DETAILED ACTION
This action is in response to the application filed 02/21/2024. Claims 1-26 are pending and have been examined.
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 .
Information Disclosure Statement
The information disclosure statements (IDS) submitted on 08/06/2024 and 04/09/2026 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
Specification
The disclosure is objected to because of the following informalities:
The mathematical formulas specified in paragraphs [0074], [0113], [0116], [0183], [0226], [0229], [0234], and [0262] are fuzzy and difficult to parse.
Appropriate correction is required.
Claim Objections
Claim 1 is objected to because of the following informalities:
“the processor being configured to: … specifying processing … and rule generation processing” is improper grammar.
Appropriate correction is required. This above issues of claim 1 manifest in the preambles of dependent claims 2-11.
Claim 6 is objected to because of the following informalities:
“in a case where the selected combination of the feature vectors is swapped is lower than a predetermined fourth threshold value” is improper grammar.
Appropriate correction is required.
Claim 7 is objected to because of the following informalities:
“in a case where the selected combination of the feature vectors is swapped is equal to or higher than a predetermined fourth similarity.” is improper grammar.
Appropriate correction is required.
Claim 9 is objected to because of the following informalities:
“wherein the processor is further configured to … reception processing of receiving” is improper grammar.
Appropriate correction is required.
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.
Claims 14-24 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 14 recites “merge, in the training processing, the second feature vectors output from the merging layer by changing a conversion rule from the first feature vectors to the second feature vectors
in the merging layer.” in its sixth limitation. This seems to specify details about the merging layer process, wherein first feature vectors are converted to second feature vectors, as recited in limitation 3. However, it refers to training processing that’s executed in response to the input of the second feature vector, as recited in limitation 5. This circular dependency renders the scope of the claim indefinite. This deficiency is inherited by dependent claims 15-24. “merge, in the training processing” is interpreted as referring to the process of the merging layer that may precede or coincide with “training processing” that depends on the second feature vector.
Claim 18 recites “a combination of the similar embedding vectors” in its second limitation. There is a lack of antecedent basis for “the similar embedding vectors”. Thus, the scope of the claim is rendered indefinite.
Claim 6 recites “select a combination of the feature vectors from the data set, and specify, as the combination of the feature vectors that are allowed to be merged, the selected combination of the feature vectors in a case where a change value of a prediction result of the provisional model in a case where the selected combination of the feature vectors is swapped is lower than a predetermined fourth threshold value” in its second limitation. “in a case where a change value of a prediction result of the provisional learning model” is improper English, as ‘where’ in this context lacks a second conjunct to link with the change value. In other words, the sentence is structured as to attach some kind of condition or description to the change value, but none is provided. This renders the meaning of the limitation unclear, and thus renders the scope of the claim indefinite.
Regarding “in a case where the selected combination of the feature vectors is swapped is lower than a predetermined fourth threshold value”, it’s unclear whether “in a case where the combination of the feature vectors is swapped” is meant to apply to the merging combinations of the second feature vectors, the change value of a prediction result, or both. It’s also unclear what, exactly is “lower than a predetermined fourth threshold value”, be it the change value, the combination itself, or something else. Thus, the scope is further rendered indefinite.
The second limitation of claim 6 is interpreted as follows: “select a combination of the feature vectors from the data set, and specify, as the combination of the feature vectors that are allowed to be merged, the selected combination of the feature vectors in a case where a value of a prediction result of the provisional model changes, and where the swapped selected combination of the feature vectors is lower than a predetermined fourth threshold value”
Claim 7 recites “select a combination of the feature vectors from the data set, and specify, as the combination of the feature vectors that are allowed to be merged, the selected combination of the feature vectors in a case where a similarity in a prediction result of the provisional model in a case where the selected combination of the feature vectors is swapped is equal to or higher than a predetermined fourth similarity” in its second limitation. “in a case where a similarity in a prediction result of the provisional model” is improper English, as ‘where’ in this context lacks a second conjunct to link with the similarity. In other words, the sentence is structured as to attach some kind of condition or description to the similarity, but none is provided. This renders the meaning of the limitation unclear, and thus renders the scope of the claim indefinite.
Regarding “in a case where the selected combination of the feature vectors is swapped is equal to or higher than a predetermined fourth similarity”, it’s unclear whether “in a case where the selected combination of the feature vectors is swapped” is meant to apply to the merging combinations of the second feature vectors, the change value of a prediction result, or both. It’s also unclear what, exactly is “higher than a predetermined fourth similarity”, be it the similarity, the combination itself, or something else. Thus, the scope is further rendered indefinite.
The second limitation of claim 7 is interpreted as follows: “select a combination of the feature vectors from the data set, and specify, as the combination of the feature vectors that are allowed to be merged, the selected combination of the feature vectors in a case where prediction results of the provisional model are similar, and where the swapped selected combination of the feature vectors is equal to or higher than a predetermined fourth similarity”.
Claim 23 recites “merge combinations of the second feature vectors that correspond to combinations of the embedding vectors in a case where a change value of a prediction result of the machine learning model in a case where the combination of the embedding vectors is swapped is lower than a predetermined seventh threshold value” in its first limitation. “in a case where a change value of a prediction result of the machine learning model” is improper English, as ‘where’ in this context lacks a second conjunct to link with the change value. In other words, the sentence is structured as to attach some kind of condition or description to the change value, but none is provided. This renders the meaning of the limitation unclear, and thus renders the scope of the claim indefinite.
Regarding “in a case where the combination of the embedding vectors is swapped is lower than a predetermined seventh threshold value”, it’s unclear whether “in a case where the combination of the embedding vectors is swapped” is meant to apply to the merging combinations of the second feature vectors, the change value of a prediction result, or both. It’s also unclear what, exactly is “lower than a predetermined seventh threshold value”, be it the change value, the combination itself, or something else. Thus, the scope is further rendered indefinite.
The claim is interpreted as follows: “The learning device according to claim 18, wherein the processor is configured to, in the training processing, merge combinations of the second feature vectors that correspond to combinations of the embedding vectors, in a case where a value of a prediction result of the machine learning model changes, and where the swapped combination of the embedding vectors is lower than a predetermined seventh threshold value”.
Claim 24 recites “merge combinations of the second feature vectors that correspond to combinations of the embedding vectors in a case where a similarity of a prediction result of the machine learning model in a case where the combination of the embedding vectors is swapped is equal to or higher than a predetermined fifth threshold value” in its first limitation. “in a case where a similarity of a prediction result of the machine learning model” is improper English, as ‘where’ in this context lacks a second conjunct to link with the similarity. In other words, the sentence is structured as to attach some kind of condition or description to the similarity, but none is provided. This renders the meaning of the limitation unclear, and thus renders the scope of the claim indefinite.
Regarding “in a case where the combination of the embedding vectors is swapped is equal to or higher than a predetermined fifth threshold value”, it’s unclear whether “in a case where the combination of the embedding vectors is swapped” is meant to apply to the merging combinations of the second feature vectors, the change value of a prediction result, or both. It’s also unclear what, exactly is “higher than a predetermined fifth threshold value”, be it the similarity, the combination itself, or something else. Thus, the scope is further rendered indefinite.
The claim is interpreted as follows: “The learning device according to claim 18, wherein the processor is configured to, in the training processing, merge combinations of the second feature vectors that correspond to combinations of the embedding vectors, in a case where prediction results of the machine learning model are similar, and where the swapped combination of the embedding vectors is higher than a predetermined fifth threshold value”.
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 1-13 are rejected under 35 U.S.C. 101 because the claimed inventions are directed to non-statutory subject matter without significantly more.
Claim 1
Step 1: The claim recites “A device”, and is therefore directed to the statutory category of machine
Step 2A Prong 1: The claim recites the following judicial exception(s)
specifying processing of specifying a combination of feature vectors that are included in a data set including a correct answer label and are allowed to be merged: This can be performed as a mental process. One can mentally decide on a desired combination of features that should be allowed to merge.
rule generation processing of generating a merging rule of the feature vectors based on a combination of the feature vectors that are allowed to be merged: This can be performed as a mental process. One can mentally decide on a merging rule for the feature vectors.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the following additional element(s)
A device for generating a data merging rule for a machine learning model, the device comprising: a processor; and a memory connected to or built in the processor: This is mere instruction to execute the recited judicial exceptions with generic computer hardware (MPEP 2106.05(f)).
Step 2B: The following additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
A device for generating a data merging rule for a machine learning model, the device comprising: a processor; and a memory connected to or built in the processor: This is mere instruction to execute the recited judicial exceptions with generic computer hardware (MPEP 2106.05(f)).
