DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Remarks
This action is in response to the amendments received on 9/8/26. Claims 1-5, 7-17, 19, 22, and 23 are pending in the application. Claims 6, 18, 20, and 21 have been cancelled and claim 23 has been added. Applicant’s arguments have been respectfully considered.
Claims 1-5, 7-17, 19, 22, and 23 are rejected under 35 U.S.C. 101.
Claim(s) 1-3, 8, 9, 11-15, 17, 19, and 23 are rejected under 35 U.S.C. 103 as being unpatentable over Lavallee (US 10,789,539), and further in view of Lee et al. (US 2020/0193153) and Araki (US 2023/0316001).
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Lavallee in view of Lee, and further in view of Qadrud-Din et al. (US 2024/0289561).
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Lavallee in view of Lee, and further in view of Vaswani et al. (US 2021/0390410).
Claims 10 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Lavallee in view of Lee, and further in view of Applicant Admitted Prior Art (AAPA).
Claim 22 is rejected under 35 U.S.C. 103 as being unpatentable over Lee et al. (US 2020/0193153), and further in view of Vaswani et al. (US 2021/0390410).
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 1-10 and 23 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 1 recites the limitation "the plurality of pairwise preference scores" in the ‘aggregating’ limitation. There is insufficient antecedent basis for this limitation in the claim.
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 1-5, 7-17, 19, 22, and 23 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 2A, Prong One asks: Is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea? See MPEP 2106.04 Part I. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. See MPEP 2106.04(a).
With respect to claims 1, 11, and 15, the limitation of “generating a respective prompt comprising a respective query”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is, nothing in the claim element precludes the step from practically being performed in the mind. For example, language, “generating” in the context of this claim encompasses the user mentally determining a problem to solve. Similarly, the limitation of “decoding, by the generative sequence processing model, one or more target tokens”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. For example, “decoding” in the context of this claim encompasses the user analyzing text. Additionally, “assigning, … a respective pairwise preference score”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. For example, “assigning” in the context of this claim encompasses the user mentally associating a score with text. The limitation of “aggregating the plurality of pairwise preference scores”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. For example, “aggregating” in the context of this claim encompasses the user combining scores.
If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
At step 2a, prong two, this judicial exception is not integrated into a practical application. The claims discuss “a generative sequence processing model”, however, this appears to be a generic computer component. Claims 11, 15, and 21 recite computing devices and a processor to execute the operations, however, this is recited as a high-level of generality (i.e., as a generic processor performing a generic computer function) such that it amounts to no more than mere instructions to apply the exception using a generic computer component. Additionally, the claims recite “prompting a generative sequence processing model with the prompt.” These elements do not integrate the abstract idea into a practical application because they do not impose a meaningful limit on the judicial exception and provide only insignificant extra solution activity that is mere data gathering in conjunction with the abstract idea.
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements amount to no more than mere instructions to apply an exception using generic computer components. Mere instructions to apply an exception using generic computer components cannot provide an inventive concept.
With respect to “prompting a generative sequence processing model with the prompt”, the courts have found limitations directed towards data gathering to be well-understood, routine, and conventional. See MPEP 2106.05(d)(II). Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information).
Considering the additional elements individually and in combination and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. The claim is not patent eligible.
With respect to claim 2, the limitations disclose “processing” input data, which is a mental process encompassing a user analyzing data. The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 3, 12, and 17, the limitations disclose “adding” the scores which is a mental process encompassing mental math. The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 4 and 5, the limitations disclose “assigning” points, which is a mental process encompassing the user mentally associating a score with text. The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 7-10, 14, and 23, the limitations further define the above mental processes and do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 13 and 19, the limitations disclose “performing” comparisons, “assigning” scores, and “aggregating” scores, which each disclose mental processes. These encompass the user comparing text, the user mentally associating a score with text, and the user combining scores, respectively. The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claim 16, the limitations disclose “providing” the text via an API. These elements do not integrate the abstract idea into a practical application because they do not impose a meaningful limit on the judicial exception and provide only insignificant extra solution activity in conjunction with the abstract idea.
