DETAILED ACTION
This Office Action is responsive to request for continued examination filed January 27th, 2026.
Claims 1-2, 7-9, 11, 16-17 and 20-22 are amended, claims 13 and 21 are cancelled, claims 23 and 24 are added. Claims 1-12, 14-17, 19, 20 and 22-24 are pending and have been examined.
Any previous objections/rejections not mentioned in this Office Actions have been withdrawn by the Examiner.
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 claims the benefit of U.S. Provisional Application 63/093524, filed October 19th, 2020. Claims 1-22 have been afforded the benefit of this filing date.
Information Disclosure Statement
The information disclosure statements (IDS) submitted with the present application are being considered by the examiner.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
Claim 24 is rejected under 35 U.S.C. 112(a) as failing to comply with the written description requirement. The claim contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor at the time the application was filed, had possession of the claimed invention.
Claim 24 recites “The computer-implemented method of claim 1, wherein re-training the second version of the machine-learned model comprises:incrementally re-training the second version of the machine-learned model to align third learned embeddings for a third version of the machine-learned model with an evolved distribution of new training data that comprises neologisms, wherein the neologisms of the evolved distribution of the new training data are not included in an initial vocabulary of the first version of the machine-learned model.”
Specification paragraph [0004], “As one example, machine learning models for natural language may attempt to model semantic meaning, interrelatedness, contextual usage, etc. of a natural language (e.g., as represented by a vocabulary of tokens such as phonemes, n-grams, and/or words). However, natural languages change over time, including word additions (e.g., new acronyms, portmanteaus and neologisms), word obsolescence, and/or the semantic drift of words. This phenomenon is particularly evident in text used on the World Wide Web (e.g., in news articles, web sites, social media, etc.) which changes quickly due to fluctuations in cultural usage and current events,” and paragraph [0024] discloses “One example aspect of the present disclosure provides techniques to evolve or update a ‘vocabulary’ of entities handled by a machine learning model over time. For example, the entities can be items, locations, users, and/or natural language tokens. In particular, at each of a number of epochs or time slices, new entities (e.g., language tokens of a natural language, object and/or image classes for image classification) which occur in high frequencies in the current time slice can be added to the vocabulary while entities which appear in low frequencies can be removed, thus keeping the vocabulary size fixed while adapting to changes in entity usage, frequency, or relevance. Example tokens include phonemes, n-grams, words, subword segments, hashtags, and/or other forms of tokens,” however, the Specification fails to particularly describe how the training data is modified to account for neologisms or other “evolving” behavior (how neologisms are identified separately from semantically shifted entities) in a way that describes the best mode of the invention.
Allowable Subject Matter
Claims 1, 8 and 16 are allowed.
The following is an examiner’s statement of reasons for allowance:
U.S. Patent 11,120,225 to Tan et al. teaches incremental training or updating of machine learning models – including sentence embedding systems and sentence classifiers – with iterative training data, wherein the training data comprises changing semantic domains and construction of the training data is based on a similarity measurement, but does not teach a weighted sampling method that is further carried out on the constructed training data.
“Mining the UK Web Archive for Semantic Change Detection” by Tsakalidis et al. teaches identification of semantic shift in vocabulary for the training of machine learning models, but does not teach iterative retraining of a model with a weighted sample set of semantically changed words.
U.S. Patent Application Publication 2015/0032761 to Pasternack teaches methods of efficient weighted sampling for the training and evaluation of machine learning models, but does not teach or suggest the application of semantic shift detection and iterative retraining.
“Efficient Training Strategies for Deep Neural Network Language Models” by Schwenk et al. provides motivation for focused training and weighted sampling, but does not teach or suggest the application of semantic shift detection and iterative retraining.
“Training Region-based Object Detectors with Online Hard Example Mining” by Shrivastava et al. teaches a method of hard example mining in model training for object detection, but does not teach or suggest semantic shift detection or iterative retraining.
China invention application CN 111626335 to Liang teaches a method for training neural networks for image processing with historical loss values and ranked hard examples, but does not teach the application of semantic shift detection and iterative retraining.
With regard to claims 1 and 8, none of the prior art, alone or in combination, teach or suggest “selecting, by the computing system and based at least in part on the identified subset of entities that have undergone the semantic shift, training examples for inclusion in a biased training dataset, by performing a weighed sampling of the new training data to designate relatively greater weight to training examples that include one or more of the identified subset of semantically shifted entities,” as well as “re-training, by the computing system, the second version of the machine-learned model with an updated distribution of the new training data comprising the biased training dataset by prioritizing the one or more of the identified subset of semantically shifted entities over other entities in the new training data.”
With regard to claim 16, none of the prior art, alone or in combination, teach or suggest “identifying a subset of the online training examples that comprise semantically shifted entities as hard examples based at least in part on the respective loss values for the pre-training loss function exhibited by the machine-learned model for the online training examples,” as well as “re-training the machine-learned model using an updated distribution of new training data comprising the identified subset of online training examples that are hard examples, by prioritizing the semantically shifted entities over other entities in the new training data.”.
Any comments considered necessary by applicant must be submitted no later than the payment of the issue fee and, to avoid processing delays, should preferably accompany the issue fee. Such submissions should be clearly labeled “Comments on Statement of Reasons for Allowance.”
Claims 2-7, 9-12, 14-15, 17, 19-20 and 22-23 are allowed as being dependent from allowable claims 1, 8 and 16.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
U.S. Patent 7,822,699 to Katariya et al.
U.S. Patent 10,216,839 to Ho et al.
U.S. Patent 10,705,796 to Doyle.
U.S. Patent 11,270,159 to Gao and Chen.
U.S. Patent 12,026,624 to Gonzalez et al.
U.S. Patent 12,292,944 to Gonzalez et al.
U.S. Pre-Grant Publication 2022/0198154 to Liu et al.
U.S. Pre-Grant Publication 2020/0151604 to Johnson et al.
U.S. Pre-Grant Publication 2020-0293944 to Furukawa et al.
U.S. Pre-Grant Publication 2019/0355346 to Bellegarda.
China Invention Application Publication 111291558 to Zhao and Sun.
China Invention Application Publication 111104518 to Liu et al.
China Invention Application Publication 110674260 to Gong.
Korean Patent Application Publication KR20200040652 to Keun et al.
"Xlnet: Generalized autoregressive pretraining for language understanding” by Yang et al.
"Normalized loss functions for deep learning with noisy labels” by Ma et al.
"Learning with bad training data via iterative trimmed loss minimization” by Shen and Sanghavi.
"Denoising neural machine translation training with trusted data and online data selection” by Wang et al.
“Loss Rank Mining: A General Hard Example Mining Method for Real-Time Detectors” by Yu et al.
“Predict the Click-Through Rate and Average Cost Per Click for Keywords Using Machine Learning Methodologies” by Shi and Li.
“Interference-Masked Loss for Deep Structured Output Learning” by Guo et al.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SEAN T SMITH whose telephone number is (571)272-6643. The examiner can normally be reached Monday - Friday 8:00am - 5:00pm.
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, PIERRE-LOUIS DESIR can be reached at (571) 272-7799. 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.
/SEAN THOMAS SMITH/Examiner, Art Unit 2659
/PIERRE LOUIS DESIR/Supervisory Patent Examiner, Art Unit 2659
Prosecution Insights