Claim 2
Step 1: The claim recites a machine, as in claim 1
Step 2A Prong 1: The claim recites the following further judicial exception(s)
the processor is configured to create a frequency distribution of a correct answer label for each of the feature vectors included in the data set: This can be performed as a mental process. One can mentally determine the frequencies of correct answer labels.
specify a combination of the feature vectors in which a similarity in the frequency distribution of the correct answer label is equal to or higher than a predetermined first threshold value, as the combination of the feature vectors that are allowed to be merged: This can be performed as a mental process. One can mentally compare the frequency distribution and a similarity threshold value to decide on which combinations should be merged.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the further additional element(s)
the processor is configured to create a frequency distribution of a correct answer label for each of the feature vectors included in the data set: This is mere instruction to execute a recited judicial exception with generic computer hardware (MPEP 2106.05(f)).
Step 2B: The further additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
the processor is configured to create a frequency distribution of a correct answer label for each of the feature vectors included in the data set: This is mere instruction to execute a recited judicial exception with generic computer hardware (MPEP 2106.05(f)).
Claim 3
Step 1: The claim recites a machine, as in claim 2
Step 2A Prong 1: The claim recites the following further judicial exception(s)
the processor is configured to further create, for a combination specified as the combination of the feature vectors that are allowed to be merged, a frequency distribution in consideration of a combination of a plurality of items: This can be performed as a mental process. One can mentally determine frequencies of values associated with some combination(s) of features.
exclude the specified combination from the combinations of the feature vectors that are allowed to be merged in a case where a similarity in the frequency distribution in consideration of the combination of the items is lower than a predetermined second threshold value: This can be performed as a mental process. One can mentally compare the frequency distribution and a similarity threshold value to decide on which combinations should be merged.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the further additional element(s)
the processor is configured to further create, for a combination specified as the combination of the feature vectors that are allowed to be merged, a frequency distribution in consideration of a combination of a plurality of items: This is mere instruction to execute a recited judicial exception with generic computer hardware (MPEP 2106.05(f)).
Step 2B: The further additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
the processor is configured to further create, for a combination specified as the combination of the feature vectors that are allowed to be merged, a frequency distribution in consideration of a combination of a plurality of items: This is mere instruction to execute a recited judicial exception with generic computer hardware (MPEP 2106.05(f)).
Claim 4
Step 1: The claim recites a machine, as in claim 1
Step 2A Prong 1: The claim recites the following further judicial exception(s)
the processor is configured to create, for each of the feature vectors included in the data set, a frequency distribution of a correct answer label in consideration of a combination of a plurality of items: This can be performed as a mental process. One can mentally determine frequencies of label values associated with some combination(s) of features.
specify a combination of the feature vectors in which a similarity in the frequency distribution of the correct answer label is equal to or higher than a predetermined seventh threshold value, as the combination of the feature vectors that are allowed to be merged: This can be performed as a mental process. One can mentally compare the frequency distribution and a similarity threshold value to decide on which combinations should be merged.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the further additional element(s)
the processor is configured to create, for each of the feature vectors included in the data set, a frequency distribution of a correct answer label in consideration of a combination of a plurality of items: This is mere instruction to execute a recited judicial exception with generic computer hardware (MPEP 2106.05(f)).
Step 2B: The further additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
the processor is configured to create, for each of the feature vectors included in the data set, a frequency distribution of a correct answer label in consideration of a combination of a plurality of items: This is mere instruction to execute a recited judicial exception with generic computer hardware (MPEP 2106.05(f)).
Claim 5
Step 1: The claim recites a machine, as in claim 1
Step 2A Prong 1: The claim recites the following further judicial exception(s)
the processor is configured to end generation of the merging rule in a case where the number of combinations of the feature vectors that are included in the merging rule and are allowed to be merged is equal to or larger than a predetermined third threshold value: This can be performed as a mental process. One can merely stop thinking about merging rules when the number of combinations included exceed a threshold value.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the further additional element(s)
the processor is configured to end generation of the merging rule in a case where the number of combinations of the feature vectors that are included in the merging rule and are allowed to be merged is equal to or larger than a predetermined third threshold value
Step 2B: The further additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
the processor is configured to end generation of the merging rule in a case where the number of combinations of the feature vectors that are included in the merging rule and are allowed to be merged is equal to or larger than a predetermined third threshold value
Claim 6
Step 1: The claim recites a machine, as in claim 1
Step 2A Prong 1: The claim recites the following further judicial exception(s)
select a combination of the feature vectors from the data set, and specify, as the combination of the feature vectors that are allowed to be merged, the selected combination of the feature vectors in a case where a change value of a prediction result of the provisional model in a case where the selected combination of the feature vectors is swapped is lower than a predetermined fourth threshold value: This can be performed as a mental process. One can merely identify a feature vector combination in a case where it changes the prediction result as recited.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the further additional element(s)
the processor is configured to: This is mere instruction to execute the recited judicial exceptions with generic computer hardware (MPEP 2106.05(f)).
generate a provisional model in which the feature vectors included in the data set are used as inputs and train the provisional model: This is mere instruction to train a model to perform a judicial exception in a generic manner (MPEP 2106.05(f)).
Step 2B: The further additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
the processor is configured to: This is mere instruction to execute the recited judicial exceptions with generic computer hardware (MPEP 2106.05(f)).
generate a provisional model in which the feature vectors included in the data set are used as inputs and train the provisional model: This is mere instruction to train a model to perform a judicial exception in a generic manner (MPEP 2106.05(f)).
Claim 7
Step 1: The claim recites a machine, as in claim 1
Step 2A Prong 1: The claim recites the following further judicial exception(s)
select a combination of the feature vectors from the data set, and specify, as the combination of the feature vectors that are allowed to be merged, the selected combination of the feature vectors in a case where a similarity in a prediction result of the provisional model in a case where the selected combination of the feature vectors is swapped is equal to or higher than a predetermined fourth similarity: This can be performed as a mental process. One can merely identify a feature vector combination in a case where it changes the prediction result as recited.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the further additional element(s)
the processor is configured to: This is mere instruction to execute the recited judicial exceptions with generic computer hardware (MPEP 2106.05(f)).
generate a provisional model in which the feature vectors included in the data set are used as inputs and train the provisional model: This is mere instruction to train a model to perform a judicial exception in a generic manner (MPEP 2106.05(f)).
Step 2B: The further additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
the processor is configured to: This is mere instruction to execute the recited judicial exceptions with generic computer hardware (MPEP 2106.05(f)).
generate a provisional model in which the feature vectors included in the data set are used as inputs and train the provisional model: This is mere instruction to train a model to perform a judicial exception in a generic manner (MPEP 2106.05(f)).
Claim 8
Step 1: The claim recites a machine, as in claim 1
Step 2A Prong 1: The claim recites the following further judicial exception(s)
candidates of the feature vectors that are allowed to be merged are determined based on at least one of an edit distance, a distribution representation, or related information of the feature vectors: This can be performed as a mental process. One can mentally decide on which feature vectors should be merged based on information about their related information, edit distance, or distribution representations of the feature vectors.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the additional element(s)
Step 2B: The additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
Claim 9
Step 1: The claim recites a machine, as in claim 1
Step 2A Prong 1: The claim recites no further judicial exception(s)
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the further additional element(s)
display processing of displaying the combination of the feature vectors that are allowed to be merged on a display unit: This amounts to mere output and is insignificant extra-solution activity (MPEP 2106.05(g)).
reception processing of receiving, from a user, whether or not to merge the combination of the feature vectors that are allowed to be merged: This amounts to mere reception of data and is insignificant extra-solution activity (MPEP 2106.05(g)).
Step 2B: The further additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
display processing of displaying the combination of the feature vectors that are allowed to be merged on a display unit: This is an instance of transmitting data over a network, a limitation known to be well-understood, routine, and conventional (MPEP 2106.05(d) II. i.)
reception processing of receiving, from a user, whether or not to merge the combination of the feature vectors that are allowed to be merged: This is an instance of receiving data over a network, a limitation known to be well-understood, routine, and conventional (MPEP 2106.05(d) II. i.)
Claim 10
Step 1: The claim recites “A learning device”, and is therefore directed to the statutory category of machine
Step 2A Prong 1: The claim recites the recited judicial exception(s) of claim 1
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the following additional element(s)
A learning device that trains a machine learning model by using a training data set obtained by performing merging according to a merging rule generated by the device for generating a data merging rule according to claim 1: This is mere instruction to train a machine learning model with the recited judicial exceptions in a generic manner (MPEP 2106.05(f)).
Step 2B: The following additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
A learning device that trains a machine learning model by using a training data set obtained by performing merging according to a merging rule generated by the device for generating a data merging rule according to claim 1: This is mere instruction to train a machine learning model with the recited judicial exceptions in a generic manner (MPEP 2106.05(f)).
