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 .
Status
This instant application No. 19/209,875 has claims 1-18 pending.
Priority / Filing Date
Applicant’s claim for priority of provisional application No. 63/651,210 (filed on May 23, 2024) is acknowledged. The effective filing date of this application is May 23, 2024.
Abstract
The abstract of the disclosure is acceptable for examination purposes.
Drawings
The drawings received on May 16, 2025 are acceptable for examination purposes.
Information Disclosure Statement
As required by M.P.E.P. 609(C), the Applicant’s submissions of the Information Disclosure Statements filed on 21 May 2025, 23 May 2025, and 2 July 2025 are acknowledged by the Examiner and the cited references have been considered in the examination of the claims. As required by M.P.E.P. 609 C(2), a copy of each of the PTOL-1449s initialed and dated by the Examiner is attached to the instant Office action.
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. See MPEP § 717.02 for applications subject to examination under the first inventor to file provisions of the AIA as explained in MPEP § 2159. See MPEP §§ 706.02(l)(1) - 706.02(l)(3) for applications not subject to examination under the first inventor to file provisions of the AIA . A terminal disclaimer must be signed in compliance with 37 CFR 1.321(b).
The USPTO Internet website contains terminal disclaimer forms which may be used. Please visit www.uspto.gov/forms/. The filing date of the application in which the form is filed determines what form (e.g., PTO/SB/25, PTO/SB/26, PTO/AIA /25, or PTO/AIA /26) 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 1-18 are rejected on the ground of nonstatutory double patenting over claims 1-20 of co-pending application No. 19/209,870.
Claims 1-18 of the instant application recite similar limitations and claims 1-20 of ‘870 as being compared in the table below. For the purpose of illustration, only claims 1-6 (system claims) of the instant application are compared to the claims of the patent (underlining are used to indicate conflict limitations). The remaining claims of the instant application recite different categories (i.e., method and medium claims) and are therefore not compared for simplicity purposes.
Instant Application
App. No. US 19/209,870
Claim 1
1. A computing system for training a tabular data model with localized context, comprising: one or more processors configured to execute instructions; and a non-transitory computer-readable storage medium containing instructions executable by the one or more processors for:
selecting a training data point from a set of training data points for a domain of tabular data;
identifying a subset of data points in the set of training data that form a neighborhood around the training data point;
selecting a context and a plurality of query points from the subset of data points that form the neighborhood around the training data point; and
training parameters of a tabular data model with a training batch including the context and the plurality of query points.
Claim 1
A computing system for tabular data models using localized context, comprising one or more processors configured to execute instructions; and a non-transitory computer-readable storage medium containing instructions executable by the one or more processors for:
identifying a query to apply a tabular data model to a query data point;
identifying a set of query domain data including a plurality of domain data points associated with a domain of the query;
selecting a local context of context points from the set of query domain data based on a distance of the context points to the query data point; and
applying the local context and query data point to a trained tabular data model to generate a data point classification of the query data point.
See further Barel and Moon below for mapping and motivation to combine with the claims of ‘870.
Claim 2
The computing system of claim 1, wherein identifying the subset of data points comprises selecting nearest-neighbors of the identified training data point as the neighborhood.
Claim 5
The computing system of claim 1, wherein selecting the local context comprises determining a number of nearest data points in the set of query domain data to the query data point.
Claim 3
The computing system of claim 1, wherein a number of the subset of data points varies based on the distance of data points to the training data point.
Claim 6
The computing system of claim 5, wherein the number of nearest data points is dynamically determined based on the distance of the respective data points in the set of query domain data to the query data points.
Claim 4
The computing system of claim 1, wherein the tabular data model is a transformer model.
Claim 4
The computing system of claim 1, wherein the tabular data model is a transformer architecture having an attention layer that attends to the local context.
Claim 5
The computing system of claim 1, wherein training parameters of the tabular data model comprises masking attention between the plurality of query points during application of the tabular data model.
See further Moon below for mapping and motivation to combine with the claims of ‘870.
Claim 6
The computing system of claim 1, wherein selecting the context and the plurality of query points comprises randomly assigning the subset of data points to the context or the plurality of query points.
See further Rangan below for mapping and motivation to combine with the claims of ‘870.
Although the conflicting claims are not identical, they are not patentably distinct from each other because they are substantially similar in scope and they use the similar limitations to produce the same end result of providing local context for context-based tabular classification. It is noted that the instant application applies the same proximity-based data selection logic to a training environment that the ‘870 application applies to an inference environment.