The courts have found limitations directed towards outputting to be well-understood, routine, and conventional. See MPEP 2106.05(d)(II). Presenting offers and gathering statistics, OIP Techs., 788 F.3d at 1362-63, 115 USPQ2d at 1092-93. Considering the additional elements individually and in combination and the claim as a whole, the additional elements do not provide significantly more than the abstract idea.
With respect to claim 22, the limitation of “assigning… a respective pairwise score” , is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. For example, “assigning” in the context of this claim encompasses the user mentally associating a score with text. The limitation of “aggregating the plurality of pairwise scores”, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. For example, “aggregating” in the context of this claim encompasses the user combining scores. Similarly, the limitation of “decoding, by the machine-learned sequence processing model, one or more target tokens”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. For example, “decoding” in the context of this claim encompasses the user analyzing text.
If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
At step 2a, prong two, this judicial exception is not integrated into a practical application. The claims discuss “one or more model instances of a machine-learned sequence processing model”, however, this appears to be a generic computer component. Additionally, the claim discloses computing devices, however, this is recited as a high-level of generality (i.e., as a generic processor performing a generic computer function) such that it amounts to no more than mere instructions to apply the exception using a generic computer component. Additionally, the claims recite “providing… a plurality of prompts.” These elements do not integrate the abstract idea into a practical application because they do not impose a meaningful limit on the judicial exception and provide only insignificant extra solution activity that is mere data gathering in conjunction with the abstract idea.
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements amount to no more than mere instructions to apply an exception using generic computer components. Mere instructions to apply an exception using generic computer components cannot provide an inventive concept.
With respect to “providing… a plurality of prompts”, the courts have found limitations directed towards data gathering to be well-understood, routine, and conventional. See MPEP 2106.05(d)(II). Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information).
Considering the additional elements individually and in combination and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. The claim is not patent eligible.
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.
Claims 1-3, 8, 9, 11-15, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Lavallee (US 10,789,539), and further in view of Lee et al. (US 2020/0193153) and Araki (US 2023/0316001).
With respect to claim 1, Lavallee teaches a computer-implemented method for ranking, the method performed by one or more computing devices and comprising:
computing a plurality of pairwise scores based on, for each respective pairwise comparison of the plurality of pairwise comparisons between a plurality of candidate results (Lavallee, Col. 11 Li. 59-61, At step 305, the computing device may use a pairwise classifier to determine scores for element pairs in the potential results & Col. 12 Li. 41-43, At step 307, the computing device may determine one or more pairwise probabilities based on the scores for the element pairs determined by the pairwise classifier):
generating a respective prompt comprising a respective query (Lavelle, Col. 10 Li. 66 – Col. 11-1, request may be received by the computing device and the request may identify previously stored text that is to be processed.), a respective first set of text associated with a respective first candidate result of the plurality of candidate results, and a respective second set of text associated with a respective second candidate result of the plurality of candidate results (Lavallee, Li. 32-41, System 200 may comprise a tagging (TE) engine 208, which may process a transcription or query for tagging. TE 208 may leverage/process device and/or user metadata that may be stored in a database and/or on the device. For example, TE 208 may parse a string of words (e.g., using grammars, named entity processing, and/or internal concept processing) to determine whether any of the words in the string match any of the user metadata, such as a name in a contact list ( e.g., a contact list stored on a user's device, such as a mobile phone). & Li. 53-54, The tagged data may be stored in one or more databases or caches, such as database 210);
prompting a generative sequence processing model with the respective prompt (Lavallee, Col. 11 Li. 1-5, the computing device may retrieve the previously stored text and provide the text as natural language input for processing in order to, for example, generate natural language understanding output.);