Claim 11
Step 1: The claim recites “A prediction device”, and is therefore directed to the statutory category of machine
Step 2A Prong 1: The claim recites the recited judicial exception(s) of claim 1
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the following additional element(s)
A prediction device that causes a machine learning model to perform prediction by using, as an input, data obtained by performing merging according to the merging rule generated by the device for generating a data merging rule according to claim 1: This is mere instruction to train a machine learning model with the recited judicial exceptions in a generic manner (MPEP 2106.05(f)).
Step 2B: The following additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
A prediction device that causes a machine learning model to perform prediction by using, as an input, data obtained by performing merging according to the merging rule generated by the device for generating a data merging rule according to claim 1: This is mere instruction to train a machine learning model with the recited judicial exceptions in a generic manner (MPEP 2106.05(f)).
Claim 12
Step 1: The claim recites “An operation method”, and is therefore directed to the statutory category of process
Step 2A Prong 1: The claim recites the following judicial exception(s)
a step of specifying a combination of feature vectors that are included in a data set including a correct answer label and are allowed to be merged: This can be performed as a mental process. One can mentally decide on a desired combination of features that should be allowed to merge.
a step of generating a merging rule of the feature vectors based on a combination of the feature vectors that are allowed to be merged: This can be performed as a mental process. One can mentally decide on a merging rule for the feature vectors.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the following additional element(s)
An operation method for a device for generating a data merging rule for a machine learning model: This is mere instruction to implement data merging for a machine learning model in a generic manner based on the recited judicial exceptions (MPEP 2106.05(f)).
Step 2B: The following additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
An operation method for a device for generating a data merging rule for a machine learning model: This is mere instruction to implement data merging for a machine learning model in a generic manner based on the recited judicial exceptions (MPEP 2106.05(f)).
Claim 13
Step 1: The claim recites “A non-transitory computer readable medium”, and is therefore directed to the statutory category of article of manufacture
Step 2A Prong 1: The claim recites the following judicial exception(s)
a step of specifying a combination of feature vectors that are included in a data set including a correct answer label and are allowed to be merged: This can be performed as a mental process. One can mentally decide on a desired combination of features that should be allowed to merge.
a step of generating a merging rule of the feature vectors based on a combination of the feature vectors that are allowed to be merged: This can be performed as a mental process. One can mentally decide on a merging rule for the feature vectors.
Step 2A Prong 2: The judicial exception(s) are not integrated into a practical application through the following additional element(s)
A non-transitory computer readable medium storing a program that generates a data merging rule for a machine learning model: This is mere instruction to implement data merging for a machine learning model in a generic manner based on the recited judicial exceptions (MPEP 2106.05(f)).
Step 2B: The following additional element(s) of the claim, taken alone or in combination, do not amount to significantly more than the recited judicial exception(s)
A non-transitory computer readable medium storing a program that generates a data merging rule for a machine learning model: This is mere instruction to implement data merging for a machine learning model in a generic manner based on the recited judicial exceptions (MPEP 2106.05(f)).
Claim Rejections - 35 USC § 102
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.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 1-5 and 10-13 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Kumar et al. (SCALABLE GENERATION OF MULTIDIMENSIONAL FEATURES FOR MACHINE LEARNING, filed 4/19/2016, US 11,295,229 B1), hereafter referred to as Kumar.
Regarding claim 1, Kumar discloses [a] device for generating a data merging rule for a machine learning model, the device comprising: a processor; and a memory connected to or built in the processor:
“Various embodiments of methods and apparatus for scalable generation of multidimensional features for machine learning data sets are described” (Kumar, column 2, Detailed Description)
“The terms "higher-order features" and "multidimensional features" may be used synonymously herein to refer to features formed by combining (merging) the values of two or more input variables (data)” (Kumar, column 4, paragraph 1)
“In the illustrated embodiment, computing device 9000 includes one or more processors 9010 coupled to a system memory 9020” (Kumar, column 22, paragraph 5)
[T]he processor being configured to: specifying processing of specifying a combination of feature vectors that are included in a data set including a correct answer label and are allowed to be merged; and rule generation processing of generating a merging rule of the feature vectors based on a combination of the feature vectors that are allowed to be merged:
“The data set may include numerous observation records, each including some number of input variables and a target variable (correct answer label)” (Kumar, column 5, paragraph 3)
“In many cases, discovering informative multidimensional features (combination of feature vectors) (e.g., higher-order features whose inclusion in a linear model's input feature set is likely to lead to higher-quality predictions) may involve obtaining the values of correlation metrics between the multidimensional features and the target variable. Computing such correlation metrics may in tum involve obtaining occurrence counts for each combination of (a) a value of the multidimensional feature 45 and (b) a value of the target variable. Such occurrence counts, associated with matched combinations of values of different variables/features, may also be referred to as cooccurrence counts” (Kumar, column 4, paragraph 3). The procedure by which informative multidimensional features are discovered comprises a set of merging rule[s].
Kumar relates to selectively training a machine learning model with combinations of features selected based on frequency distributions and is analogous to the claimed invention.
Regarding claim 2, the rejection of claim 1 is incorporated. Kumar further discloses a device, wherein, in the specifying processing, the processor is configured to
create a frequency distribution of a correct answer label for each of the feature vectors included in the data set:
“Example data set 202, which is to be used to train a model, comprises ten million observation records labeled OR1-OR10000000. Each observation record comprises four input variables A, B, C and D (feature vectors), and a target variable T. Some combinations of the input variables A, B, C and D may be highly correlated with the target variable, but which specific combinations are highly correlated may not be known at the 45 start of the feature exploration. The exhaustive list 210 of possible features (all combinations of feature vectors) which can be derived from the input variables (including the input variables, considered singly, themselves) comprises a total of fifteen features: the four single-variable features 211, six two-variable or quadratic 50 features 212 {AxB, AxC, AxD, BxC, BxD, CxD (where the symbol xis used to denote the combination of the two input variables on either side of the x)}, four three-variable features 213 {AxBxC, AxBxD, AxCxD and BxCxD} and one four-variable feature 214 {AxBxCxD}” (Kumar, column 11, paragraph 3)
“a two-phase technique based on efficient hashing-based approximations may be used in the depicted embodiments to select a useful set of features (i.e., a smaller set of features than shown in list 210 that are likely to be the most highly-correlated with the target variable T). In the first phase, feature processing coordinator 230 may perform min-wise hashing-based approximate correlation analysis 233, resulting in the identification of a pruned high-predictive-value candidate feature set 235” (Kumar, column 12, paragraph 2)
“One example of a correlation metric which may be used in at least some embodiments is symmetric uncertainty; other metrics such as different types of mutual information metrics may be employed in other embodiments. In order to compute the exact symmetric uncertainty U(f, T) with respect to a single input variable or feature f (combination feature vector) and a target variable T (correct answer label), three entropy values H(f), H(T) and H(f,T) may have to be computed, which in turn require the computation of the probabilities shown in equations E1.1, E1.2 and E1.3 of equation set E1. In these equations, N is the total number of observations,
N
u
is the count of observations in which the unidimensional feature f has the value u,
N
t
is the count of observations in which the target variable T has the value t, and
N
t
u
is the count of observations (frequency distribution) in which the target variable has the value t and the feature f has the value u” (Kumar, column 12, paragraph 4). A frequency distribution (count) is created for each combination of feature vectors, including the single-dimensional feature vectors.
specify a combination of the feature vectors in which a similarity in the frequency distribution of the correct answer label is equal to or higher than a predetermined first threshold value, as the combination of the feature vectors that are allowed to be merged:
“Using these approximate counts for
N
t
u
v
, an approximate symmetric uncertainty metric (or other correlation metrics) can be obtained for quadratic features, and these approximate metrics (similarit[ies]) can be used to select a set of candidate quadratic features (combination[s] of the feature vectors)” (Kumar, column 14, paragraph 1)
“Then, these approximate correlation values (similarit[ies]) may be used to select those multidimensional features (combination[s] of features) which meet a first correlation threshold criterion. For example, only those multidimensional features whose approximate symmetric uncertainty exceeds SUthreshold1 may be selected, while those multidimensional features which have an approximate symmetric uncertainty value less than or equal to SUthreshold1 may be rejected. The selected metrics may be included in a candidate feature set (combination[s] of the feature vectors that are allowed to be merged)” (Kumar, column 6, paragraph 2). Only combinations with correlation scores above the threshold are kept (i.e., only these combinations are allowed to stay merged).
