DETAILED ACTION
This correspondence is responsive to the application filed on July 15, 2024. Claims 1-14 are pending in the case with claims 1 and 8 in independent form.
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 .
Priority
Applicant’s claim for the benefit of a prior-filed application under 35 U.S.C. 119(e) or under 35 U.S.C. 120, 121, 365(c), or 386(c) is acknowledged.
Summary of Detailed Action
Claims 1-14 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Claims 1-3 and 8-10 are rejected under 35 U.S.C. 103 as being unpatentable over Kurata et al. in view of Hoffmeister and He et al.
Claims 4 and 11 are rejected under 35 U.S.C. 103 as being unpatentable over Kurata in view of Hoffmeister and He, and further in view of Huang et al.
Claims 5, 7, 12 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Kurata in view of Hoffmeister and He, and further in view of Gao et al.
Claims 6 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Kurata in view of Hoffmeister and He, and further in view of Purnell et al.
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 1-14 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claim(s) recite(s) subject matter a general, high-level of a method to identify one or more matching strings, from a set of strings, that match a query string, comparing the query string to one or more respective strings of the set of strings, based on said comparing, identifying the one or more matching strings, which are mental processes or concepts that can be performed in the human mind, including observation, evaluation, judgment or opinion, or by a human using pen and paper. MPEP 210604(a)(2)(III). This judicial exception is not integrated into a practical application and the claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claims 1-14 recite one of the four statutory categories of patent able subject matter and belong to the statutory class(es) of a process (method claims 1-7), a machine (system/apparatus claims ), and an article of manufacture (non-transitory computer readable media claims 8-14).
Claim 1 recites a method, thus a process and one of the four statutory categories of patentable subject matter. However, claim 1 further recites to identify one or more matching strings, from a set of strings, that match a query string, comparing the query string to one or more respective strings of the set of strings, based on said comparing, identifying the one or more matching strings, which are mental processes or concepts that can be performed in the human mind, including observation, evaluation, judgment or opinion, or by a human using pen and paper. MPEP 210604(a)(2)(III).
The claim does not include any additional elements which integrate the abstract idea into a practical application since the additional elements consist of:
A computer-executed (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer a vector representation output that represents a string input; (This additional element amounts to merely the words to “apply it” (or an equivalent) or are mere instructions to implement an abstract idea or other exception on a computer. MPEP 2106.05(f).) Also, this additional element amounts to no more than generally linking the use of the judicial exception to a particular technologic environment or field of use - The application or use of the judicial exception in this manner does not meaningfully limit the claim by going beyond generally linking the use of the judicial exception to a particular technological environment. MPEP 2106.05(h)).
executing a query (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
wherein executing the query comprises:(an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; (This additional element amounts to merely the words to “apply it” (or an equivalent) or are mere instructions to implement an abstract idea or other exception on a computer. MPEP 2106.05(f).) Also, this additional element amounts to no more than generally linking the use of the judicial exception to a particular technologic environment or field of use - The application or use of the judicial exception in this manner does not meaningfully limit the claim by going beyond generally linking the use of the judicial exception to a particular technological environment. MPEP 2106.05(h)).
the vector representation of(an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
one or more vector representations of (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
returning results from said executing the query; (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
wherein the method is performed by one or more computing devices. (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
Thus, the claim is directed to the abstract idea.
Further, the additional elements, alone or in combination, do not provide significantly more than the abstract idea itself, because implementation on a computer (MPEP 2106.05(f)) cannot provide significantly more, and generally linking the use of the judicial exception to a particular technological field of use does not meaningfully limit the claims (MPEP 2106.04(d)) and the combination of additional elements does not provide an inventive concept. Thus, the claim is ineligible.
Claim 2, dependent on claim 1, does not include any additional elements which integrate the abstract idea into a practical application since the additional elements consist of:
wherein applying the trained RNN model comprises inferring the vector representation of the query string based on n-grams of a plurality of words in the query string. (This additional element amounts to merely the words to “apply it” (or an equivalent) or are mere instructions to implement an abstract idea or other exception on a computer. MPEP 2106.05(f).) Also, this additional element amounts to no more than generally linking the use of the judicial exception to a particular technologic environment or field of use - The application or use of the judicial exception in this manner does not meaningfully limit the claim by going beyond generally linking the use of the judicial exception to a particular technological environment. MPEP 2106.05(h)).
Claim 3, dependent on claim 2, does not include any additional elements which integrate the abstract idea into a practical application since the additional elements consist of:
wherein applying the trained RNN model comprises identifying a vector representation of each word, of the plurality of words, using a gated recurrent unit (GRU) operating over embeddings of the n-grams of said each word to produce a plurality of word vector representations. (This additional element amounts to merely the words to “apply it” (or an equivalent) or are mere instructions to implement an abstract idea or other exception on a computer. MPEP 2106.05(f).) Also, this additional element amounts to no more than generally linking the use of the judicial exception to a particular technologic environment or field of use - The application or use of the judicial exception in this manner does not meaningfully limit the claim by going beyond generally linking the use of the judicial exception to a particular technological environment. MPEP 2106.05(h)).
Claim 4, dependent on claim 3, does not include any additional elements which integrate the abstract idea into a practical application since the additional elements consist of:
wherein applying the trained RNN model comprises producing the vector representation of the query string based on a dense layer that generates the vector representation of the query string based on the plurality of word vector representations. (This additional element amounts to merely the words to “apply it” (or an equivalent) or are mere instructions to implement an abstract idea or other exception on a computer. MPEP 2106.05(f).) Also, this additional element amounts to no more than generally linking the use of the judicial exception to a particular technologic environment or field of use - The application or use of the judicial exception in this manner does not meaningfully limit the claim by going beyond generally linking the use of the judicial exception to a particular technological environment. MPEP 2106.05(h)).
