DETAILED ACTION
This Office Action is in response to the correspondence filed by the applicant on 7/10/2026.
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 .
Response to Arguments
Regarding 112 rejections, applicant’s arguments (pg. 8) have been fully considered and are persuasive. The rejection is now withdrawn.
Regarding 101 rejections, applicant’s arguments (pgs. 10-12) have been fully considered, but they are not persuasive. Under the patent eligibility guidance, the claims are directed to an abstract idea.
For Step 1, Examiner determines that the claims fall into statutory category. For example, the independent claim 1 recites a series of steps, therefore, is a process.
For Step 2A Prolong 1, Examiner determines that claims recite judicial exception. The independent claims 21, 28, and 35 recite, “receiving, by a machine learning system, a sentence comprising a plurality of words, respectively provided as tokens of the sentence to input to an encoder of a machine learning model; and training, by the machine learning system, the machine learning model using a vector of probabilities determined from respective bias scores for each token in the input sentence to determine one or more tokens of the input sentence to replace with a mask token.” These limitations, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. That is, other than reciting “by a machine learning system”, nothing in the claim element preclude the step from practically being performed in the mind. For example, a person can calculate model parameters using a vector determined from bias scores for each word in the input sentence. The limitations, as drafted, are processes that, under its broadest reasonable interpretation, cover performance of the limitations in the mind. Thus, the claims recite a mental process.
For Step 2A Prolong 2, Examiner determines that claims are not integrated in to a practical application. The claim recite additional elements: “at an information handling device”; “a processor; a memory device that stores instructions executable by the processor”; and “a storage device that stores code, the code being executable by a processor”
Each of the additional elements and/or the combination of the additional elements is no more than mere instructions to apply the exception using generic computer components. Accordingly, even in combination, these additional elements do not integrate the abstract idea into a practical application because they do not impose any meaning limits on practicing the abstract idea. The claims are thus directed to the abstract idea.
For Step 2B, Examiner determines that claims do not provide an inventive concept. As discussed with respect to Step 2A Prolong Two, the additional elements in the claim amount to no more than mere instructions to apply the exception using a generic computer component. The same analysis applies here in 2B, i.e., mere instructions to apply an exception on a generic computer cannot integrate a judicial exception into a practical application at Step 2A or provide an inventive concept in Step 2B. The claims are ineligible.
Even though the disclosed invention is described in the specification as improving model training, the claim provides no meaningful limitations such that this improvement is realized. Therefore, the claim does not amount to significantly more than the abstract idea itself.
Regarding the rejection of claim 21 under 35 U.S.C. 103, Applicant’s arguments (pg. 9-10) with respect to rejections have been fully considered, but they are not persuasive.
Applicant asserts, “there is no teaching of a ‘bias score’”. However, Examiner respectfully disagrees. In section 5.3 (pgs. 55-56), BADJATIYA states, “We use identical Convolutional Neural Network (CNN) [8] implemented in Keras [6] for all the experiments. Fig. 2 shows the architecture diagram of the neural network. We use 100D GloVe [19] embeddings to encode text and CNN architecture with 128 filters of size 5×5×5 for each convolution layer, dropout of 0.3, max-sequence length as 250, categorical-cross-entropy as the loss function and RMSPROP as optimizer with learning-rate of 0.00005. We use the best parameter setting as described in Dixon et al. [8] and train the classifiers for our task with batch size 128 and early stopping to a maximum of 20 epochs, keeping other hyperparameters constant throughout the experiments.) In other words, the CNN in Figure 2 is trained to minimizes the “categorical-cross-entropy” loss function using RMSPROP optimizer. As one of ordinary skill in the art would know the categorical-cross-entropy is defined as “ – summation of yi log (y’I ) for i=1 … k, where yi is the true probability of the i-th class and y’i is the predicted probability of the i-th class.” This value measures the difference between the predicted and true probability distributions across multiple classes. The categories of the BADJATIYA are biased (hateful and offensive) and neutral (pg. 51 Table 2 and section 3.1; pg. 54 Section 5.1). Since the classifier can be trained to classify a sentence-level input, the word-level probability indicates the “bias” score for each word, and a vector of probabilities (i.e., bias score of the words in the sentence; e.g., a sentence with a word “muslims” with probability of 0.81. and the words “she” and “woman” in “is Alex acting weird because she is a woman?” as described on pg. 51 2nd Col.) is determined based on the word-level probability. In other words, the word level bias scores are used to determine the sentence-level probabilities. And the sentence-level probabilities are used to calculate the predicted probabilities for the categorical-cross-entropy function for training the sentence-level classifier. Thus, BADJATIYA teaches the limitations.
