and the respective outcome vector. - mathematical concept (relationships, formulas or equations, calculations) of using a dot product
Dependent claim 8 and 18 recites, apply a generalized matrix-vector multiply between the result vector and an outcome matrix comprising the outcome vectors to accumulate the respective outcome metrics. - mathematical concept (relationships, formulas or equations, calculations) of using a generalized matric-vector multiply between the result vector and outcome matrix
Dependent claim 9 and 19 recites, initialize the bank of processing elements to load the outcome matrix into memory cells associated with the processing elements. - this element constitutes “mere instructions to apply an exception.” (MPEP § 2106.05(f)).
Dependent claim 10 and 20 recites,
compare each outcome metric to a depth value for the potential outcome; - mental process (observation, evaluation, judgement) as a human mind can compare outcome metrics
and normalize the outcome metrics; - mathematical concept (relationships, formulas or equations, calculations) to use normalization on outcome metrics
multiply each normalized outcome metric by a respective outcome identifier; - mathematical concept (relationships, formulas or equations, calculations) of using multiplication
and return the outcome identifier identifying the determined outcome. - this element constitutes “mere instructions to apply an exception.” (MPEP § 2106.05(f)).
The dependent claims do not integrate the abstract idea into a practical application, nor do they amount to significantly more than the abstract idea.
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
The following are the references being used:
Nakandala et al., (A Tensor Compiler for Unified Machine Learning Prediction Serving, herein
Nakandala)
Claims 1-4, 6-14, and 16-20 are rejected under 35 U.S.C. 102 (a)(1) as being anticipated by Nakandala.
Regarding claim 1,
Nakandala teaches,
A computing device comprising: a bank of processing elements; (Nakandala, page 8, section 6, “an Intel Xeon CPU E5-2690 v4 @2.6GHz(6 virtual cores), and an NVIDIA P100 GPU.” Page 9, section 6.1.1, “we use all six cores in the machine, while for request/response experiments we use one core.”, page 11, section 6.1.1, “we also were able to run HB on the new Graph core IPU[15] over a single decision tree.”, note: one machine holds many compute units. Nakandala switches between using all six cores and using one core, which shows the cores are separately usable units rather than on indivisible processor.)
a controller interconnected with the bank of processing elements, the controller configured to: (Nakandala, page 4, section 3.2, “the Optimizer extracts the parameters of each operator via the referenced extractor function and stores them in the container… The generated module is then exported into the target runtime format” section 3.3, “HB is currently limited by single GPU memory execution.” And page 14, section 6.3, “data movements between CPU and GPU memory.”, note: Nakandala pulls the tree’s parameters out of the trained model and stored them, then builds a module of tensor operations from those stored parameters. At run time the host CPU works through that module, handing each operation to the compute units and holding the intermediate result in between.)
obtain an input vector having a plurality of input attributes, the input vector to be processed by a decision tree to identify a determined outcome for the input vector; (Page 4-5, section 4.1, “we assume all decision nodes perform < comparisons… Given these tensors, Algorithm 1 presents how we perform tree scoring for a batch of input records X...
PNG
media_image1.png
70
412
media_image1.png
Greyscale
”, note: X holds n records, each with F feature values. One row of X is one input vector and its values are the input attributes. The algorithm’s stated output is a predicted class label for each row, and Nakandala states that what Algorithm 1 performs is tree scoring. That label is the determined outcome. A single row is a single input vector.)
control the bank of processing elements to process the input vector to obtain a result vector, wherein each input attribute is processed by one of the processing elements in the bank to obtain a result, and wherein the result vector comprises a combination of the results; (Nakandala, page 5, section 4.1, “The first GEMM is used to match each input feature with the internal node(s) using it. The following < operations is used to evaluate all the internal decision nodes and produces a tensor of 0s and 1s based on the false/ true outcome of the conditions…
PNG
media_image2.png
134
438
media_image2.png
Greyscale
”, Table 3, “
PNG
media_image3.png
80
166
media_image3.png
Greyscale
” note: Every question in the tree asks about one particular feature. Table A is a lookup that says which question asks about which feature. The first multiple uses A to deliver each of the input’s values to the question that asks about it. Then Table B holds each question’s cutoff number, and the < B step checks every value against its cutoff, giving a yes or no for every question in the tree. Each yes or no is a result. They are all in T which is the result vector.)