Claim 5, dependent on claim 1, recites additional abstract subject matter for wherein comparing the query string to the one or more respective strings of the set of strings comprises: comparing the query string to a particular string of the set of strings, which are mental processes or concepts that can be performed in the human mind, including observation, evaluation, judgment or opinion, or by a human using pen and paper. MPEP 210604(a)(2)(III).
The claim does not include any additional elements which integrate the abstract idea into a practical application since the additional elements consist of:
the vector representation of (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
the one or more vector representations of (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
the vector representation of (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
a particular vector representation of (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
by producing a dot product of the vector representation of the query string and the particular vector representation of the particular string. (This additional element amounts to merely the words to “apply it” (or an equivalent) or are mere instructions to implement an abstract idea or other exception on a computer. MPEP 2106.05(f).) Also, this additional element amounts to no more than generally linking the use of the judicial exception to a particular technologic environment or field of use - The application or use of the judicial exception in this manner does not meaningfully limit the claim by going beyond generally linking the use of the judicial exception to a particular technological environment. MPEP 2106.05(h)).
Claim 6, dependent on claim 1, does not include any additional elements which integrate the abstract idea into a practical application since the additional elements consist of:
wherein the training data set is a first training data set; (This additional element amounts to merely the words to “apply it” (or an equivalent) or are mere instructions to implement an abstract idea or other exception on a computer. MPEP 2106.05(f).) Also, this additional element amounts to no more than generally linking the use of the judicial exception to a particular technologic environment or field of use - The application or use of the judicial exception in this manner does not meaningfully limit the claim by going beyond generally linking the use of the judicial exception to a particular technological environment. MPEP 2106.05(h)).
generating a second training data set based on user feedback; (This additional element amounts to merely the words to “apply it” (or an equivalent) or are mere instructions to implement an abstract idea or other exception on a computer. MPEP 2106.05(f).) Also, this additional element amounts to no more than generally linking the use of the judicial exception to a particular technologic environment or field of use - The application or use of the judicial exception in this manner does not meaningfully limit the claim by going beyond generally linking the use of the judicial exception to a particular technological environment. MPEP 2106.05(h)).
generating a second training data set based on user feedback; (This additional element amounts to merely the words to “apply it” (or an equivalent) or are mere instructions to implement an abstract idea or other exception on a computer. MPEP 2106.05(f).) Also, this additional element amounts to no more than generally linking the use of the judicial exception to a particular technologic environment or field of use - The application or use of the judicial exception in this manner does not meaningfully limit the claim by going beyond generally linking the use of the judicial exception to a particular technological environment. MPEP 2106.05(h)).
after applying the trained RNN model against the query string, training the trained RNN model based on the second training data set to produce a second trained RNN model. (This additional element amounts to merely the words to “apply it” (or an equivalent) or are mere instructions to implement an abstract idea or other exception on a computer. MPEP 2106.05(f).) Also, this additional element amounts to no more than generally linking the use of the judicial exception to a particular technologic environment or field of use - The application or use of the judicial exception in this manner does not meaningfully limit the claim by going beyond generally linking the use of the judicial exception to a particular technological environment. MPEP 2106.05(h)).
Claim 7, dependent on claim 1, recites only additional abstract subject matter for
wherein the set of strings comprises name strings and wherein the query string is a name string which are mental processes or concepts that can be performed in the human mind, including observation, evaluation, judgment or opinion, or by a human using pen and paper. MPEP 210604(a)(2)(III).
Claim 8 recites a computer-readable media, thus an article of manufacture and one of the four statutory categories of patentable subject matter. However, claim 8 further recites to identify one or more matching strings, from a set of strings, that match a query string; comparing the query string to one or more respective strings of the set of strings; based on said comparing, identifying the one or more matching strings, which are mental processes or concepts that can be performed in the human mind, including observation, evaluation, judgment or opinion, or by a human using pen and paper. MPEP 210604(a)(2)(III).
The claim does not include any additional elements which integrate the abstract idea into a practical application since the additional elements consist of:
One or more non-transitory computer-readable media storing one or more sequences of instructions that, when executed by one or more processors, cause (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer a vector representation output that represents a string input; (This additional element amounts to merely the words to “apply it” (or an equivalent) or are mere instructions to implement an abstract idea or other exception on a computer. MPEP 2106.05(f).) Also, this additional element amounts to no more than generally linking the use of the judicial exception to a particular technologic environment or field of use - The application or use of the judicial exception in this manner does not meaningfully limit the claim by going beyond generally linking the use of the judicial exception to a particular technological environment. MPEP 2106.05(h)).
executing a query (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
wherein executing the query comprises: (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; (This additional element amounts to merely the words to “apply it” (or an equivalent) or are mere instructions to implement an abstract idea or other exception on a computer. MPEP 2106.05(f).) Also, this additional element amounts to no more than generally linking the use of the judicial exception to a particular technologic environment or field of use - The application or use of the judicial exception in this manner does not meaningfully limit the claim by going beyond generally linking the use of the judicial exception to a particular technological environment. MPEP 2106.05(h)).
the vector representation of (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
one or more vector representations of (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
returning results from said executing the query. (an additional element merely recites the words “apply it” (or an equivalent) with the judicial exception, or merely includes instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See also, MPEP 2106.05(f), MPEP 2106.04(d), 2019 Guidance, 84 FR 50 at 55, footnote 30.).