Double Patenting
The nonstatutory double patenting rejection is based on a judicially created doctrine grounded in public policy (a policy reflected in the statute) so as to prevent the unjustified or improper timewise extension of the “right to exclude” granted by a patent and to prevent possible harassment by multiple assignees. A nonstatutory double patenting rejection is appropriate where the claims at issue are not identical, but at least one examined application claim is not patentably distinct from the reference claim(s) because the examined application claim is either anticipated by, or would have been obvious over, the reference claim(s). See, e.g., In re Berg, 140 F.3d 1428, 46 USPQ2d 1226 (Fed. Cir. 1998); In re Goodman, 11 F.3d 1046, 29 USPQ2d 2010 (Fed. Cir. 1993); In re Longi, 759 F.2d 887, 225 USPQ 645 (Fed. Cir. 1985); In re Van Ornum, 686 F.2d 937, 214 USPQ 761 (CCPA 1982); In re Vogel, 422 F.2d 438, 164 USPQ 619 (CCPA 1970); and In re Thorington, 418 F.2d 528, 163 USPQ 644 (CCPA 1969).
A timely filed terminal disclaimer in compliance with 37 CFR 1.321(c) or 1.321(d) may be used to overcome an actual or provisional rejection based on a nonstatutory double patenting ground provided the reference application or patent either is shown to be commonly owned with this application, or claims an invention made as a result of activities undertaken within the scope of a joint research agreement. A terminal disclaimer must be signed in compliance with 37 CFR 1.321(b).
The USPTO internet Web site contains terminal disclaimer forms which may be used. Please visit http://www.uspto.gov/forms/. The filing date of the application will determine what form should be used. A web-based eTerminal Disclaimer may be filled out completely online using web-screens. An eTerminal Disclaimer that meets all requirements is auto-processed and approved immediately upon submission. For more information about eTerminal Disclaimers, refer to http://www.uspto.gov/patents/process/file/efs/guidance/eTD-info-I.jsp.
Claims 21-40 are rejected on the ground of nonstatutory double patenting as being unpatentable over Claims 1-20 of US PAT 12,106,050 Although the claims, at issue are not identical, they are not patentably distinct from each other because the claims of the instant application are rejected as being unpatentable over the claims of the US PAT. Please see below for the mapping in the table, where the bolded limitations indicate the corresponding limitations between the US PAT and instant application.
Instant application: 18/817,147
US PAT 12,106,050
21. A system, comprising:
at least one processor; a memory, comprising program instructions that when executed by the at least one processor cause the at least one processor to implement a machine learning system, the machine learning system configured to:
receive a sentence comprising a plurality of words, respectively provided as tokens of the sentence to input to an encoder of a machine learning model; and
train the machine learning model using a vector of probabilities determined from respective bias scores for each token in the input sentence to determine one or more tokens of the input sentence to replace with a mask token.
1. A system, comprising:
at least one processor;
a memory, comprising program instructions that when executed by the at least one processor cause the at least one processor to implement a machine learning system, the machine learning system configured to:
receive a sentence comprising a plurality of words, respectively to be provided as tokens of the sentence to input to an encoder of a machine learning model;
determine a token-wise correlation using a semantic orientation for a given bias attribute to determine a bias score for each of the tokens in the sentence;
determine a respective probability of dropout for each token based on the bias score determined for each of the tokens in the sentence; and
selectively perform dropout for each of the tokens in the sentence as part of training or tuning the machine learning model using the sentence based on the respective probability of dropout determined for each of the tokens in the sentence.
Other independent claims 28 and 35 are also similar to the independent claims 8 and 15 of the US PAT.