control the bank of processing elements to accumulate the result vector with an outcome vector for each potential outcome of the decision tree to obtain a respective outcome metric for each potential outcome; (Nakandala, section 4.1, Table 3, “
PNG
media_image4.png
66
348
media_image4.png
Greyscale
” Algorithm 1, “
PNG
media_image5.png
166
420
media_image5.png
Greyscale
… The second GEMM operation generates an encoding for the path composed by the true internal nodes,” note: Each leaf of the tree is one answer the tree could give, one potential outcome. Table C has one column for every leaf. That column is an instruction for reaching that leaf, for each question in the tree it says +1 if you need a yes to get there, -1 if you need a no, and 0 if that question is not on the way at all. That column is the outcome vector for that leaf. Now multiple the yes or no answers against every column at once. For each leaf, there is one number that says how well the actual answers lined up with that leaf’s instruction. Those numbers are the outcome metrics. Adding and multiplying is accumulating.)
and select one potential outcome as the determined outcome of the decision tree for the input vector based on the respective outcome metrics for each potential outcome. (Nakandala, section 4.1, Algorithm 1, “
PNG
media_image6.png
198
434
media_image6.png
Greyscale
… the successive == operation returns the leaf node selected by the encoded path. Finally, the third GEMM operation maps the selected leaf node to the class label.”, node: Each leaf now has a score. Compare each score against the number in D for that leaf. Exactly one leaf will match, that is the leaf he input would have landed on if the tree is navigated the normal way. Picking that one leaf out of all of them, using the scores maps to the claims.
Regarding claim 2,
The computing device of claim 1, wherein the decision tree comprises a plurality of nodes, each node configured to process a given input attribute of the input vector. (Nakandala, section 4.1, “
PNG
media_image7.png
222
538
media_image7.png
Greyscale
… A captures the relationship between input features and internal nodes.”, note: Nakandala’s tree is a list of nodes. Table A says which feature each node looks at. So the tree has many nodes, and each one handles a particular input attribute.)
Regarding claim 3,
The computing device of claim 2, wherein the controller is further configured to: assign each node of the decision tree to one of the processing elements to process the given input attribute to obtain the result. (Nakandala, section 4.1, Algorithm 1, “
PNG
media_image8.png
236
406
media_image8.png
Greyscale
”, Table 3, “
PNG
media_image9.png
36
202
media_image9.png
Greyscale
… The first GEMM is used to match each input feature with the internal node(s) using it.”, note: A has one column per question node, marking which feature that node asks about. The annotation shows the output is I wide, one slop per node, so slow 3 receives the value of the feature node 3 asks about. Then < B compares slot 3 against node 3’s own threshold. So slow 3 does node 3’s entire job, gets its feature, applies its cutoff, produces its result. The host fixes that pairing when it build A and B, before anything runs.)
Regarding claim 4,
The computing device of claim 1, wherein to process the input attribute, the processing element is configured to compare the input attribute to a predefined threshold for the input attribute (Nakandala, page 5, section 4.1, “
PNG
media_image8.png
236
406
media_image8.png
Greyscale
”, Table 3, “
PNG
media_image10.png
22
164
media_image10.png
Greyscale
… “B is set to the threshold value of each internal node,” and section 3.2, “extract the parameters of each operator (e.g., weights of a linear regression, thresholds of a decision tree)” note: The processing is a less-than check against B, and B is nothing but a list of cutoff numbers, one per question. Those cutoffs are predefined because Nakandala pulls them out of a model that was already trained before any of it runs.)