Thus, the claim is directed to the abstract idea.
Further, the additional elements, alone or in combination, do not provide significantly more than the abstract idea itself, because implementation on a computer (MPEP 2106.05(f)) cannot provide significantly more, and generally linking the use of the judicial exception to a particular technological field of use does not meaningfully limit the claims (MPEP 2106.04(d)) and the combination of additional elements does not provide an inventive concept. Thus, the claim is ineligible.
Dependent claims 9-14 are comparably rejected as set forth above with respect to dependent claims 2-7.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1-3 and 8-10 are rejected under 35 U.S.C. 103 as being unpatentable over Kurata et al. (Pub. No. US 2018/0082167 A1, published March 22, 2018) hereinafter Kurata in view of Hoffmeister (US Patent No. 10,332,508, issued June 25, 2019) and He et al. (Pub. No. US 2019/0370398 A1, published December 5, 2019) hereinafter He.
Regarding claim 1, Kurata/Huang teaches:
A computer-executed method comprising (i.e., a method is provided for processing a Recurrent Neural Network (RNN)… by a hardware processor,.. According to another embodiment of the present invention, an apparatus and a computer program product are provided for implementing the above method. Kurata, Figs 1, 10, para 3.):
training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer a vector representation output that represents a string input;
Kurata teaches that, RNN is used for text processing. For example, a text output from a speech recognition system can be input to the RNN. Although the speech recognition system can output not only a 1-best result but candidates of results of the system, the RNN processes a single input, usually the 1-best result from the speech recognition system. Kurata, para 2, 16, 18. The training section 160 can train the RNN based on the training data (training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer . In an embodiment, the training section 160 can train the RNN such that an error between the candidate data and the correct output is reduced. The training section 160 can utilize results of processing of the calculating section 120, the merging section 130, and the outputting section 150 for the training. Kurata, Figs 1-2, 5, para 23, 16,18, 30, 38, 69-70, 72. [0030] In some embodiments, the RNN can input text data including a plurality of candidates and stochastic information of the candidates and output an answer of a slot filling problem, an answer of key word spotting, and/or a translated text (training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer . Kurata, Figs 1-2, 5, 9, para 30, 23, 16,18, 30, 38, 69-70.
As discussed above, Kurata teaches training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer output that represents a string input. Kurata does not specifically disclose a vector representation.
However, Hoffmeister teaches in the field related to speech recognition. Hoffmeister, col 1:5-23. Hoffmeister, which is analogous to the claimed invention because Hoffmeister is directed to speech and text processing and using RNN, teaches that, Text for a search result may be encoded (1376) (for example using an RNN Encoder 950) to create an encoded feature vector 910 representing the text (a vector representation) of the search result. For example, an encoder 950 may be used to encode each sentence of a search result into a feature vector. Hoffmeister, Fig 9, 13, col 26:5-44; col 15:46-62.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement the method and system for training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer output that represents a string input of Kurata using the vector representation of Hoffmeister, with a reasonable expectation of success, in order to provide speech processing that may also convert a user's speech into text data which may then be provided to various text-based software applications. Hoffmeister, col 1:5-23, col 26:5-44. This would have provided the advantages of providing data for use text-based applications.
executing a query to identify one or more matching strings, from a set of strings, that match a query string; wherein executing the query comprises: applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; comparing the vector representation of the query string to one or more vector representations of one or more respective strings of the set of strings; based on said comparing, identifying the one or more matching strings; returning results from said executing the query; wherein the method is performed by one or more computing devices.
Kurata teaches that a method is provided for processing a Recurrent Neural Network (RNN)… by a hardware processor,.. According to another embodiment of the present invention, an apparatus and a computer program product are provided for implementing the above method (wherein the method is performed by one or more computing devices). Kurata, Figs 1, 10, para 3.
Kurata in view of Hoffmeister does not specifically disclose executing a query to identify one or more matching strings, from a set of strings, that match a query string; wherein executing the query comprises: applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; comparing the vector representation of the query string to one or more vector representations of one or more respective strings of the set of strings; based on said comparing, identifying the one or more matching strings; returning results from said executing the query.