With respect to the dependent claims, each of the claims maps to a corresponding dependent claim of the US PAT or are found within the scope of the independent 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 21-22, 25-26, 28-29, 32-33, 35-36, and 39 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The independent claims 21, 28, and 35 recite receiving, by a machine learning system, a sentence comprising a plurality of words, respectively provided as tokens of the sentence to input to an encoder of a machine learning model; and training, by the machine learning system, the machine learning model using a vector of probabilities determined from respective bias scores for each token in the input sentence to determine one or more tokens of the input sentence to replace with a mask token.
The recited limitations, 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, other than reciting “by a machine learning system”, nothing in the claim element precludes the step from practically being performed in the mind. For example, a person can calculate model parameters using a vector determined from bias scores for each word in the input sentence. The limitations, as drafted, are processes that, under its broadest reasonable interpretation, cover performance of the limitations in the mind.
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.
This judicial exception is not integrated into a practical application. In particular, the claims only recite additional elements – “by a machine learning system”. The additional elements in both steps is recited at a high-level of generality (i.e., as a generic processor performing a generic computer function of the recited steps) such that it amounts no more than mere instructions to apply the exception using a generic computer component. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. The claim is directed to an 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 element of using a computer (i.e., machine learning system) to perform the recited steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding the dependent claims, claims 22, 29, and 36 recite determining an absolute value difference and adding the absolute value difference; claims 25, 32, and 39 recite applying vector for one or more epochs; claims 26 and 33 recite a dropout for each token for tuning the model.
Even though the disclosed invention is described in the specification as improving computer technology, the claim provides no meaningful limitations such that this improvement is realized. Therefore, the claim does not amount to significantly more than the abstract idea itself.
Accordingly, the limitations of the Claims, whether considered individually or as an ordered combination, are not sufficient to add significantly more to improve technological functionality. As such, claims 21-22, 25-26, 28-29, 32-33, 35-36, and 39 are rejected under 35 U.S.C. 101 as being directed to non-statutory subject matter.
Allowable Subject Matter
Claims 22-24, 29-31, and 36-38 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims, and if rewritten to overcome the 101 rejections.
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 of this title, 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 21, 25, 28, 32, 35, and 39 are rejected under 35 U.S.C. 103 as being unpatentable over BADJATIYA (Badjatiya P, Gupta M, Varma V. Stereotypical bias removal for hate speech detection task using knowledge-based generalizations. InThe world wide web conference 2019 May 13 (pp. 49-59).), and in further view of KURITA (Kurita K, Vyas N, Pareek A, Black AW, Tsvetkov Y. Measuring bias in contextualized word representations. InProceedings of the first workshop on gender bias in natural language processing 2019 Aug (pp. 166-172).).
REGARDING CLAIM 21, BADJATIYA discloses a discloses a system, comprising: at least one processor; a memory, comprising program instructions that when executed by the at least one processor cause the at least one processor to implement a machine learning system, the machine learning system (Abst – “social media platforms … bias mitigation from unstructured text data … methods leveraging knowledge-based generalizations for bias-free learning. … real-world datasets … of size ~96k and a Twitter dataset of size ~24k”) configured to:
receive a sentence comprising a plurality of words, respectively provided as tokens of the sentence to input to an encoder of a machine learning model (Pg. 56 Figure 2.; Section 5.3 – “We use identical Convolutional Neural Network (CNN) [8] implemented in Keras [6] for all the experiments. Fig. 2 shows the architecture diagram of the neural network. We use 100D GloVe [19] embeddings to encode text and CNN architecture with 128 filters of size 5×5×5 for each convolution layer, dropout of 0.3, max-sequence length as 250, categorical-cross-entropy as the loss function and RMSPROP as optimizer with learning-rate of 0.00005. We use the best parameter setting as described in Dixon et al. [8] and train the classifiers for our task with batch size 128 and early stopping to a maximum of 20 epochs, keeping other hyperparameters constant throughout the experiments.”); and