Regarding claim 6,
The computing device of claim 1, wherein the controller is configured to assign each respective outcome metric to be accumulated by one of the processing elements in the bank. (Nakandala, section 4.1, Algorithm 1, “
PNG
media_image8.png
236
406
media_image8.png
Greyscale
”, table 3, “
PNG
media_image11.png
46
74
media_image11.png
Greyscale
” note: C has one column per leaf and the output is L wide, one slop per leaf. Slot 2 is built by running down column 2, multiplying each entry by the matching yes or no answer and adding them up. Column 2 feeds nothing but slot 2, and slot 2 draws on nothing but column 2. That self-contained running total is the claimed accumulation at one element, and the host fixes the pairing when it builds C.)
Regarding claim 7,
The computing device of claim 1, wherein the respective outcome metric comprises a dot product between the result vector and the respective outcome vector. (Nakandala, page 5, section 4.1, “
PNG
media_image8.png
236
406
media_image8.png
Greyscale
” Table 3, “
PNG
media_image12.png
34
92
media_image12.png
Greyscale
… We cast the evaluation of a tree as a series of three Generic Matrix Multiplication (GEMM) operations interleaved by two element-wise logical operations.”, note: Multiplying tables together is doing dot products. Each number that comes out of GEMM (T, C) is the dot product of the yes or no answers with one column of C, and each column is one leaf’s outcome vector. So every outcome metric is a dot product.)
Regarding claim 8,
The computing device of claim 7, wherein the controller is configured to apply a generalized matrix-vector multiply between the result vector and an outcome matrix comprising the outcome vectors to accumulate the respective outcome metrics. (Nakandala, page 5, section 4.1, “
PNG
media_image8.png
236
406
media_image8.png
Greyscale
” Table 3, “
PNG
media_image12.png
34
92
media_image12.png
Greyscale
… a series of three Generic Matrix Multiplication (GEMM) operations interleaved by two element-wise logical operations” note: C is a table whose columns are the outcome vectors, that is the outcome matric the claim asks for. Nakandala gets the metrics by multiplying against it and calls the operation a generic matrix multiply by name. When only one record is being scored, one side of that multiply is a single vector, which makes it a matrix-vector multiply)
Regarding claim 9,
The computing device of claim 8, wherein the controller is configured to initialize the bank of processing elements to load the outcome matrix into memory cells associated with the processing elements. (Nakandala, page 4, section 3.2, “The generated module is then exported into the target runtime format.”, section 3.3, “HB is currently limited by single GPU memory execution”, section 6.3, “data movements between CPU and GPU memory.”, note: The multiply can’t run at all unless C is somewhere the compute units can read, a precondition. What caps the system is how much fits in GPU memory, which only limits anything if the model’s tables are sitting there. The module built beforehand is the initialize.)
Regarding claim 10,
The computing device of claim 1, wherein to select the determined outcome, the controller is configured to:
compare each outcome metric to a depth value for the potential outcome; (Nakandala, page 5, section 4.1, “
PNG
media_image8.png
236
406
media_image8.png
Greyscale
”, Table 3, “
PNG
media_image13.png
44
268
media_image13.png
Greyscale
… D captures the count of the internal nodes in the path from a leaf node to the tree root, for which the internal node is the left child of its parent.”, note: Every leaf comes out of claim 1 with a score. D holds one number per leaf, obtained by walking that leaf up to the root and counting nodes along the way. == D tests each score against its own leaf’s number, and exactly one leaf matches)
and normalize the outcome metrics; (Nakandala, page 5, section 4.1, “the successive == operation returns the leaf node selected by the encoded path.”, Algorithm 1, “
PNG
media_image8.png
236
406
media_image8.png
Greyscale
”, Table 3, “
PNG
media_image14.png
48
140
media_image14.png
Greyscale
” note: == D is an equality test, so it displays a true/false per leaf, one leak marked, the rest not, rather than carrying forward the original spread of scores. The operation returns the leaf node selected, a marker of which leaf won. That the marks are binary follows from the algorithm’s own declared output. R is annotated {0,1} and E is 0/1 indicator matrix, so the values entering GEMM(T,E) must themselves by 0/1 for R to come out that way. Scores that arrived on an arbitrary numeric range how all sit on the same two-point scale, which is normalization.