However, He teaches in the field related to elates to natural language processing in human-machine interaction, in particular, to methods and apparatus for searching historical data based on natural language understanding. He, para 1. He, which is analogous to the claimed invention because He is directed to natural language processing in human interactions, teaches that, [0059] At block 51, if two or more sub-classification predictions exceed the threshold (e.g., when the audio input is “navigate home and play music” which corresponds to two intents) (executing a query to identify one or more matching strings (audio query separated into input text string query request intent including to find, identify, match and retrieve request intent information), from a set of strings, that match a query string (from set of database intent strings that match a query string); wherein executing the query comprises wherein executing the query comprises:), the flowchart may proceed to block 52, where a neural network (e.g., feedforward neural network (FNN), recurrent neural network (RNN)) model may be applied to (wherein executing the query comprises: applying the trained RNN model)(1: following from block 51) separate the corresponding input texts into various text strings based on the multiple sub-classification predictions (executing a query (audio query request intent including to find, match and retrieve request intent information which is separated into input text string query) to identify one or more matching strings, from a set of strings, that match a query string; wherein executing the query comprises wherein executing the query comprises:)and/or (2: following from block 50) extract a sub-classification prediction. If just one sub-classification prediction exceeds the threshold, after the multiple sub-classification predictions are separated, or after the sub-classification prediction is extracted, the flowchart may proceed to block 53 where a N-gram model may be applied to convert the each text string (which corresponds to the sub-classification prediction) for approximate matching. By converting the sequence of text strings to a set of N-grams, the sequence can be embedded in a vector space, thus allowing the sequence to be compared to other sequences (e.g., preset intentions) in an efficient manner (applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; comparing the vector representation of the query string to one or more vector representations of one or more respective strings of the set of strings; based on said comparing, identifying the one or more matching strings; returning results from said executing the query). Accordingly, at block 54, the converted set of N-grams (corresponding to the sub-classification prediction) may be compared against an intent database to obtain a matching intent in the intent database (based on said comparing, identifying the one or more matching strings; returning results from said executing the query). The matching intent(s) may be obtained as the sub-classification(s) of the audio input at block 57. He, Figs 3A-3C, 4, 5, para 59, 60-61.[0061] If the intent match is unsuccessful at block 54, a feedforward neural network model may be applied at block 55. At block 56, the outputs of the block 49 and the block 55 may be compared. If the two outputs are the same, the flowchart may proceed to block 57(based on said comparing, identifying the one or more matching strings; returning results from said executing the query); otherwise, the second machine learning model group 326 may render output 303e (e.g., a fail message) The naive bayes model, the TF-IDF model, the N-gram model, the FNN, and the RNN, and their training are incorporated herein by reference. Based on the context of the current interface, one or more weights of the class associated with the user's intention in the each machine learning model can be dynamically adjusted, thus improving the accuracy of the classification. He, Figs 3A-3C, 4, 5, para 61, 59, 60.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement the method and system for training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer output that represents a string input of Kurata using the vector representation of Hoffmeister and the executing a query to identify one or more matching strings, from a set of strings, that match a query string; wherein executing the query comprises: applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; comparing the vector representation of the query string to one or more vector representations of one or more respective strings of the set of strings; based on said comparing, identifying the one or more matching strings; returning results from said executing the query of He, with a reasonable expectation of success, in order to provide speech processing that may also convert a user's speech into text data which may then be provided to various text-based software applications and in order to provide human-machine interactions that allow people to use their voices to effectuate control and help effectuate the control without any physical or visual contact with the device. Hoffmeister, col 1:5-23, col 26:5-44. He, para 2. This would have provided the advantages of providing data for use text-based applications and improving interactions with applications.
Regarding claim 2, which depends from claim 1 and recites:
wherein applying the trained RNN model comprises inferring the vector representation of the query string based on n-grams of a plurality of words in the query string.
Kurata in view of Hoffmeister and He teaches the method of claim 1 from which claim 2 depends, including applying the trained RNN model comprises inferring the vector representation of the query string. Kurata in view of Hoffmeister does not specifically disclose based on n-grams of a plurality of words in the query string.
However, He teaches that, At block 51, if two or more sub-classification predictions exceed the threshold (e.g., when the audio input is “navigate home and play music” which corresponds to two intents), the flowchart may proceed to block 52, where a neural network (e.g., feedforward neural network (FNN), recurrent neural network (RNN)) model may be applied to (applying the trained RNN model) (1: following from block 51) separate the corresponding input texts into various text strings based on the multiple sub-classification predictions and/or (2: following from block 50) extract a sub-classification prediction. If just one sub-classification prediction exceeds the threshold, after the multiple sub-classification predictions are separated, or after the sub-classification prediction is extracted, the flowchart may proceed to block 53 where a N-gram model may be applied to convert the each text string (which corresponds to the sub-classification prediction) for approximate matching. By converting the sequence of text strings to a set of N-grams, the sequence can be embedded in a vector space, thus allowing the sequence to be compared to other sequences (e.g., preset intentions) in an efficient manner (applying the trained RNN model comprises inferring the vector representation of the query string based on n-grams of a plurality of words in the query string). Accordingly, at block 54, the converted set of N-grams (corresponding to the sub-classification prediction) may be compared against an intent database to obtain a matching intent in the intent database. The matching intent(s) may be obtained as the sub-classification(s) of the audio input at block 57. He, Figs 3A-3C, 4, 5, para 59, 60-61.[0061] If the intent match is unsuccessful at block 54, a feedforward neural network model may be applied at block 55. At block 56, the outputs of the block 49 and the block 55 may be compared. If the two outputs are the same, the flowchart may proceed to block 57; otherwise, the second machine learning model group 326 may render output 303e (e.g., a fail message) The naive bayes model, the TF-IDF model, the N-gram model, the FNN, and the RNN, and their training are incorporated herein by reference. Based on the context of the current interface, one or more weights of the class associated with the user's intention in the each machine learning model can be dynamically adjusted, thus improving the accuracy of the classification. He, Figs 3A-3C, 4, 5, para 61, 59, 60.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement the method and system for training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer output that represents a string input of Kurata using the vector representation of Hoffmeister and the executing a query to identify one or more matching strings, from a set of strings, that match a query string; wherein executing the query comprises: applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; comparing the vector representation of the query string to one or more vector representations of one or more respective strings of the set of strings; based on said comparing, identifying the one or more matching strings; returning results from said executing the query and applying the trained RNN model comprises inferring the vector representation of the query string based on n-grams of a plurality of words in the query string of He, with a reasonable expectation of success, in order to provide speech processing that may also convert a user's speech into text data which may then be provided to various text-based software applications and in order to provide human-machine interactions that allow people to use their voices to effectuate control and help effectuate the control without any physical or visual contact with the device. Hoffmeister, col 1:5-23, col 26:5-44. He, para 2. This would have provided the advantages of providing data for use text-based applications and improving interactions with applications.