wherein the respective pairwise comparison between the respective first set of text and the respective second set of text based on the query (Lavallee, Col. 11 Li. 59-61, At step 305, the computing device may use a pairwise classifier to determine scores for element pairs in the potential results); and
generating, by the generative sequence processing model based on the respective pairwise comparison, a respective output indicating whether the respective first set of text is preferred to the respective second set of text or the respective second set text is preferred to the respective first set of text (Lavallee, Col. 12 Li. 41-43, At step 307, the computing device may determine one or more pairwise probabilities based on the scores for the element pairs determined by the pairwise classifier); and
assigning, when the one or more target tokens indicates that the respective first set of text is preferred to the respective second set of text (Lavallee, Col. 12 Li. 53-61, with respect to scores produced by pairwise classifiers, such as those described herein in connection with step 305 of FIG. 3, each score for an element pair may be processed through an algorithm in order to determine a corresponding pairwise probability. With respect to a score produced by a pairwise classifier for element pair (A, B), the score may be processed so that a pairwise probability is determined, which represents the probability that A is better than B.), a respective pairwise preference score for the respective first set of text, or, when the one or more target tokens indicates that the respective second set of text is preferred to the respective first set of text, the respective pairwise preference score for the respective second set of text (Lavallee, Col. 13 Li. 27-33, At step 309, the computing device may determine, for each of the potential results, an estimation of the probability that the current result is the top ranked, or best, result among the potential results. In other words, the probability estimate being determined at step 309 is the probability that a particular element in the potential results is the best choice given all the alternatives in the potential results.).
aggregating the plurality of pairwise preference scores into an aggregate score for each candidate result based on a cumulative number of pairwise comparisons in which the one or more target tokens identifies that candidate result as the preferred candidate result (Lavallee, Col. 12 Li. 61-64, where the potential results includes three elements (A, B and C), the scores may be processed to determine pairwise probability estimates, which include the probability that A is better than B, 65 P(A>B); the probability that B is better than C, P(B>C); and the probability that A is better than C, P(A>C).) to generate a ranking for the plurality of candidate results (Lavallee, Col. 18 Li. 5-7, At step 311, the computing device may determine a ranking of the potential results based on the top-rank probability estimates. For example, the potential results may be sorted so that the element with the highest top-rank probability is first in the list and is followed by the element with the second highest top-rank probability, and so forth).
Lavallee doesn't expressly discuss wherein the plurality of pairwise comparisons are performed in parallel.
Lee teaches wherein the plurality of pairwise comparisons are performed in parallel (Lee, pa 0209, The comparisons may be performed partly or completely in parallel to enable a fast processing time.).
It would have been obvious at the effective filing date of the invention to a person having ordinary skill in the art to which said subject matter pertains to have modified Lavallee with the teachings of Lee because it may significantly improve the data processing speed (Lee, pa 0209).
Lavallee in view of Lee doesn’t expressly discuss wherein the respective pairwise comparison between the respective first set of text and the respective second set of text based on the query is performed through decoding, by the generative sequence processing model, one or more target tokens identifying whether the respective first candidate result or the respective second candidate result is a preferred candidate result.
Araki teaches wherein the respective pairwise comparison between the respective first set of text and the respective second set of text based on the query is performed through decoding, by the generative sequence processing model, one or more target tokens identifying whether the respective first candidate result or the respective second candidate result is a preferred candidate result (Araki, pa 0025, the machine learning system 140 may convert [Y] to a sequence of mask tokens “[MASK] . . . [MASK]” and choose the final prediction based on the confidence by enumerating from 1 to M mask tokens in [Y], where M is the maximum number of mask tokens allowed for the answer slot. For example, as expressed in equation 1, the confidence score may be denoted as c(m) for “m” mask tokens.);
It would have been obvious at the effective filing date of the invention to a person having ordinary skill in the art to which said subject matter pertains to have modified Lavallee in view of Lee with the teachings of Araki because it can indicate the most likely term for that particular domain (Araki, pa 0025).