train the machine learning model using a vector of probabilities determined from respective bias scores for each token in the input sentence (pg. 51 2nd Col – “For example, if a classifier labels the sentence with just a word “muslims” as “Hateful” with probability 0.81, then we say that the word “muslims” is a BSW for the classifier.’;Pg. 56 Figure 2.; Section 5.3 – “We use identical Convolutional Neural Network (CNN) [8] implemented in Keras [6] for all the experiments. Fig. 2 shows the architecture diagram of the neural network. We use 100D GloVe [19] embeddings to encode text and CNN architecture with 128 filters of size 5×5×5 for each convolution layer, dropout of 0.3, max-sequence length as 250, categorical-cross-entropy as the loss function and RMSPROP as optimizer with learning-rate of 0.00005. We use the best parameter setting as described in Dixon et al. [8] and train the classifiers for our task with batch size 128 and early stopping to a maximum of 20 epochs, keeping other hyperparameters constant throughout the experiments.”; Pg. 56 Section 5.4.1 BSWs identified by Various Strategies. – “detecting biased words using Skewed Predicted Class Probability Distribution (SPCPD) … ”) to determine one or more tokens of the input sentence to replace with [a mask] tag token (Abst. – “Knowledge-based generalization provides an effective way to encode knowledge because the abstraction they provide not only generalizes content but also facilitates retraction of information from the hate speech detection classifier, thereby reducing the imbalance”; Pg. 50 1st Col – “Detection and Replacement. In the first stage (Detection), we propose skewed prediction probability and class distribution imbalance based novel heuristics for bias sensitive word (BSW) detection. Further, in the second stage (Replacement), we present novel bias removal strategies leveraging knowledge-based generalizations. These include replacing BSWs with generalizations like their Part-of-Speech (POS) tags, Named Entity tags, WordNet [17] based linguistic equivalents, and word embedding based generalizations.”; Pg. 53 4.3.1 Replacing with Part-of-speech (POS) tags. – “Example: Replace the word ‘Muhammad’ with POS tag ‘NOUN’ in the sentence - Muhammad set the example for his followers, and his example shows him to be a cold-blooded murderer. POS replacement substitutes specific information about the word ‘Muhammad’ from the text and only exposes partial information about the word, denoted by its POS tag ‘NOUN’, forcing it to use signals that do not give significant importance to the word ‘Muhammad’ to improve its predictions.”).
BADJATIYA does not explicitly teach the [square-bracketed] limitation and teaches the underlined feature instead.
KURITA disclose the [square-bracketed] limitation. KURITA discloses a method/system for measuring bias in contextualized word representations, train the machine learning model using a vector of probabilities determined from respective bias scores for each token in the input sentence to determine one or more tokens of the input sentence to replace with [a mask] token (KURITA Section 2 Quantifying Bias in BERT – “BERT is trained using a masked language modelling objective i.e. to predict masked tokens, denoted as [MASK], in a sentence given the entire context. We use the predictions for these [MASK] tokens to measure the bias encoded in the actual representations. We directly query the underlying masked language model in BERT2 to compute the association between certain targets (e.g., gendered words) and attributes (e.g. career-related words). For example, to compute the association between the target male gender and the attribute programmer, we feed in the masked sentence “[MASK] is a programmer” to BERT, and compute the probability assigned to the sentence ‘he is a programmer” (ptgt). To measure the association, however, we need to measure how much more BERT prefers the male gender association with the attribute programmer, compared to the female gender. We thus re-weight this likelihood ptgt using the prior bias of the model towards predicting the male gender. To do this, we mask out the attribute programmer and query BERT with the sentence “[MASK] is a [MASK]”, then compute the probability BERT assigns to the sentence ‘he is a [MASK]” (pprior). Intuitively, pprior represents how likely the word he is in BERT, given the sentence structure and no other evidence. Finally, the difference between the normalized predictions for the words he and she can be used to measure the gender bias in BERT for the programmer attribute.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the method/system of BADJATIYA to include replacing with a mask token, as taught by KURITA.
One of ordinary skill would have been motivated to include replacing with a mask token, in order to more effectively predict biased tokens within a sentence.
REGARDING CLAIM 25, BADJATIYA in view of KURITA discloses the system of claim 21, wherein the vector of probabilities is applied for one or more epochs (BADJATIYA Pg. 56 – “We use the best parameter setting as described in Dixon et al. [8] and train the classifiers for our task with batch size 128 and early stopping to a maximum of 20 epochs, keeping other hyperparameters constant throughout the experiments.”).
Regarding Claim 28, it is similar to Claim 21; thus, it is rejected under the same rationale.
Regarding Claim 32, it is similar to Claim 25; thus, it is rejected under the same rationale.