Regarding claim 3, which depends from claim 2 and recites:
wherein applying the trained RNN model comprises identifying a vector representation of each word, of the plurality of words, using a gated recurrent unit (GRU) operating over embeddings of the n-grams of said each word to produce a plurality of word vector representations.
Kurata in view of Hoffmeister and He teaches the method of claim 2 from which claim 3 depends, including wherein applying the trained RNN model comprises inferring the vector representation of the query string based on n-grams of a plurality of words in the query string. Kurata teaches that, the calculating section can calculate a temporal next state of a recurrent layer in the RNN, by inputting candidate selected at S120 and a current state before inputting the candidate to the RNN. In an embodiment, the calculating section can calculate the temporal next state of a recurrent layer in LSTM. In an embodiment, the calculating section can adopt at least one of a variety of types of LSTM (e.g., LSTM described in Gers & Schmidhuber (2000), Cho, et al. (2014), Koutnik, et al. (2014), Yao et al. (2015), Greff, et al. (2015), or Jozefowicz, et al (2015). In some embodiments, the calculating section can adopt GRU (applying the trained RNN model comprises using a gated recurrent unit (GRU) operating over each word (each word, see also para 42, 30, 36-37, 53, 57)) instead of LSTM, as described by Junyoung Chung, Caglar Gulcehre, KyungHyun Cho, Yoshua Bengio, Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling which can be obtained at “https://arxiv.org/abs/1412.3555”. Kurata, Figs 5, 6, para 43, 42, 30, 36-37, 53, 57. Thus, Kurata teaches applying the trained RNN model comprises using a gated recurrent unit (GRU) operating over each word. Kurata in view of Hoffmeister does not specifically disclose identifying a vector representation of each word, of the plurality of words, operating over embeddings of the n-grams of said each word to produce a plurality of word vector representations.
However, He teaches that, At block 51, if two or more sub-classification predictions exceed the threshold (e.g., when the audio input is “navigate home and play music” which corresponds to two intents), the flowchart may proceed to block 52, where a neural network (e.g., feedforward neural network (FNN), recurrent neural network (RNN)) model may be applied to (applying the trained RNN model) (1: following from block 51) separate the corresponding input texts into various text strings based on the multiple sub-classification predictions and/or (2: following from block 50) extract a sub-classification prediction. If just one sub-classification prediction exceeds the threshold, after the multiple sub-classification predictions are separated, or after the sub-classification prediction is extracted, the flowchart may proceed to block 53 where a N-gram model may be applied to convert the each text string (which corresponds to the sub-classification prediction) for approximate matching. By converting the sequence of text strings to a set of N-grams, the sequence can be embedded in a vector space, thus allowing the sequence to be compared to other sequences (e.g., preset intentions) in an efficient manner (applying the trained RNN model comprises identifying a vector representation of each word of the plurality of words, operating over embeddings of the n-grams of said each word to produce a plurality of word vector representations). Accordingly, at block 54, the converted set of N-grams (corresponding to the sub-classification prediction) may be compared against an intent database to obtain a matching intent in the intent database. The matching intent(s) may be obtained as the sub-classification(s) of the audio input at block 57. He, Figs 3A-3C, 4, 5, para 59, 60-61.[0061] If the intent match is unsuccessful at block 54, a feedforward neural network model may be applied at block 55. At block 56, the outputs of the block 49 and the block 55 may be compared. If the two outputs are the same, the flowchart may proceed to block 57; otherwise, the second machine learning model group 326 may render output 303e (e.g., a fail message) The naive bayes model, the TF-IDF model, the N-gram model, the FNN, and the RNN, and their training are incorporated herein by reference. Based on the context of the current interface, one or more weights of the class associated with the user's intention in the each machine learning model can be dynamically adjusted, thus improving the accuracy of the classification. He, Figs 3A-3C, 4, 5, para 61, 59, 60.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement the method and system for training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer output that represents a string input and applying the trained RNN model comprises using a gated recurrent unit (GRU) operating over each word of Kurata using the vector representation of Hoffmeister and the executing a query to identify one or more matching strings, from a set of strings, that match a query string; wherein executing the query comprises: applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; comparing the vector representation of the query string to one or more vector representations of one or more respective strings of the set of strings; based on said comparing, identifying the one or more matching strings; returning results from said executing the query and applying the trained RNN model comprises inferring the vector representation of the query string based on n-grams of a plurality of words in the query string and identifying a vector representation of each word, of the plurality of words, operating over embeddings of the n-grams of said each word to produce a plurality of word vector representations of He, with a reasonable expectation of success, in order to provide speech processing that may also convert a user's speech into text data which may then be provided to various text-based software applications and in order to provide human-machine interactions that allow people to use their voices to effectuate control and help effectuate the control without any physical or visual contact with the device. Hoffmeister, col 1:5-23, col 26:5-44. He, para 2. This would have provided the advantages of providing data for use text-based applications and improving interactions with applications.
Claims 8-10 recite one or more non-transitory computer-readable media that parallel the method of claims 1-3, respectively. Therefore, the analysis discussed above with respect to claims 1-3 also applies to claims 8-10, respectively. Accordingly, claims 8-10 are rejected based substantially the same rationale set forth above with respect to claims 1-3, respectively. More specifically regarding One or more non-transitory computer-readable media storing one or more sequences of instructions that, when executed by one or more processors, (i.e., Kurata, Figs 1, 10, para 16-17,95-101.).