Regarding claim 3, the rejection of claim 2 is incorporated. Kumar further discloses a system, wherein, in the specifying processing, the processor is configured to
further create, for a combination specified as the combination of the feature vectors that are allowed to be merged, a frequency distribution in consideration of a combination of a plurality of items, and exclude the specified combination from the combinations of the feature vectors that are allowed to be merged in a case where a similarity in the frequency distribution in consideration of the combination of the items is lower than a predetermined second threshold value:
“In the second phase of the technique, the candidate features of set 235 (combination[s] of the feature vectors that are allowed to be merged) may be subjected to an exact correlation analysis 236, in which the hashing-based approximation approach of the first phase is not used. The exact correlation values determined for the candidate features may in some cases be used to further prune the candidate features, resulting in an even smaller final approved feature set 237. In the depicted example, the approved feature set includes only A, B, C and BxD; AxC has been pruned as a result of the exact correlation analysis. Rationale for Using Min-Wise Hashing” (Kumar, column 12, paragraph 2)
“The candidate feature set may be evaluated further in at least one embodiment, e.g., by computing exact rather than approximate population counts (frequency distribution[s]) and exact rather than approximate correlation metrics (similarit[ies]) with respect to the target variable. Those candidate feature set members which meet a second correlation threshold criterion SUthreshold2 may be retained in an approved feature set” (Kumar, column 6, paragraph 3). Feature combinations with correlation scores below the threshold are excluded from the approved feature set.
Regarding claim 4, the rejection of claim 1 is incorporated. Kumar further discloses a system, wherein, in the specifying processing, the processor is configured to
create, for each of the feature vectors included in the data set, a frequency distribution of a correct answer label in consideration of a combination of a plurality of items:
“Example data set 202, which is to be used to train a model, comprises ten million observation records labeled OR1-OR10000000. Each observation record comprises four input variables A, B, C and D (feature vectors), and a target variable T. Some combinations of the input variables A, B, C and D may be highly correlated with the target variable, but which specific combinations are highly correlated may not be known at the 45 start of the feature exploration. The exhaustive list 210 of possible features (all combinations of feature vectors) which can be derived from the input variables (including the input variables, considered singly, themselves) comprises a total of fifteen features: the four single-variable features 211, six two-variable or quadratic 50 features 212 {AxB, AxC, AxD, BxC, BxD, CxD (where the symbol xis used to denote the combination of the two input variables on either side of the x)}, four three-variable features 213 {AxBxC, AxBxD, AxCxD and BxCxD} and one four-variable feature 214 {AxBxCxD}” (Kumar, column 11, paragraph 3)
“a two-phase technique based on efficient hashing-based approximations may be used in the depicted embodiments to select a useful set of features (i.e., a smaller set of features than shown in list 210 that are likely to be the most highly-correlated with the target variable T). In the first phase, feature processing coordinator 230 may perform min-wise hashing-based approximate correlation analysis 233, resulting in the identification of a pruned high-predictive-value candidate feature set 235” (Kumar, column 12, paragraph 2)
“One example of a correlation metric which may be used in at least some embodiments is symmetric uncertainty; other metrics such as different types of mutual information metrics may be employed in other embodiments. In order to compute the exact symmetric uncertainty U(f, T) with respect to a single input variable or feature f (combination of a plurality of items (features)) and a target variable T (correct answer label), three entropy values H(f), H(T) and H(f,T) may have to be computed, which in turn require the computation of the probabilities shown in equations E1.1, E1.2 and E1.3 of equation set E1. In these equations, N is the total number of observations,
N
u
is the count of observations in which the unidimensional feature f has the value u,
N
t
is the count of observations in which the target variable T has the value t, and
N
t
u
is the count of observations (frequency distribution) in which the target variable has the value t and the feature f has the value u” (Kumar, column 12, paragraph 4).
specify a combination of the feature vectors in which a similarity in the frequency distribution of the correct answer label is equal to or higher than a predetermined seventh threshold value, as the combination of the feature vectors that are allowed to be merged:
“Using these approximate counts for
N
t
u
v
, an approximate symmetric uncertainty metric (or other correlation metrics) can be obtained for quadratic features, and these approximate metrics (similarit[ies]) can be used to select a set of candidate quadratic features (combination[s] of the feature vectors)” (Kumar, column 14, paragraph 1)
“Then, these approximate correlation values (similarit[ies]) may be used to select those multidimensional features (combination[s] of features) which meet a first correlation threshold criterion. For example, only those multidimensional features whose approximate symmetric uncertainty exceeds SUthreshold1 may be selected, while those multidimensional features which have an approximate symmetric uncertainty value less than or equal to SUthreshold1 may be rejected. The selected metrics may be included in a candidate feature set (combination[s] of the feature vectors that are allowed to be merged)” (Kumar, column 6, paragraph 2). Only combinations with correlation scores above the threshold are kept (i.e., only these combinations are allowed to stay merged).
Regarding claim 5, the rejection of claim 1 is incorporated. Kumar further discloses a system, wherein, in the rule generation processing, the processor is configured to end generation of the merging rule in a case where the number of combinations of the feature vectors that are included in the merging rule and are allowed to be merged is equal to or larger than a predetermined third threshold value: “in at least some embodiments the feature exploration request may include parameters such as (a) the maximum number of features to be combined when considering multidimensional features” (Kumar, column 8, paragraph 3).
Regarding claim 8, the rejection of claim 1 is incorporated. Kumar further discloses a system, wherein, in the specifying processing, candidates of the feature vectors that are allowed to be merged are determined based on at least one of an edit distance, a distribution representation, or related information of the feature vectors:
“Accordingly, in at least some embodiments, a technique involving the use of a min-wise hashing algorithm to obtain approximate occurrence counts and corresponding approximate correlation metrics may be employed … In a first stage, a candidate feature set (candidates of the feature vectors) comprising a selected subset of multidimensional features may be constructed efficiently using min-wise hashing. During this phase, multidimensional features which are unlikely to be highly correlated with the target variable may be discarded from further consideration. The candidate feature set members to be retained for further evaluation may be selected using approximate occurrence counts which rely on the set-similarity-detection capabilities of min-wise hashing algorithms” (Kumar, column 4, paragraph 4)
“The data set may include numerous observation records, each including some number of input variables and a target variable. Using signatures (distribution representation[s] / related information) obtained by applying a plurality of hash functions (or other similar transformation functions) to the observation records, approximate population counts for various subsets of the data set may be determined, where the member observation records of each subset meet a particular co-occurrence criterion.” (Kumar, column 5, paragraph 3)
“approximate correlation values may be used to select those multidimensional features (combination[s] of features) which meet a first correlation threshold criterion … The selected metrics may be included in a candidate feature set (combination[s] of the feature vectors that are allowed to be merged)” (Kumar, column 6, paragraph 2). Only combinations with correlation scores above the threshold are kept (i.e., only these combinations are allowed to stay merged).
Regarding claim 10, the rejection of claim 1 is incorporated. Kumar further discloses [a] learning device that trains a machine learning model by using a training data set obtained by performing merging according to a merging rule generated by the device for generating a data merging rule according to claim 1:
“A final approved feature set which can be used to train a linear model may be obtained based on the results of these exact calculations” (Kumar, column 5, paragraph 1)
“A number of different types of models (machine learning model[s]s) may be trained using the approved feature sets in various embodiments, including binary classification models and multi-class classification models. In at least one embodiment a regression model may also be trained using at least some multidimensional features selected using the min-wise hashing approach” (Kumar, column 5, paragraph 2)
Regarding claim 11, the rejection of claim 1 is incorporated. Kumar further discloses [a] prediction device that causes a machine learning model to perform prediction by using, as an input, data obtained by performing merging according to the merging rule generated by the device for generating a data merging rule according to claim:
“According to one embodiment, a data set to be used to train a machine learning model may be identified, e.g., in response to a client's model generation request submitted to” (Kumar, column 5, paragraph 3)
“A detailed evaluation of the candidate features may then be performed (e.g., by obtaining exact co-occurrence counts and exact correlation metrics) (element 319). Those features which meet a second threshold criterion may be retained in a finalized or approved feature set (element 322). The approved feature set may optionally be used to train a model ( e.g., a linear model) to predict values of the target variable (element 325) using any desired training algorithm” (Kumar, column 16, paragraph 4)
Regarding claim 12, Kumar discloses [a]n operation method for a device for generating a data merging rule for a machine learning model, the method comprising: a step of specifying a combination of feature vectors that are included in a data set including a correct answer label and are allowed to be merged; and a step of generating a merging rule of the feature vectors based on a combination of the feature vectors that are allowed to be merged:
“The data set may include numerous observation records, each including some number of input variables and a target variable (correct answer label)” (Kumar, column 5, paragraph 3)
“In many cases, discovering informative multidimensional features (combination of feature vectors) (e.g., higher-order features whose inclusion in a linear model's input feature set is likely to lead to higher-quality predictions) may involve obtaining the values of correlation metrics between the multidimensional features and the target variable. Computing such correlation metrics may in tum involve obtaining occurrence counts for each combination of (a) a value of the multidimensional feature 45 and (b) a value of the target variable. Such occurrence counts, associated with matched combinations of values of different variables/features, may also be referred to as cooccurrence counts” (Kumar, column 4, paragraph 3). The procedure by which informative multidimensional features are discovered comprises a set of merging rule[s].
Kumar relates to selectively training a machine learning model with combinations of features selected based on frequency distributions and is analogous to the claimed invention.