With respect to claim 2, Lavallee in view of Lee and Araki teaches the computer-implemented method of claim 1, wherein performing, by the generative sequence processing model, the respective pairwise comparison comprises:
processing a first input comprising the first set of text ordered before the second set of text; and processing a second input comprising the second set of text ordered before the first set of text (This was well known way to perform pairwise comparisons at the time of the invention. See Lavalle Col. 3 Li. 64-67).
With respect to claim 3, Lavallee in view of Lee and Araki teaches the computer-implemented method of claim 1, wherein aggregating the respective pairwise score into the aggregate score comprises adding the respective pairwise score to a cumulative total of pairwise scores associated with the respective first set of text being the preferred set of text or to a cumulative total of pairwise scores associated with the respective second set of text being the preferred set of text (Lee, pa 0254, The user interface presented to the user may include components for allowing the user accept or reject the portions of text presented as similar. Acceptance indicates that the user agrees that the portion of text identified by the search engine is in fact similar to the portion of text that the user had entered. Rejection indicates that the user does not think that the portions of text are similar. The acceptances and rejections from the user may be used to train and improve the machine learning model for text similarity. When an acceptance is received, the value of the association between the two portions may be increased and the machine learning model may be trained to make it more likely to recommend the stored text segment for the input text segment, or text like it, in the future).
With respect to claim 8, Lavallee in view of Lee and Araki teaches the computer-implemented method of claim 1, wherein the output corresponds to a likelihood of the generative sequence processing model generating a token that identifies the set of text as the preferred text (Lavallee, Col. 12 Li. 41-43, At step 307, the computing device may determine one or more pairwise probabilities based on the scores for the element pairs determined by the pairwise classifier).
With respect to claim 9, Lavallee in view of Lee and Araki teaches the computer-implemented method of claim 8, wherein the output comprises text corresponding to the token (Lavallee, Col. 18 Li. 12-17, the potential results may be sorted so that the person determined to have the greatest top-rank probability estimate is first in the list (e.g., Andy Smith is sorted to be first in the list if the top-rank probability estimate for Andy Smith is greater than the top-rank probability estimate for Andy Jones).).
With respect to claim 11, Lavallee teaches a computer-implemented method for prompt-based ranking, the method performed by one or more computing devices and comprising:
providing, for a plurality of pairwise comparisons by a generative sequence processing model (Lavallee, Col. 11 Li. 59-61, At step 305, the computing device may use a pairwise classifier to determine scores for element pairs in the potential results & Col. 12 Li. 41-43, At step 307, the computing device may determine one or more pairwise probabilities based on the scores for the element pairs determined by the pairwise classifier), a plurality of sets of text, each set of text of the plurality of sets of text associated with a candidate result of a plurality of candidate results (Lavalle, Li. 32-41, System 200 may comprise a tagging (TE) engine 208, which may process a transcription or query for tagging. TE 208 may leverage/process device and/or user metadata that may be stored in a database and/or on the device. For example, TE 208 may parse a string of words (e.g., using grammars, named entity processing, and/or internal concept processing) to determine whether any of the words in the string match any of the user metadata, such as a name in a contact list ( e.g., a contact list stored on a user's device, such as a mobile phone). & Li. 53-54, The tagged data may be stored in one or more databases or caches, such as database 210);
assigning, for each of the plurality of pairwise comparisons, a pairwise score for the respective first set of text or for the respective second set of text based on the one or more target tokens (Lavallee, Col. 13 Li. 38-40, Once each probability estimate or approximate has been determined, the elements may be sorted into a rank order by the obtained probability estimates.); and
generating, based on the assigned pairwise scores (Lavallee, Col. 12 Li. 61-64, where the potential results includes three elements (A, B and C), the scores may be processed to determine pairwise probability estimates, which include the probability that A is better than B, 65 P(A>B); the probability that B is better than C, P(B>C); and the probability that A is better than C, P(A>C).), an aggregate score for each of the plurality of sets of text by accumulating the assigned pairwise scores across the plurality of pairwise comparisons to generate a ranking of the plurality of sets of text (Lavallee, Col. 18 Li. 5-7, At step 311, the computing device may determine a ranking of the potential results based on the top-rank probability estimates. For example, the potential results may be sorted so that the element with the highest top-rank probability is first in the list and is followed by the element with the second highest top-rank probability, and so forth).