Claim(s) 4 and 11 are rejected under 35 U.S.C. 103 as being unpatentable over Kurata in view of Hoffmeister and He as applied to claims 3 and 10 above, and further in view of Huang et al. (Patent No. US 10,431,210 B1, published October 1, 2019) hereinafter Huang.
Regarding claim 4, which depends from claim 3 and recites:
wherein applying the trained RNN model comprises producing the vector representation of the query string based on a dense layer that generates the vector representation of the query string based on the plurality of word vector representations.
Kurata in view of Hoffmeister and He teaches the method of claim 3 from which claim 4 depends, including applying the trained RNN model comprises producing a vector representation of the query string and identifying a vector representation of each word, of the plurality of words, using a gated recurrent unit (GRU) operating over embeddings of the n-grams of said each word to produce a plurality of word vector representations of the query string. Kurata in view of Hoffmeister and He does not specifically disclose based on a dense layer.
However, Huang teaches in the field related to computing systems and more particularly to implementing a whole sentence a recurrent neural network language model for natural language processing. Huang, col 1:10-13. Huang, which is analogous to the claimed invention because Huang is directed to natural language processing using LSTM RNNs, teaches that, In one example, in a one layer BiLSTM 700, an LSTM layer 730 may be loaded once from beginning to end and once from end to beginning, which may increase the speed at which BiLSTM learns a sequential task in comparison with a one directional LSTM. For example, BiLSTM 700 may receive each of inputs w.sub.1, w.sub.2, . . . , w.sub.T 710 at an embedding layer 720, with an embedding node for each word w. In one example, each word is loaded through the embedding layer to two LSTM within LSTM layer 730, one at the beginning of a loop and one at the end of a loop. In one example, the first and last LSTM outputs from LSTM layer 730 may feed forward outputs to a concatenation layer 740. In one example, concatenation layer 740 may represent a layer of NN scorer 420. Concatenation layer 740 may concatenate the outputs, providing double the number of outputs to a next fully connected (FC) 742 (based on a dense layer (outputs based on fully connected dense layer)). FC 742 obtains the final score of the sentence (based on a dense layer (outputs based on fully connected dense layer)). In one example, BiLSTM 700 may include additional or alternate sizes of embedding layer 720 and LSTM layer 730, such as include an embedding size of two hundred in embedding layer 720, with seven hundred hidden LSTM units in LSTM layer 730. While in the example, concatenation layer 740 is illustrated receiving the first and last LSTM outputs from LSTM layer 730 and concatenating the outputs, in additional or alternate examples, concatenation layer 740 may receive additional LSTM outputs and in additional or alternate examples, concatenation layer 740 may be replaced by an alternative NN scoring layer that applies one or more scoring functions to multiple outputs from LSTM layer 730. Huang, col 11:44-col:12:5.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement the method and system for training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer output that represents a string input and applying the trained RNN model comprises using a gated recurrent unit (GRU) operating over each word of Kurata using the vector representation of Hoffmeister and the executing a query to identify one or more matching strings, from a set of strings, that match a query string; wherein executing the query comprises: applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; comparing the vector representation of the query string to one or more vector representations of one or more respective strings of the set of strings; based on said comparing, identifying the one or more matching strings; returning results from said executing the query and applying the trained RNN model comprises inferring the vector representation of the query string based on n-grams of a plurality of words in the query string and identifying a vector representation of each word, of the plurality of words, operating over embeddings of the n-grams of said each word to produce a plurality of word vector representations of He and the outputs based on fully connected dense layer of Huang, with a reasonable expectation of success, in order to provide speech processing that may also convert a user's speech into text data which may then be provided to various text-based software applications and in order to provide human-machine interactions that allow people to use their voices to effectuate control and help effectuate the control without any physical or visual contact with the device and in order to facilitating use of RNNs for processing tasks where prior inputs need to be considered, such as natural language processing (NLP) tasks and processing a whole sentence. Hoffmeister, col 1:5-23, col 26:5-44. He, para 2. Huang, col 1:17-25, col 11:44-col:12:5. This would have provided the advantages of providing data for use text-based applications and improving interactions with applications and facilitating natural language processing tasks.
Claim 11 recites one or more non-transitory computer-readable media that parallels the method of claim 4. Therefore, the analysis discussed above with respect to claim 4 also applies to claim 11. Accordingly, claim 11 is rejected based substantially the same rationale set forth above with respect to claim 4.
Claim(s) 5, 7, 12 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Kurata in view of Hoffmeister and He as applied to claims 1 and 8 above, and further in view of Gao et al. (Pub. No. US 2020/0327136 A1, provisional application filed April 14, 2019) hereinafter Gao.
Regarding claim 5, which depends from claim 1 and recites:
wherein comparing the vector representation of the query string to the one or more vector representations of the one or more respective strings of the set of strings comprises: comparing the vector representation of the query string to a particular vector representation of a particular string, of the set of strings, by producing a dot product of the vector representation of the query string and the particular vector representation of the particular string.
Kurata in view of Hoffmeister and He teaches the method of claim 1 from which claim 5 depends, including comparing the vector representation of the query string to the one or more vector representations of the one or more respective strings of the set of strings comprises. Kurata does not specifically disclose comparing the vector representation of the query string to a particular vector representation of a particular string, of the set of strings.