Regarding claim 13, Kumar discloses [a] non-transitory computer readable medium storing a program that generates a data merging rule for a machine learning model: “In some embodiments, system memory 9020 may be one embodiment of a computer-accessible medium configured to store program instructions and data as described above for FIG. 1 through FIG. 11 for implementing embodiments of 60 the corresponding methods and apparatus. However, in other embodiments, program instructions and/or data may be received, sent or stored upon different types of computer-accessible media. Generally speaking, a computer-accessible medium may include non-transitory storage media or 65 memory media such as magnetic or optical media,” (Kumar, column 23, paragraph 4)
Kumar’s program causing a computer to execute a process comprising: a step of specifying a combination of feature vectors that are included in a data set including a correct answer label and are allowed to be merged; and a step of generating a merging rule of the feature vectors based on a combination of the feature vectors that are allowed to be merged:
“The data set may include numerous observation records, each including some number of input variables and a target variable (correct answer label)” (Kumar, column 5, paragraph 3)
“In many cases, discovering informative multidimensional features (combination of feature vectors) (e.g., higher-order features whose inclusion in a linear model's input feature set is likely to lead to higher-quality predictions) may involve obtaining the values of correlation metrics between the multidimensional features and the target variable. Computing such correlation metrics may in tum involve obtaining occurrence counts for each combination of (a) a value of the multidimensional feature 45 and (b) a value of the target variable. Such occurrence counts, associated with matched combinations of values of different variables/features, may also be referred to as cooccurrence counts” (Kumar, column 4, paragraph 3). The procedure by which informative multidimensional features are discovered comprises a set of merging rule[s].
Kumar relates to selectively training a machine learning model with combinations of features selected based on frequency distributions and is analogous to the claimed invention.
Claim 25 is rejected under 35 U.S.C. 102(a)(2) as being anticipated by Fisher et al. (Merge and Label: A novel neural network architecture for nested NER, published 2019, Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp. 5840-5850), hereafter referred to as Fisher.
Regarding claim 25, Fisher discloses [a]n operation method for a learning device for a machine learning model including a merging layer that converts first feature vectors into second feature vectors and outputs the second feature vectors:
“An overview of the architecture used to predict the structure and labels is shown in Figure 2 … The input tensor, X, holds the word embeddings (first feature vectors) of dimension e, for every word in the input of sequence length, s … The Static Layer updates the token embeddings using contextual information” Fisher, page 5841, left column, paragraph 4). The updated token embeddings are second feature vectors.
“Next, for u repetitions, we go through a series of building the structure using the Structure Layer, and then use this structure to continue updating the individual token embeddings using the Update Layer, giving an output
X
u
” (Fisher, page 5841, left column, paragraph 4).
“The Structure Layer is responsible for three tasks. Firstly, deciding which token embeddings should be merged at each level, expressed as real values between 0 and 1” (Fisher, page 5843, left column, paragraph 3)
PNG
media_image1.png
382
377
media_image1.png
Greyscale
(Fisher, page 5841, right column, Figure 2). The static layer, structure layer, and update layers constitute merging layers, or, equivalently, one large merging layer used to merge token embeddings in the system.
Fisher’s method comprising:
a step of training the machine learning model by using the second feature vectors:
“During training we initialize the weights of the network using the identity function. As a result, the default behaviour of FFEU prior to training is to pass on the word embedding unchanged, which is then updated during via backpropagation” (Fisher, page 5842, right column, paragraph 3)
“For both datasets, during training, we replace all “B-” labels with their corresponding “I-” label. At evaluation, all predictions which are the first word in a merged entity have the “B-” added back on. As the trained model’s merging weights, M, can take any value between 0 and 1, we have to set a cutoff at eval time when deciding which words are in the same entity. We perform a grid search over cutoff values using the dev set, with a value of 0.75 proving optimal” (Fisher, page 5845, left column, paragraph 5)
“The model is trained to predict the correct merge decisions, held in the tensor M of dimension [b, s- 1, L] and the correct class labels given these decisions, C. The merge decisions are trained directly using the mean absolute error (MAE)” (Fisher, page 5845, right column, paragraph 2)
wherein the step of training the machine learning model includes a step of merging the second feature vectors output from the merging layer by changing a conversion rule from the first feature vectors to the second feature vectors in the merging layer:
“The Structure Layer is responsible for three tasks. Firstly, deciding which token embeddings (second feature vectors) should be merged at each level, expressed as real values between 0 and 1, and denoted M. Secondly, given these merge values M, deciding how the separate token embeddings should be combined in order to give the embeddings for each entity, T” (Fisher, page 5843, left column, paragraph 3)
PNG
media_image2.png
132
1041
media_image2.png
Greyscale
”Trained model’s representation of nested entities, after thresholding the merge values” (Fisher, page 2, Figure 1)
Fisher relates to merging features based on feature embeddings and is analogous to the claimed invention.
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.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 6-7 are rejected under 35 U.S.C. 103 as being unpatentable over Kumar et al. (SCALABLE GENERATION OF MULTIDIMENSIONAL FEATURES FOR MACHINE LEARNING, filed 4/19/2016, US 11,295,229 B1), hereafter referred to as Kumar, in view of Mehanna et al. (SELECTION AND MODIFICATION OF FEATURES USED BY ONE OR MORE MACHINE LEARNED MODELS USED BY AN ONLINE SYSTEM, published 3/31/2016, US 2016/0092786 A1), hereafter referred to as Mehanna.
Regarding claim 6, the rejection of claim 1 is incorporated. While Kumar fails to disclose the further limitations of the claim, Mehanna discloses a device configured to:
generate a provisional model in which the feature vectors included in the data set are used as inputs and train the provisional model: “the machine learning module 230 applies a machine learned model (provisional model) to data identified from a set of training data and associated with features included in an identified group” (Mehanna, [0033])
select a combination of the feature vectors from the data set, and specify, as the combination of the feature vectors that are allowed to be merged, the selected combination of the feature vectors in a case where a change value of a prediction result of the provisional model in a case where the selected combination of the feature vectors is swapped is lower than a predetermined fourth threshold value:
“When generating the groups of features (combination[s] of the feature vectors), the machine learning module 230 determines a measure of feature impact for the machine learned model associated with each feature in the set. A measure of feature impact associated with a feature provides a measure of the feature's importance to the machine learned model. In one embodiment, a measure of feature impact is proportional to the importance of the feature to the machine learned model, so features that are more important to the machine learned model (e.g., features resulting in a larger change to an error term (prediction result) of the machine learned model if removed (swapped) from the machine learned model) are associated with higher measures of feature impact” (Mehanna, [0030])
“Based on the measures of feature impact associated with features from the set of features, the online system 140 generates 320 various groups each including different features from the set of features. For example, different groups include different numbers of features or include different features. In various embodiments, the online system 140 generates 320 a group including each feature form the set of features, a group including features originally received by the machine learned model, and one or more intermediate groups each having different numbers of features from the set of features. The online system 140 may associate an intermediate group (selected combination of the feature vectors) with a threshold measure of feature impact, so the intermediate group includes features having measures of feature impact equaling or exceeding the threshold measure of feature impact but does not include features having measures of feature impact less than the threshold measure of feature impact” (Mehanna, [0039])
Mehanna relates to automatic feature selection for machine learning models and is analogous to the claimed invention. The existing combination teaches a device capable of combining feature vectors together. The claimed invention improves upon this method by merging vectors in response to a change in prediction results. Mehanna teaches a system able to discard features with significant negative performance impacts from groupings, applicable to the existing combination. A person of ordinary skill in the art would have recognized that filtering negative impact features out of feature groups would lead to the predictable result of utilizing only a subset of the best-performing features to train the model, and would improve the known device by increasing model performance (MPEP 2143 I. (D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results).
Regarding claim 7, the rejection of claim 1 is incorporated. While Kumar fails to disclose the further limitations of the claim, Mehanna discloses a device configured to:
generate a provisional model in which the feature vectors included in the data set are used as inputs and train the provisional model: “the machine learning module 230 applies a machine learned model (provisional model) to data identified from a set of training data and associated with features included in an identified group” (Mehanna, [0033])
select a combination of the feature vectors from the data set, and specify, as the combination of the feature vectors that are allowed to be merged, the selected combination of the feature vectors in a case where a similarity in a prediction result of the provisional model in a case where the selected combination of the feature vectors is swapped is equal to or higher than a predetermined fourth similarity: “Additional information may also be used by the machine learning module 230 to generate intermediate groups (merge[d] combinations of the second feature vectors). For example, a measure of correlation (similarity) between different features may be determined, and additional features having greater than a threshold measure of correlation to a feature in the set may be excluded from inclusion in a group (selected combination of the vectors) including the feature.” (Mehanna, [0032]). As would be known by one of ordinary skill in the art, the addition of highly correlated features to a machine learning model result in similar prediction result[s]. Thus, features that would lead to similar prediction results are being culled from these groups.