multiply each normalized outcome metric by a respective outcome identifier; (Nakandala, page 5, section 4.1, “For regression tasks, we initialize E with label values.”, Algorithm 1, “
PNG
media_image8.png
236
406
media_image8.png
Greyscale
”, Table 3, “
PNG
media_image14.png
48
140
media_image14.png
Greyscale
” note: E holds one row per leaf recording what the leaf returns, so each leaf’s mark is multiplied by its own row, the winner is kept and everything else zeroes out. In the regression variant E is load with label values, so the operation is metric x identifier)
and return the outcome identifier identifying the determined outcome. (Nakandala, page 5, section 4.1, “Finally, the third GEMM operation maps the selected leaf node to the class label.”, and Algorithm 1, “
PNG
media_image8.png
236
406
media_image8.png
Greyscale
”, note: One nonzero value survives the multiply. R is what the algorithm hands back, and it holds the class label of the leaf that was selected.)
Claims 11-14, and 16-20 is a method claim, that corresponds to machine claim 1-4, and 6 -10, respectively. Otherwise, they are not patentably distinguishable. Therefore, claims 11-14, and 16-20 are rejected for the same reasons as claims 1-4, and 6 -10 respectively.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The following are the references used:
Pedretti et al., (Tree-based machine learning performed in-memory with memristive analog CAM, herein Pedretti)
Claim(s) 5 and 15 is/are rejected under 35 U.S.C. 103 as being unpatentable over
Nakandala in view of Pedretti.
Regarding claim 5,
Pedretti teaches,
The computing device of claim 4, wherein the controller is further configured to initialize the bank of processing elements to store the predefined threshold for each input attribute in a respective corresponding memory cell of the processing element. (Pedretti, page 4, “DT can be mapped to analog CAM by directly programming each root-to-leaf path into an array row. Feature vectors f are given as input to the columns DL,” page 3, “The lower and the higher bound of the searching range is stored as conductance in the RRAM device in our analog CAM.”, page 5, “In the case of only one threshold decision for a particular feature, one of the memristors is kept as wildcard (LRS or HRS) and the other is programmed at an intermediate threshold value, implementing a ’less than or equal to’ with a high threshold (a left branch), while a greater than is programmed in the opposite case (a right branch).”, note: Pedretti’s array is programmed with the tree’s thresholds before any input arrives, which is the recited initialization, and because feature values are fed to the columns, each column belongs to one feature and its cell therefore holds that feature’s threshold which maps to the claims for each input attribute. The threshold in Pedretti is not stored somewhere and retrieved, it is stored as the cells own conductance, and that conductance is physically what determines whether an arriving value falls above or below it, so the storage and the comparison are the same device, which maps to the memory cell of the processing element.)
It would be obvious to one of ordinary skill in the art before the effective filing date of the
claimed invention to combine the teaching of Nakandala and Pedretti because Nakandala keeps its threshold in general memory separate from the units that compare against them, while Pedretti teaches storing each threshold inside the element that uses it. Pedretti explains that separating storage from computing is what costs time and energy, and reports that keeping the values in place makes the system faster and lower energy per decision.
Claims 15 is a method claim, that corresponds to machine claim 5, respectively. Otherwise, they are not patentably distinguishable. Therefore, claims 15 are rejected for the same reasons as claim 5 respectively.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to UYEN-NHU PHAM TRAN whose telephone number is (571)272-1559. The examiner can normally be reached Monday - Friday 7:30-5.
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, Miranda Huang can be reached at (571) 270-7092. 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.
/U.P.T./Examiner, Art Unit 2124
/MIRANDA M HUANG/Supervisory Patent Examiner, Art Unit 2124
Prosecution Insights