Regarding Claim 35, it is similar to Claim 21; thus, it is rejected under the same rationale.
Regarding Claim 39, it is similar to Claim 25; thus, it is rejected under the same rationale.
Claims 26 and 33 are rejected under 35 U.S.C. 103 as being unpatentable over BADJATIYA in view of KURITA, and in further view of WEBSTER (Webster K, Wang X, Tenney I, Beutel A, Pitler E, Pavlick E, Chen J, Chi E, Petrov S. Measuring and reducing gendered correlations in pre-trained models. arXiv preprint arXiv:2010.06032. 2020 Oct 12.).
REGARDING CLAIM 26, BADJATIYA in view of KURITA discloses the system of claim 21.
BADJATIYA in view of KURITA does not explicitly teach a dropout for tuning a model.
WEBSTER discloses a method/system for de-biasing a language model wherein the machine learning model is pre-trained such that the dropout for each of the tokens in the sentence (Section 5.1 – “Dropout regularization is used when training large models to reduce over-fitting. BERT uses a standard application of dropout for regularization, but ALBERT, having fewer parameters, does not apply any. Given dropout interrupts the attention mechanism that reinforces associations between words in a sentence, we hypothesis it might also be useful for reducing gendered (and potentially other) correlations.”) is performed as part of tuning the pre-trained machine learning model (Section 6 – “To separate the effect of the underlying model from that of the task data, we initialize fine-tuning (step = 0) with a frozen model, simply using it as a feature extractor, before unfreezing for steps > 0 and finetuning all layers with the STS-B training set. Both the accuracy and correlation metrics start low, and increase (or steadfastly remain zero) as fine-tuning progresses. The correlation metric remains lower for the checkpoints to which mitigation has been applied, compared to the public BERT model.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the method/system of BADJATIYA in view of KURITA to include a dropout for tuning a pre-trained model, as taught by KURITA.
One of ordinary skill would have been motivated to include a dropout for tuning a pre-trained model, in order to more effectively avoid overfitting a model during fine-tuning and/or effectively de-biasing a pre-trained model.
Regarding Claim 33, it is similar to Claim 26; thus, it is rejected under the same rationale.
Claims 27, 34, and 40 are rejected under 35 U.S.C. 103 as being unpatentable over BADJATIYA in view of KURITA, and in further view of SATHEESH (US 2022/0351039 A1).
REGARDING CLAIM 27, BADJATIYA in view of KURITA discloses the system of claim 21.
BADJATIYA in view of KURITA does not explicitly teach a user interaction to receive a trained model.
SATHEESH discloses a method/system for training natural language models, wherein the machine learning system is further configured to:
receive, via an interface (SAHEESH Fig. 8 Network Interface), the machine learning model from a client (SATHEESH Fig. 5 512 Report distilled model; Par 56 – “First user 302 and second user 304 each distill their respective local models at 510 and 514, and each report their distilled models to the central node or server 102 at 512 and 516.”); and
return, via the interface (SAHEESH Fig. 8 Network Interface), the trained machine learning model to the client (SATHEESH Fig. 5 520 Report global model; Par 56 – “Having received the model reports, central node or server 102 may construct or update the global model 318 (e.g., as described in disclosed embodiments). Central node or server 102 then reports the global model to the first user 302 and second user 304 at 520 and 522. In turn, first user 302 and second user 304 then update their respective local model based on the global model (e.g., as described in disclosed embodiments) at 524 and 526.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the method/system of BADJATIYA in view of KURITA to include receiving an updated model from a server in response to sending a model, as taught by SATHEESH.
One of ordinary skill would have been motivated to include receiving an updated model from a server in response to sending a model, in order to provide a user a more reliable model base (Par 63).
Regarding Claim 34, it is similar to Claim 27; thus, it is rejected under the same rationale.
Regarding Claim 40, it is similar to Claim 27; thus, it is rejected under the same rationale.
Conclusion
THIS ACTION IS MADE FINAL. 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 JONATHAN C KIM whose telephone number is (571)272-3327. The examiner can normally be reached Monday to Friday 8:00 AM thru 4:00 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, Andrew C Flanders can be reached at 571-272-7516. 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.
/JONATHAN C KIM/Primary Examiner, Art Unit 2655
Prosecution Insights