Lavallee doesn't expressly discuss wherein the plurality of pairwise comparisons are performed in parallel.
Lee teaches wherein the plurality of pairwise comparisons are performed in parallel (Lee, pa 0209, The comparisons may be performed partly or completely in parallel to enable a fast processing time.).
It would have been obvious at the effective filing date of the invention to a person having ordinary skill in the art to which said subject matter pertains to have modified Lavallee with the teachings of Lee because it may significantly improve the data processing speed (Lee, pa 0209).
Lavallee in view of Lee doesn’t expressly discuss wherein each of the plurality of pairwise comparisons is performed through decoding, by the generative sequence processing model, one or more target tokens indicating preference between a pair of the sets of texts conditioned on a respective prompt comprising a query and the pair of the sets of text.
Araki teaches wherein each of the plurality of pairwise comparisons is performed through decoding, by the generative sequence processing model, one or more target tokens indicating preference between a pair of the sets of texts conditioned on a respective prompt comprising a query and the pair of the sets of text (Araki, pa 0025, the machine learning system 140 may convert [Y] to a sequence of mask tokens “[MASK] . . . [MASK]” and choose the final prediction based on the confidence by enumerating from 1 to M mask tokens in [Y], where M is the maximum number of mask tokens allowed for the answer slot. For example, as expressed in equation 1, the confidence score may be denoted as c(m) for “m” mask tokens.);
It would have been obvious at the effective filing date of the invention to a person having ordinary skill in the art to which said subject matter pertains to have modified Lavallee in view of Lee with the teachings of Araki because it can indicate the most likely term for that particular domain (Araki, pa 0025).
With respect to claims 12 and 14, the limitations are essentially the same as claims 3 and 8, and are rejected for the same reasons.
With respect to claim 13, Lavallee in view of Lee and Araki teaches the computer implemented method of claim 11, further comprising: performing a plurality of pairwise comparisons for a plurality of sets of text in parallel; assigning pairwise scores for the preferred sets of texts for each of the plurality of pairwise comparisons (Lavallee, Col. 13 Li. 38-40, Once each probability estimate or approximate has been determined, the elements may be sorted into a rank order by the obtained probability estimates.); and aggregating the pairwise scores for each of the plurality of sets of text into an aggregate score for each of the plurality of sets of text to generate the ranking (Lee, pa 0224, The candidate combinations may then be scored based on the scoring criteria so that a highest scoring candidate combination is selected as the final combination.).
With respect to claims 15 and 21, the limitations are essentially the same as claim 11, and are rejected for the same reasons.
With respect to claim 23, Lavallee in view of Lee and Araki teaches the computer-implemented method of claim 1, wherein decoding the one or more target tokens comprises autoregressively decoding the one or more target tokens from an output vocabulary of the generative sequence processing model based on a probability distribution conditioned on the respective prompt (Araki, pa 0024, the machine learning system 140 includes an autoregressive language model, which is pre-trained. The language model is configured to predict a next word (e.g., target entity) based on the previous context (e.g., the prompt), thereby modeling natural language (e.g., English). For example, as shown in FIG. 3, upon receiving a prompt as input, the machine learning system 140 is configured to perform factual knowledge retrieval by predicting the answer slot [Y], given an input prompt x.sub.prompt. In this regard, the KRETC system 200 is considered to have the factual knowledge queried by x.sub.prompt if the final output 9 produced by the machine learning system 140 (e.g., the language model) is the same as the ground truth y.).
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Lavallee in view of Lee, and further in view of Qadrud-Din et al. (US 2024/0289561).
With respect to claim 4, Lavallee in view of Lee teaches the computer-implemented method of claim 3, as discussed above.