However, He teaches that, At block 51, if two or more sub-classification predictions exceed the threshold (e.g., when the audio input is “navigate home and play music” which corresponds to two intents), the flowchart may proceed to block 52, where a neural network (e.g., feedforward neural network (FNN), recurrent neural network (RNN)) model may be applied to (applying the trained RNN model) (1: following from block 51) separate the corresponding input texts into various text strings based on the multiple sub-classification predictions and/or (2: following from block 50) extract a sub-classification prediction. If just one sub-classification prediction exceeds the threshold, after the multiple sub-classification predictions are separated, or after the sub-classification prediction is extracted, the flowchart may proceed to block 53 where a N-gram model may be applied to convert the each text string (which corresponds to the sub-classification prediction) for approximate matching. By converting the sequence of text strings to a set of N-grams, the sequence can be embedded in a vector space, thus allowing the sequence to be compared to other sequences (e.g., preset intentions) in an efficient manner (wherein comparing the vector representation of the query string to the one or more vector representations of the one or more respective strings of the set of strings comprises: comparing the vector representation of the query string to a particular vector representation of a particular string (to a particular other vector representation of a particular string), of the set of strings). Accordingly, at block 54, the converted set of N-grams (corresponding to the sub-classification prediction) may be compared against an intent database to obtain a matching intent in the intent database. The matching intent(s) may be obtained as the sub-classification(s) of the audio input at block 57. He, Figs 3A-3C, 4, 5, para 59, 60-61.[0061] If the intent match is unsuccessful at block 54, a feedforward neural network model may be applied at block 55. At block 56, the outputs of the block 49 and the block 55 may be compared. If the two outputs are the same, the flowchart may proceed to block 57; otherwise, the second machine learning model group 326 may render output 303e (e.g., a fail message) The naive bayes model, the TF-IDF model, the N-gram model, the FNN, and the RNN, and their training are incorporated herein by reference. Based on the context of the current interface, one or more weights of the class associated with the user's intention in the each machine learning model can be dynamically adjusted, thus improving the accuracy of the classification. He, Figs 3A-3C, 4, 5, para 61, 59, 60.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement the method and system for training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer output that represents a string input of Kurata using the vector representation of Hoffmeister and the executing a query to identify one or more matching strings, from a set of strings, that match a query string; wherein executing the query comprises: applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; comparing the vector representation of the query string to one or more vector representations of one or more respective strings of the set of strings; based on said comparing, identifying the one or more matching strings; returning results from said executing the query and wherein comparing the vector representation of the query string to the one or more vector representations of the one or more respective strings of the set of strings comprises: comparing the vector representation of the query string to a particular vector representation of a particular string of He, with a reasonable expectation of success, in order to provide speech processing that may also convert a user's speech into text data which may then be provided to various text-based software applications and in order to provide human-machine interactions that allow people to use their voices to effectuate control and help effectuate the control without any physical or visual contact with the device. Hoffmeister, col 1:5-23, col 26:5-44. He, para 2. This would have provided the advantages of providing data for use text-based applications and improving interactions with applications.
Thus, Kurata in view of Hoffmeister and He teaches the query string and the particular vector representation of the particular string and comparing the vector representation of the query string to the one or more vector representations of the one or more respective strings of the set of strings comprises: comparing the vector representation of the query string to a particular vector representation of a particular string of the set of strings.
Kurata in view of Hoffmeister and He does not specifically disclose comparing by producing a dot product of the string vector representations.
However, Gao teaches in the field related to information management technology and specifically to predictive analytics technology. Gao, para 2. Gao, which is analogous to the claimed invention because Gao is directed to automated name matching and comparing strings, teaches that, [0079] If the similarity scoring function ranges from 0 to 1 and the weighting functions are normalized over all tokens in the string, then this technique may be interpreted to be a cosine similarity between two vector representations, because score(t.sub.c.sup.i, t.sub.e.sup.j) is a multiplication over two comparable components (comparing by producing a dot product of the string vector representations), and then the sum over component multiplications behaves as a dot product operation. And of course, cosine similarity is equivalent to the dot product of two unit-length normalized vectors (comparing by producing a dot product of the string vector representations). However, the difference of this technique is that the component multiplications do not need to happen on exactly matching components, but rather on fuzzily matched components through the similarity function. Gao, para 79.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement the method and system for training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer output that represents a string input of Kurata using the vector representation of Hoffmeister and the executing a query to identify one or more matching strings, from a set of strings, that match a query string; wherein executing the query comprises: applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; comparing the vector representation of the query string to one or more vector representations of one or more respective strings of the set of strings; based on said comparing, identifying the one or more matching strings; returning results from said executing the query and wherein comparing the vector representation of the query string to the one or more vector representations of the one or more respective strings of the set of strings comprises: comparing the vector representation of the query string to a particular vector representation of a particular string of He and the comparing by producing a dot product of the string vector representations of Gao, with a reasonable expectation of success, in order to provide speech processing that may also convert a user's speech into text data which may then be provided to various text-based software applications and in order to provide human-machine interactions that allow people to use their voices to effectuate control and help effectuate the control without any physical or visual contact with the device and in order to automate matching names in noisy datasets. Hoffmeister, col 1:5-23, col 26:5-44. He, para 2. Gao, para 3-5, 70, 79. This would have provided the advantages of providing data for use text-based applications and improving interactions with applications and improving management of noisy names information in datasets.
Regarding claim 7, which depends from claim 1 and recites:
wherein the set of strings comprises name strings and wherein the query string is a name string.