Mehanna relates to automatic feature selection for machine learning models and is analogous to the claimed invention. The existing combination teaches a device capable of combining feature vectors together. The claimed invention improves upon this method by merging vectors in response to the similarity of prediction results. Mehanna teaches a system able to discard features significant correlations with other features from groupings, applicable to the existing combination. A person of ordinary skill in the art would have recognized that filtering highly correlated features out of feature groups would lead to the predictable result of utilizing a smaller subset of features to attain similar predictive results from the model, and would improve the known device by increasing decreasing model training complexity and length (MPEP 2143 I. (D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results).
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Kumar et al. (SCALABLE GENERATION OF MULTIDIMENSIONAL FEATURES FOR MACHINE LEARNING, filed 4/19/2016, US 11,295,229 B1), hereafter referred to as Kumar, in view of Romba (AUTOMATED ASSEMBLY OF ECU CALIBRATIONS, published 3/12/2016, US 2016/0131067 A1).
Regarding claim 9, the rejection of claim 1. Kumar further discloses a system, wherein the processor is further configured to: … reception processing of receiving, from a user, whether or not to merge the combination of the feature vectors that are allowed to be merged: “Using the programmatic interfaces 150, a client 174 (user) may submit an implicit or explicit request to identify a set of features to be used to train a model using a specified data set 108 in the depicted embodiment. The request may be implicit in some cases, in that the client may simply request the generation or creation of a predictive model for a target variable of the data set 108, without specifically indicating that multidimensional features should be evaluated. The administrative components of the machine learning service, such as the feature processing coordinator 130 or the model training coordinator 142 may in some cases determine, in response to receiving such a request and performing a preliminary analysis of at least a portion of the data set, that an evaluation of multidimensional features is appropriate. In other cases an explicit feature exploration request may be submitted, e.g., comprising the logical equivalent of the request "please determine which, if, any quadratic or other multidimensional features should be generated to help develop a linear model with a high level of predictive accuracy for data set 108A". Parameters such as a target budget constraint (expressed in any of various units such as resource usage, elapsed time, a currency, or the like), the types of multidimensional features to be considered, parallelism parameters for one or more phases of the analysis, and so on, may be included in the explicit or implicit request.” (Kumar, column 9, paragraph 3).
While Kumar fails to disclose the further limitations of the claim, Romba discloses a system, wherein the processor is further configured to: display processing of displaying the combination of the feature vectors that are allowed to be merged on a display unit: “At step 106, features selected by the user at step 104 may be combined pursuant to the rules identified in the decision tree to identify allowable combinations of features (combination[s] of the feature vectors that are allowed to be merged) and/or the permissible engine calibrations for those combined features. At step 108, these permissible combinations may be provided to the output 18, such as, for example, being displayed in a window 80 on the graphical user interface 50 (display unit)” (Romba, [0024])
Romba relates to selective feature combination searches and is analogous to the claimed invention. Kumar teaches a system that generates selective feature combinations based, in-part, on user requests. Romba teaches a system that displays selected feature combinations to a user display device. It would have been obvious to one of ordinary skill in the art to combine Kumar and Romba by displaying the selected combinations of features from Romba’s system to a user display device. This would achieve the predictable result of allowing the user to observe the most relevant combined features pertinent to their query, with Kumar’s feature combination selection system and Romba’s feature combination displaying system performing the same together as they did separately. (MPEP 2143 I. (A) Combining prior art elements according to known methods to yield predictable results).
Claims 14-16, 18-19, 21-22, and 26 are rejected under 35 U.S.C. 103 as being unpatentable over Fisher et al. (Merge and Label: A novel neural network architecture for nested NER, published 2019, Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp. 5840-5850), hereafter referred to as Fisher, in view of Kumar et al. (SCALABLE GENERATION OF MULTIDIMENSIONAL FEATURES FOR MACHINE LEARNING, filed 4/19/2016, US 11,295,229 B1), hereafter referred to as Kumar.
Regarding claim 14, Fisher discloses [a] learning device for a machine learning model … wherein the machine learning model includes a merging layer that converts first feature vectors into second feature vectors and outputs the second feature vectors:
“An overview of the architecture used to predict the structure and labels is shown in Figure 2 … The input tensor, X, holds the word embeddings (first feature vectors) of dimension e, for every word in the input of sequence length, s … The Static Layer updates the token embeddings using contextual information” Fisher, page 5841, left column, paragraph 4). The updated token embeddings are second feature vectors.
“Next, for u repetitions, we go through a series of building the structure using the Structure Layer, and then use this structure to continue updating the individual token embeddings using the Update Layer, giving an output
X
u
” (Fisher, page 5841, left column, paragraph 4).
“The Structure Layer is responsible for three tasks. Firstly, deciding which token embeddings should be merged at each level, expressed as real values between 0 and 1” (Fisher, page 5843, left column, paragraph 3)
PNG
media_image1.png
382
377
media_image1.png
Greyscale
(Fisher, page 5841, right column, Figure 2). The static layer, structure layer, and update layers constitute merging layers, or, equivalently, one large merging layer used to merge token embeddings in the system.
Fisher further discloses instructions to:
execute training processing of training the machine learning model in response to an input of the second feature vector:
“During training we initialize the weights of the network using the identity function. As a result, the default behaviour of FFEU prior to training is to pass on the word embedding unchanged, which is then updated during via backpropagation” (Fisher, page 5842, right column, paragraph 3)
“For both datasets, during training, we replace all “B-” labels with their corresponding “I-” label. At evaluation, all predictions which are the first word in a merged entity have the “B-” added back on. As the trained model’s merging weights, M, can take any value between 0 and 1, we have to set a cutoff at eval time when deciding which words are in the same entity. We perform a grid search over cutoff values using the dev set, with a value of 0.75 proving optimal” (Fisher, page 5845, left column, paragraph 5)
“The model is trained to predict the correct merge decisions, held in the tensor M of dimension [b, s- 1, L] and the correct class labels given these decisions, C. The merge decisions are trained directly using the mean absolute error (MAE)” (Fisher, page 5845, right column, paragraph 2)
merge, in the training processing, the second feature vectors output from the merging layer by changing a conversion rule from the first feature vectors to the second feature vectors in the merging layer:
“The Structure Layer is responsible for three tasks. Firstly, deciding which token embeddings (second feature vectors) should be merged at each level, expressed as real values between 0 and 1, and denoted M. Secondly, given these merge values M, deciding how the separate token embeddings should be combined in order to give the embeddings for each entity, T” (Fisher, page 5843, left column, paragraph 3)
PNG
media_image2.png
132
1041
media_image2.png
Greyscale
”Trained model’s representation of nested entities, after thresholding the merge values” (Fisher, page 2, Figure 1)
Fisher relates to merging features based on feature embeddings and is analogous to the claimed invention.
While Fisher fails to disclose the further limitations of the claim, Kumar discloses [a] learning device for a machine learning model, the device comprising: a processor; and a memory connected to or built in the processor:
“Various embodiments of methods and apparatus for scalable generation of multidimensional features for machine learning data sets are described” (Kumar, column 2, Detailed Description)
“In the illustrated embodiment, computing device 9000 includes one or more processors 9010 coupled to a system memory 9020” (Kumar, column 22, paragraph 5)
Kumar relates to selectively training a machine learning model with combinations of features selected based on frequency distributions and is analogous to the claimed invention. Fisher teaches a system for merging token embeddings. The claimed invention improves upon this method by storing it in the form of instructions on computer hardware. Kumar teaches computer hardware capable of running feature merging systems, applicable to Fisher. A person of ordinary skill in the art would have recognized that storing Fisher’s method as computer instructions on Kumar’s hardware would lead to the predictable result of the method being executable by a computing system, and would improve the known device by allowing it to be performed with real data (MPEP 2143 I. (D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results).