Qadrud-Din teaches assigning one point to the preferred set of text (Qadrud-Din, pa 0186-0191, 2. If you extracted any passages, assign each one a score from 1 to 5, representing how the passage relates to the question: [0187] 5 (complete answer) [0188] 4 (one piece of a multipart answer) [0189] 3 (relevant definition or fact) [0190] 2 (useful context) [0191] 1 (marginally related)).
It would have been obvious at the effective filing date of the invention to a person having ordinary skill in the art to which said subject matter pertains to have modified Lavallee in view of Lee with the teachings of Qadrud-Din because in assists in finding relevant text (Qadrud-Din, pa 0183).
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Lavallee in view of Lee, and further in view of Vaswani et al. (US 2021/0390410).
With respect to claim 7, Lavallee in view of Lee teaches the computer-implemented method of claim 6, as discussed above.
Vaswani teaches wherein the generative sequence processing model processes a batch of input requests in parallel, the batch of input requests comprising the respective prompt (Vaswani, each layer input includes a “batch” dimension, where the neural network processes a batch of multiple input images in parallel and the layer input 210 includes a respective index along the batch dimension for each input image in the batch.).
It would have been obvious at the effective filing date of the invention to a person having ordinary skill in the art to which said subject matter pertains to have modified Lavallee in view of Lee with the teachings of Vaswani because it significantly reduces the time and memory required to both train the neural network and perform inference using the neural networks (Vaswani, pa 0006).
Claims 10 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Lavallee, and further in view of Applicant Admitted Prior Art (AAPA).
With respect to claim 10, Lavallee in view of Lee teaches the computer-implemented method of claim 8, as discussed above.
AAPA teaches wherein the output is received via an application programming interface that does not output calibrated prediction probabilities (AAPA, specification, pa 0005, with the existing methods, ranking requires generative sequence processing models to output calibrated prediction probabilities before sorting, which is very difficult and not supported by generation only APIs).
It would have been obvious at the effective filing date of the invention to a person having ordinary skill in the art to which said subject matter pertains to have modified Lavallee in view of Lee with the teachings of AAPA because this is how existing methods operate (AAPA, pa 0005).
With respect to claim 16, the limitations are essentially the same as claim 10, and are rejected for the same reasons.
Claim 22 is rejected under 35 U.S.C. 103 as being unpatentable over Lee et al. (US 2020/0193153), and further in view of Vaswani et al. (US 2021/0390410) and Araki (US 2023/0316001).
With respect to claim 22, Lee teaches a computer-implemented method for ranking, the method performed by one or more computing devices and comprising:
providing, for input to one or more model instances of a machine-learned sequence processing model, a plurality of prompts, wherein each respective prompt of the plurality of prompts corresponds to a respective pairwise comparison between a plurality of candidate results and comprises a respective query, a respective first set of text associated with a respective first candidate result, and a respective second set of text associated with a respective second candidate result (Lee, pa 0089-0092, In step 302, an input text block may be received from a user. In step 303, the input text block may be split into input text segments. In step 304, text similarity matching may be performed on each input text segment using the machine learning model. … the input text segments may be considered query text segments and the stored text segments in document storage 124 may be considered candidate text segments.);
assigning, when the one or more target tokens indicates that the respective first set of text is preferred to the respective second set of text, a respective pairwise score for the respective first set of text, or, when the one or more target tokens indicates that the respective second set of text is preferred to the respective first set of text, the respective pairwise score for the respective second set of text (Lee, pa 0224, The candidate combinations may then be scored based on the scoring criteria so that a highest scoring candidate combination is selected as the final combination.); and
aggregating the plurality of pairwise scores into an aggregate score by accumulating the assigned pairwise scores across the plurality of prompts distributed across the batch dimension to generate a ranking for the plurality of candidate results (Lee, pa 0224, multiple candidate combinations may be output rather than selecting a single combination for output, in order to provide more options to the user.).