Kurata in view of Hoffmeister and He teaches the method of claim 1 from which claim 7 depends, including the set of strings and query string. Kurata in view of Hoffmeister and He does not specifically disclose name strings.
However, Gao teaches that, For comparing string phrases, such as company relaxed names (a name string, comparing name strings), the company matching system implements a technique that takes into consideration both the importance of the individual words that make up the phrases, as well as the similarity between words in the case of inexact match, in essence allowing for fuzziness. Gao, para 70.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement the method and system for training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer output that represents a string input of Kurata using the vector representation of Hoffmeister and the executing a query to identify one or more matching strings, from a set of strings, that match a query string; wherein executing the query comprises: applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; comparing the vector representation of the query string to one or more vector representations of one or more respective strings of the set of strings; based on said comparing, identifying the one or more matching strings; returning results from said executing the query of He and a name string, comparing name strings of Gao, with a reasonable expectation of success, in order to provide speech processing that may also convert a user's speech into text data which may then be provided to various text-based software applications and in order to provide human-machine interactions that allow people to use their voices to effectuate control and help effectuate the control without any physical or visual contact with the device and in order to automate matching names in noisy datasets. Hoffmeister, col 1:5-23, col 26:5-44. He, para 2. Gao, para 3-5, 70.This would have provided the advantages of providing data for use text-based applications and improving interactions with applications and improving management of noisy names information in datasets.
Claims 12 and 14 recite one or more non-transitory computer-readable media that parallel the methods of claim 5 and 7, respectively. Therefore, the analysis discussed above with respect to claims 5 and 7 also applies to claims 12 and 14, respectively. Accordingly, claims 1 and 14 are rejected based substantially the same rationale set forth above with respect to claims 5 and 7, respectively.
Claim(s) 6 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Kurata in view of Hoffmeister and He as applied to claims 1 and 8 above, and further in view of Purnell et al. (Pub. No. US 2021/0004440 A1, filed July 2, 2019) hereinafter Purnell.
Regarding claim 6, which depends from claim 1 and further recites:
wherein the training data set is a first training data set; generating a second training data set based on user feedback; after applying the trained RNN model against the query string, training the trained RNN model based on the second training data set to produce a second trained RNN model.
Kurata in view of Hoffmeister and He teaches the method of claim 1 from which claim 6 depends, including training data set is a first training data set and after applying the trained RNN model against the query string. Kurata in view of Hoffmeister and He does not specifically disclose generating a second training data set based on user feedback, training the trained RNN model based on the second training data set to produce a second trained RNN model.
However, Purnell teaches in the field related to automated language processing. Purnell, para 1. Purnell, which is analogous to the claimed invention because Purnell is directed to natural language processing, machine learning models, word embedding, teaches that, In some cases, the server 505 may update the machine learning model performing the vector remapping. For example, this machine learning model may be an example of a supervised machine learning model, such as a CNN, an RNN (RNN model), an LSTM model, or any other type of machine learning model. The server 505 may receive feedback based on the output result (e.g., a user operating the user device 510 may tag the result with the correct output value) and may train the supervised machine learning model based on the tag (generating a second training data set based on user feedback, training the trained RNN model based on the second training data set to produce a second trained RNN model). Additionally or alternatively, the machine learning model may be updated to handle additional vocabularies of interest, additional languages, additional scoring functions, or some combination of these. The server 505 may receive updated training data and may retrain the supervised model using an updated vocabulary of interest, a new word embedding operation for an additional supported language, a new machine learning model for scoring an input String, or any combination of these or other updated training inputs (generating a second training data set based on user feedback, training the trained RNN model based on the second training data set to produce a second trained RNN model). Purnell, para 46.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement the method and system for training a recurrent neural network (RNN) model based on a training data set to produce a trained RNN model that is configured to infer output that represents a string input of Kurata using the vector representation of Hoffmeister and the executing a query to identify one or more matching strings, from a set of strings, that match a query string; wherein executing the query comprises: applying the trained RNN model against the query string to generate output that comprises a vector representation of the query string; comparing the vector representation of the query string to one or more vector representations of one or more respective strings of the set of strings; based on said comparing, identifying the one or more matching strings; returning results from said executing the query of He and generating a second training data set based on user feedback, training the trained RNN model based on the second training data set to produce a second trained RNN mode of Purnell, with a reasonable expectation of success, in order to provide speech processing that may also convert a user's speech into text data which may then be provided to various text-based software applications and in order to provide human-machine interactions that allow people to use their voices to effectuate control and help effectuate the control without any physical or visual contact with the device and in order to provide more accurate, robust natural language processing systems. Hoffmeister, col 1:5-23, col 26:5-44. He, para 2. Purnell, para 2. This would have provided the advantages of providing data for use text-based applications and improving interactions with applications and improving accuracy of machine learning models and natural language processing.
Claim 13 recites one or more non-transitory computer-readable media that parallels the method of claim 6. Therefore, the analysis discussed above with respect to claim 6 also applies to claim 13. Accordingly, claim 13 is rejected based substantially the same rationale set forth above with respect to claim 6.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. US-20180114108-A1, US-20190213284-A1, US-20200026908-A1, US-20200380403-A1, US-20220172049-A1.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BARBARA LEVEL whose telephone number is (303)297-4748. The examiner can normally be reached Monday through Friday 8:00 AM - 5:00 PM MT.
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, Mariela Reyes can be reached at (571) 270-1006. 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.
/BARBARA M LEVEL/ Examiner, Art Unit 2142
Prosecution Insights