Regarding claim 15, the rejection of claim 14 is incorporated. Fisher further discloses instructions, wherein the processor is configured to, in the training processing, change the conversion rule in the merging layer by using an algorithm in which a score is given based on a value of a loss function used for training the machine learning model:
“There are four separate outputs from the Structure Layer … Finally, the fourth output, M , stores the merge values (score[s]) for every level. It is used in the loss function, to directly incentivize the correct merge decisions at the correct levels” (Fisher, page 5844, right column, paragraph 2)
“The merge decisions are trained directly using the mean absolute error (MAE):
PNG
media_image3.png
54
231
media_image3.png
Greyscale
This is then weighted by a scalar
w
M
, and added to the usual Cross Entropy (CE) loss from the predictions of the classes, CEC, giving a final loss function of the form
PNG
media_image4.png
27
271
media_image4.png
Greyscale
” (Fisher, page 5845, right column, paragraph 2)
Regarding claim 16, the rejection of claim 15 is incorporated. Fisher further discloses instructions, wherein the score of the algorithm includes the number of the second feature vectors to be merged in the merging layer: “The Structure Layer is responsible for three tasks. Firstly, deciding which token embeddings (second feature vectors) should be merged at each level, expressed as real values between 0 and 1, and denoted M (score[s])” (Fisher, page 5843, left column, paragraph 3)
Regarding claim 18, the rejection of claim 14 is incorporated. Fisher further discloses instructions, wherein:
the machine learning model further includes an embedding layer that outputs embedding vectors corresponding to the second feature vectors: “The Static Layer (embedding layer) updates the token embeddings using contextual information” (Fisher, page 5841, left column, paragraph 3)
the processor is configured to, in the training processing, make a combination of the similar embedding vectors more similar:
“We pass the embeddings (X) of each pair of adjacent words through a feedforward NN FFS to give directions D [b, s-1, d] and merge values M [b, s-1, 1] between each pair. If FFS predicts M(1;2) to be close to 0, this indicates that tokens 1 and 2 are part of the same entity (similar) on this level.” (Fisher, page 5843, left column, paragraph 4)
“M holds, for every pair of adjacent words (s - 1 given input length s) and every output level (L levels), a value between 0 and 1. A value close to 0 denotes that the two (adjacent) tokens/entities from the previous level are likely to be merged on this level to form an entity (combination of similar embedding vectors); nested entities emerge when entities from lower levels are used.” (Fisher, page 5841, right column, paragraph 2)
“The model is trained to predict the correct merge decisions, held in the tensor M of dimension [b, s-1, L] and the correct class labels given these decisions, C. The merge decisions are trained directly using the mean absolute error (MAE):
PNG
media_image3.png
54
231
media_image3.png
Greyscale
This is then weighted by a scalar
w
M
, and added to the usual Cross Entropy (CE) loss from the predictions of the classes, CEC, giving a final loss function of the form
PNG
media_image4.png
27
271
media_image4.png
Greyscale
” (Fisher, page 5845, right column, paragraph 2)
Regarding claim 19, the rejection of claim 18 is incorporated. Fisher further discloses instructions, wherein the processor is configured to, in the training processing, introduce a term that makes the combination of the similar embedding vectors more similar, to a loss function used for training the machine learning model: “The model is trained to predict the correct merge decisions, held in the tensor M of dimension [b, s-1, L] and the correct class labels given these decisions, C. The merge decisions are trained directly using the mean absolute error (MAE) (term that makes the combination of the similar embedding vectors more similar):
PNG
media_image3.png
54
231
media_image3.png
Greyscale
This is then weighted by a scalar
w
M
, and added to the usual Cross Entropy (CE) loss from the predictions of the classes, CEC, giving a final loss function of the form
PNG
media_image4.png
27
271
media_image4.png
Greyscale
” (Fisher, page 5845, right column, paragraph 2)
Regarding claim 21, the rejection of claim 18 is incorporated. Fisher further discloses instructions, wherein the processor is configured to, in the training processing, add a correction value for making a combination of the embedding vectors more similar, to at least one of combinations of the embedding vectors having a similarity equal to or higher than a predetermined third similarity:
“the fourth output, M, stores the merge values for every level. It is used in the loss (correction value) function, to directly incentivize the correct merge decisions at the correct levels.” (Fisher, page 5844, right column, paragraph 2)
“As the trained model’s merging weights, M, can take any value between 0 and 1, we have to set a cutoff (predetermined similarity threshold) at eval time when deciding which words are in the same entity. We perform a grid search over cutoff values using the dev set, with a value of 0.75 proving optimal.” (Fisher, page 5845, left column, paragraph 5)
“progressively larger entities to be formed by combining smaller entities from the previous levels” (Fisher, page 5844, left column, paragraph 2)
PNG
media_image2.png
132
1041
media_image2.png
Greyscale
”Trained model’s representation of nested entities, after thresholding the merge values” (Fisher, page 2, Figure 1)
The merge values M are corrected via the loss function as the system is trained, such that merges between related (similar) entities are more likely. As shown in Figure 1, entities at level 1 and above of processing each consist of a combination of the embedding vectors. Thus, the influence of loss on M encourages merging between related combinations by pushing associated M values above the predetermined similarity (the cutoff value).
Regarding claim 22, the rejection of claim 18 is incorporated. Fisher further discloses instructions, The learning device according to claim 18, wherein the processor is configured to, in the training processing, merge combinations of the second feature vectors that correspond to combinations of the embedding vectors having a similarity equal to or higher than a predetermined first similarity: “As the trained model’s merging weights, M, can take any value between 0 and 1, we have to set a cutoff (threshold) at eval time when deciding which words are in the same entity. We perform a grid search over cutoff values using the dev set, with a value of 0.75 proving optimal.” (Fisher, page 5845, left column, paragraph 5)
Regarding claim 26, Fisher discloses instructions for training a machine learning model including a merging layer that converts first feature vectors into second feature vectors and outputs the second feature vectors:
“An overview of the architecture used to predict the structure and labels is shown in Figure 2 … The input tensor, X, holds the word embeddings (first feature vectors) of dimension e, for every word in the input of sequence length, s … The Static Layer updates the token embeddings using contextual information” Fisher, page 5841, left column, paragraph 4). The updated token embeddings are second feature vectors.
“Next, for u repetitions, we go through a series of building the structure using the Structure Layer, and then use this structure to continue updating the individual token embeddings using the Update Layer, giving an output
X
u
” (Fisher, page 5841, left column, paragraph 4).
“The Structure Layer is responsible for three tasks. Firstly, deciding which token embeddings should be merged at each level, expressed as real values between 0 and 1” (Fisher, page 5843, left column, paragraph 3)
PNG
media_image1.png
382
377
media_image1.png
Greyscale
(Fisher, page 5841, right column, Figure 2). The static layer, structure layer, and update layers constitute merging layers, or, equivalently, one large merging layer used to merge token embeddings in the system.
Fisher’s program causing a computer to execute a process comprising:
a step of training the machine learning model by using the second feature vectors:
“During training we initialize the weights of the network using the identity function. As a result, the default behaviour of FFEU prior to training is to pass on the word embedding unchanged, which is then updated during via backpropagation” (Fisher, page 5842, right column, paragraph 3)
“For both datasets, during training, we replace all “B-” labels with their corresponding “I-” label. At evaluation, all predictions which are the first word in a merged entity have the “B-” added back on. As the trained model’s merging weights, M, can take any value between 0 and 1, we have to set a cutoff at eval time when deciding which words are in the same entity. We perform a grid search over cutoff values using the dev set, with a value of 0.75 proving optimal” (Fisher, page 5845, left column, paragraph 5)
“The model is trained to predict the correct merge decisions, held in the tensor M of dimension [b, s- 1, L] and the correct class labels given these decisions, C. The merge decisions are trained directly using the mean absolute error (MAE)” (Fisher, page 5845, right column, paragraph 2)
wherein the step of training the machine learning model includes a step of merging the second feature vectors output from the merging layer by changing a conversion rule from the first feature vectors to the second feature vectors in the merging layer.
“The Structure Layer is responsible for three tasks. Firstly, deciding which token embeddings (second feature vectors) should be merged at each level, expressed as real values between 0 and 1, and denoted M. Secondly, given these merge values M, deciding how the separate token embeddings should be combined in order to give the embeddings for each entity, T” (Fisher, page 5843, left column, paragraph 3)
PNG
media_image2.png
132
1041
media_image2.png
Greyscale
”Trained model’s representation of nested entities, after thresholding the merge values” (Fisher, page 2, Figure 1)
Fisher relates to merging features based on feature embeddings and is analogous to the claimed invention.
While Fisher fails to disclose the further limitations of the claim, Kumar discloses A non-transitory computer readable medium storing a program for training a machine learning model:
“Various embodiments of methods and apparatus for scalable generation of multidimensional features for machine learning data sets are described” (Kumar, column 2, Detailed Description)
“In the illustrated embodiment, computing device 9000 includes one or more processors 9010 coupled to a system memory 9020” (Kumar, column 22, paragraph 5)
“In some embodiments, system memory 9020 may be one embodiment of a computer-accessible medium configured to store program instructions and data as described above for FIG. 1 through FIG. 11 for implementing embodiments of 60 the corresponding methods and apparatus. However, in other embodiments, program instructions and/or data may be received, sent or stored upon different types of computer-accessible media. Generally speaking, a computer-accessible medium may include non-transitory storage media” (Kumar, column 23, paragraph 4)
Kumar relates to selectively training a machine learning model with combinations of features selected based on frequency distributions and is analogous to the claimed invention. Fisher teaches a system for merging token embeddings. The claimed invention improves upon this method by storing it in the form of instructions on computer hardware. Kumar teaches computer hardware capable of running feature merging systems, applicable to Fisher. A person of ordinary skill in the art would have recognized that storing Fisher’s method as computer instructions on Kumar’s hardware would lead to the predictable result of the method being executable by a computing system, and would improve the known device by allowing it to be performed with real data (MPEP 2143 I. (D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results).