Lee doesn't expressly discuss wherein the one or more model instances of the machine-learned sequence processing model are configured with an input structure having a batch dimension for processing inputs in parallel, the plurality of prompts distributed across the batch dimension of the input structure.
Vaswani teaches wherein the one or more model instances of the machine-learned sequence processing model are configured with an input structure having a batch dimension for processing inputs in parallel, the plurality of prompts distributed across the batch dimension of the input structure (Vaswani, pa 0034, each layer input includes a “batch” dimension, where the neural network processes a batch of multiple input images in parallel and the layer input 210 includes a respective index along the batch dimension for each input image in the batch.).
It would have been obvious at the effective filing date of the invention to a person having ordinary skill in the art to which said subject matter pertains to have modified Lee with the teachings of Vaswani because it significantly reduces the time and memory required to both train the neural network and perform inference using the neural networks (Vaswani, pa 0006).
Lee in view of Vaswani doesn’t expressly discuss wherein each of the plurality of pairwise comparisons is performed through decoding, by the generative sequence processing model, one or more target tokens indicating preference between a pair of the sets of texts conditioned on a respective prompt comprising a query and the pair of the sets of text.
Araki teaches wherein each of the plurality of pairwise comparisons …is performed through decoding, by the machine-learned sequence processing model, one or more target tokens identifying a preferred candidate result between the respective first candidate result or the respective second candidate result (Araki, pa 0025, the machine learning system 140 may convert [Y] to a sequence of mask tokens “[MASK] . . . [MASK]” and choose the final prediction based on the confidence by enumerating from 1 to M mask tokens in [Y], where M is the maximum number of mask tokens allowed for the answer slot. For example, as expressed in equation 1, the confidence score may be denoted as c(m) for “m” mask tokens.).
It would have been obvious at the effective filing date of the invention to a person having ordinary skill in the art to which said subject matter pertains to have modified Lee in view of Vaswani with the teachings of Araki because it can indicate the most likely term for that particular domain (Araki, pa 0025).
Response to Arguments
35 U.S.C. 101
Applicant argues that the claims reflect an improvement of generative model document ranking and computer functionality, as disclosed in the Specification. The Examiner respectfully disagrees. Applicant has not pointed to specific limitations in the claim that reflect the technical improvements. Applicant does not identify any specific error in examiner’s written rejection identifying specific limitations of the claims that recite abstract ideas. MPEP 2106.04(I) directs examiners to “identify that the claimed concept (the specific claim limitation(s) that the examiner believes may recite an exception) aligns with at least one judicial exception.” This is reiterated in MPEP 2106.04(a), “Examiners should determine whether a claim recites an abstract idea by (1) identifying the specific limitation(s) in the claim under examination that the examiner believes recites an abstract idea, and (2) determining whether the identified limitations(s) fall within at least one of the groupings of abstract ideas listed above.” MPEP 2106.04(d) then states “Accordingly, after determining that a claim recites a judicial exception in Step 2A Prong One, examiners should evaluate whether the claim as a whole integrates the recited judicial exception into a practical application of the exception in Step 2A Prong Two.” See also MPEP 2106.07(a). The examiner has done so here in the rejection, but Appellant’s argument does not address the specific limitations identified in the rejection as reciting an abstract idea and merely summarizes the scope of various claims. Appellant’s analysis that is detached from the actual claim language (“what is recited in the claim”) cannot be persuasive of any examiner error in the formulation of the 101 rejection.
35 U.S.C. 103
Applicant seems to argue a newly amended limitation. Applicant’s amendment has rendered the previous rejection moot. Upon further consideration of the amendment, a new grounds of rejection is made in view of Araki (US 2023/0316001).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BRITTANY N ALLEN whose telephone number is (571)270-3566. The examiner can normally be reached M-F 9 am - 5:00 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, Sherief Badawi can be reached at 571-272-9782. 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.
/BRITTANY N ALLEN/ Primary Examiner, Art Unit 2169