Claim 17 is rejected under 35 U.S.C. 103 as being unpatentable over Fisher et al. (Merge and Label: A novel neural network architecture for nested NER, published 2019, Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp. 5840-5850), hereafter referred to as Fisher, in view of Kumar et al. (SCALABLE GENERATION OF MULTIDIMENSIONAL FEATURES FOR MACHINE LEARNING, filed 4/19/2016, US 11,295,229 B1), hereafter referred to as Kumar, and further in view of Thirumalai et al. (GENERATING SIMILARITY SCORES FOR MATCHING NON-IDENTICAL DATA STRINGS, patented 10/12/2010, US 7,814,107 B1), hereafter referred to as Thirumalai.
Regarding claim 17, the rejection of claim 15 is incorporated. While the aforementioned references fail to disclose the further limitations of the claim, Thirumalai discloses instructions, wherein an initial value of the score of the algorithm is determined based on at least one of an edit distance, a distribution representation, or related information of the first feature vectors which are input to the merging layer: “the edit distance function compares and scores relative similarity between tokens (first feature vectors) based, in part, on the type of token. When a document is tokenized, as mentioned above, each token is categorized into one of three types: an alphabetic token in which all characters are alphabetic characters; a numeric token where the characters collectively identify a numeric value; and alpha-numeric tokens comprising mixed alphabetic and numeric characters (as well as other characters). Knowing these token types, when comparing and scoring the similarity between two tokens, the following rules may be applied: tokens of different types, when compared, have a similarity score of zero; alpha-numeric tokens are compared for exact matches (after converting the alphabetic characters to lower case characters); numeric tokens are compared numerically for exact matches (after having been converted to their numeric value); and alphabetic characters are compared and scored using an edit distance algorithm” (Thirumalai, column 9, paragraph 6)
Thirumalai relates to grouping token features based on edit distance similarity metrics and is analogous to the claimed invention. The existing combination teaches a system that groups tokens by measuring a similarity score for each pair using a neural network. The claimed invention differs from this method by using edit distance, distribution representations, or related information of feature vectors to calculate the score. Thirumalai teaches a system that calculates a similarity score for tokens using edit distance. Because both the existing combination and Thirumalai teach the use of algorithmically matching similar tokens using a quantitative scoring metric, it would have been obvious to a person of ordinary skill in the art to substitute the existing combination’s neural network scoring for Thirumalai’s edit distance scoring to achieve the predictable result of grouping tokens together by proximity of textual construction / transformation (MPEP 2143 I. (B) Substituting one known element for another for predictable results).
Claims 20 and 23-24 are rejected under 35 U.S.C. 103 as being unpatentable over Fisher et al. (Merge and Label: A novel neural network architecture for nested NER, published 2019, Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp. 5840-5850), hereafter referred to as Fisher, in view of Kumar et al. (SCALABLE GENERATION OF MULTIDIMENSIONAL FEATURES FOR MACHINE LEARNING, filed 4/19/2016, US 11,295,229 B1), hereafter referred to as Kumar, and further in view of Mehanna et al. (SELECTION AND MODIFICATION OF FEATURES USED BY ONE OR MORE MACHINE LEARNED MODELS USED BY AN ONLINE SYSTEM, published 3/31/2016, US 2016/0092786 A1), hereafter referred to as Mehanna.
Regarding claim 20, the rejection of claim 18 is incorporated. While the aforementioned references fail to disclose the further limitations of the claim, Mehanna discloses a instructions, wherein the processor is configured to, in the training processing, swap a combination of the embedding vectors having a similarity equal to or higher than a predetermined second similarity with a predetermined probability: “Additional information may also be used by the machine learning module 230 to generate intermediate groups (combinations of the vectors). For example, a measure of correlation (similarity) between different features may be determined, and additional features having greater than a threshold (predetermined second similarity) measure of correlation to a feature in the set may be excluded (swapped) from inclusion in a group (combination of the vectors) including the feature.” (Mehanna, [0032]). Feature vectors with correlations higher than the threshold are excluded from groups with a predetermined probability of 1.
Mehanna relates to automatic feature selection for machine learning models and is analogous to the claimed invention. The existing combination teaches a device capable of combining feature vectors together. The claimed invention improves upon this method by merging vectors in response to the similarity of prediction results. Mehanna teaches a system able to discard features significant correlations with other features from groupings, applicable to the existing combination. A person of ordinary skill in the art would have recognized that filtering highly correlated features out of feature groups would lead to the predictable result of utilizing a smaller subset of features to attain similar predictive results from the model, and would improve the known device by increasing decreasing model training complexity and length (MPEP 2143 I. (D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results).
Regarding claim 23, the rejection of claim 18 is incorporated. While the aforementioned references fail to disclose the further limitations of the claim, Mehanna discloses a device, wherein the processor is configured to, in the training processing, merge combinations of the second feature vectors that correspond to combinations of the embedding vectors in a case where a change value of a prediction result of the machine learning model in a case where the combination of the embedding vectors is swapped is lower than a predetermined seventh threshold value:
“When generating the groups of features (combination[s] of the feature vectors), the machine learning module 230 determines a measure of feature impact for the machine learned model associated with each feature in the set. A measure of feature impact associated with a feature provides a measure of the feature's importance to the machine learned model. In one embodiment, a measure of feature impact is proportional to the importance of the feature to the machine learned model, so features that are more important to the machine learned model (e.g., features resulting in a larger change to an error term (prediction result) of the machine learned model if removed (swapped) from the machine learned model) are associated with higher measures of feature impact” (Mehanna, [0030])
“Based on the measures of feature impact associated with features from the set of features, the online system 140 generates 320 various groups (merge[d] combinations of the second feature vectors) each including different features from the set of features. For example, different groups include different numbers of features or include different features. In various embodiments, the online system 140 generates 320 a group including each feature form the set of features, a group including features originally received by the machine learned model, and one or more intermediate groups each having different numbers of features from the set of features. The online system 140 may associate an intermediate group (combination of the vectors) with a threshold measure of feature impact, so the intermediate group includes features having measures of feature impact equaling or exceeding the threshold measure of feature impact but does not include features having measures of feature impact less than the threshold measure of feature impact” (Mehanna, [0039])
Mehanna relates to automatic feature selection for machine learning models and is analogous to the claimed invention. The existing combination teaches a device capable of combining feature vectors together. The claimed invention improves upon this method by merging vectors in response to a change in prediction results. Mehanna teaches a system able to discard features with significant negative performance impacts from groupings, applicable to the existing combination. A person of ordinary skill in the art would have recognized that filtering negative impact features out of feature groups would lead to the predictable result of utilizing only a subset of the best-performing features to train the model, and would improve the known device by increasing model performance (MPEP 2143 I. (D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results).
Regarding claim 24, the rejection of claim 18 is incorporated. While the aforementioned references fail to disclose the further limitations of the claim, Mehanna discloses a device, wherein the processor is configured to, in the training processing, merge combinations of the second feature vectors that correspond to combinations of the embedding vectors in a case where a similarity of a prediction result of the machine learning model in a case where the combination of the embedding vectors is swapped is equal to or higher than a predetermined fifth threshold value: “Additional information may also be used by the machine learning module 230 to generate intermediate groups (merge[d] combinations of the second feature vectors). For example, a measure of correlation (similarity) between different features may be determined, and additional features having greater than a threshold measure of correlation to a feature in the set may be excluded from inclusion in a group (combination of the vectors) including the feature.” (Mehanna, [0032]). As would be known by one of ordinary skill in the art, the addition of highly correlated features to a machine learning model result in similar prediction result[s]. Thus, features that would lead to similar prediction results are being culled from these groups.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Dvijotham et al. (ROBUSTNESS TO ADVERSARIAL BEHAVIOR FOR TEXT CLASSIFICATION MODELS, filed 4/23/2021, US 2021/0334459 A1) discloses a method of forming combined feature representations in embedding space for training a machine learning system.
Best et al. (OPTIMIZING DATA REDUCTION, SECURITY AND ENCRYPTION REQUIREMENTS IN A NETWORK ENVIRONMENT, published 8/23/2018, US 20180241777 A1) discloses a method of selectively combinations of features for training a machine learning model such that combinations don’t result in a significant reduction in performance.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Aaron P Gormley whose telephone number is (571)272-1372. The examiner can normally be reached Monday - Friday 12:00 PM - 8:00 PM EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Michelle T Bechtold can be reached at (571) 431-0762. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/AG/Examiner, Art Unit 2148 /MICHELLE T BECHTOLD/Supervisory Patent Examiner, Art Unit 2148
Prosecution Insights