It would have been obvious to a person with ordinary skills in the art at the time of the invention was made to modify the elements of claims 1-20 of ‘161 with any combination of the cited references below to arrive at claims 1-18 of the instant application for the purpose of utilizing tabular transform to apply on proximity-based neighborhood to create a localized structure for scaling in large data tables to achieve a more resource-efficient model.
Further, it would have been obvious to a person with ordinary skills in the art at the time of the invention was made to modify or to omit the additional elements of claims 1-20 of ‘870 to arrive at claims 1-18 of the instant application because the person would have realized that the remaining element would perform the same functions as before. “Omission of element and its function in combination is obvious expedient if the remaining elements perform same functions as before.” See In re Karlson (CCPA) 136 USPQ 184, decide Jan 16, 1963, Appl. No. 6857, U.S. Court of Customs and Patent Appeals.
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.
The claimed invention in claims 1-18 are directed to a judicial exception (i.e., an abstract idea) without significantly more.
Claims 1-18 pass step 1 of the 35 U.S.C. 101 analysis since each claim is either directed to a method; a system comprising one or more processor and a non-transitory computer-readable storage medium; or a non-transitory computer-readable medium.
Claims 1, 7, and 13 recite each, in part, elements that are directed to an abstract idea (“Courts have examined claims that required the use of a computer and still found that the underlying, patent-ineligible invention could be performed via pen and paper or in a person’s mind.” Versata Dev. Group v. SAP Am., Inc., 793 F.3d 1306, 1335, 115 USPQ2d 1681, 1702 (Fed. Cir. 2015)). Each claim recites the limitations of selecting a training data point from a set of training data points…; identifying a subset of data points…that form a neighborhood…; selecting a context and a plurality of query points… The 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 (e.g., mentally selecting a training data point, mentally identifying a subset of data points, and mentally selecting a context and a plurality of query points). That is, other than reciting generic components (e.g., processor, memory, and computer-executable instructions), nothing in the claim precludes the limitations from being performed in the human mind to implement mathematical algorithms and proximity calculations per step 2A – prong 1 of the Abstract Idea Analysis.
Each claim further recites an additional step of training parameters of a tabular data model with a training batch… which is an extra-solution activity to implement a high-level functional requirement that does not prove a specific technical improvement to the computer’s internal potation. The step only uses the computer as a tool to perform the mathematical optimization relating to accuracy or efficiency of the data classification result. The claims are drafted at a high level of generality to describe what the system does (e.g., selecting data for a local context to train a generic tabular model) rather than how the computer’s hardware or software architecture is fundamentally improved. Thus, the claim does not pass step 2A – prong 2 of the Abstract Idea Analysis since each of the additional limitation(s) is no more than mere instructions to apply the exception using a generic computer component (e.g., processor, memory, and computer-executable instructions).
The extra-solution activity in step 2A - prong 2 are reevaluated in step 2B to determining if each limitation is more than what is well-understood, routine, conventional activity in the field (i.e., nearest-neighbor selection or kNN is a well-known and conventional technique in data science). The background of the limitations does not provide any indication that the computer components (e.g., processor, memory, and computer-executable instructions) are not off-the-shelf computer components. The Symantec, TLI, and OOP Techs court decisions cited in MPEP 2106.05(d)(II) indicate that mere receiving, generating, storing, determining, identifying, and transmitting of data over a network are a well-understood, routine, and conventional functions when claimed in a merely generic manner (as it is here). Accordingly, a conclusion that the claims are well-understood, routine, conventional (WURC) activity is supported under Berkheimer Option 2. For these reasons, there is no inventive concept in each claim, thus, the claims are ineligible.
Claims 2, 8, and 14 further recite in each claim …identifying the subset…comprises selecting nearest-neighbors…as the neighborhood which can be implemented in a human mind and/or with the aid of pen/paper similar to the above analysis (e.g., mentally selecting the nearest-neighbors as the neighborhood). Since kNN is a well-known and foundational mathematical algorithm in statistical analysis, adding this specific algorithm to the abstract idea does not provide an inventive concept. Thus, the claims are ineligible.
Claims 3, 9, and 15 further recite in each claim a number of the subset of data points varies based on the distance… which is a further mathematical refinement of the proximity logic. Adjusting a sample size based on a numerical threshold is a basic logical operation that does not amount to a technical improvement in computer operation. Thus, the claims are ineligible.
Claims 4, 10, and 16 further recite in each claim the tabular data model is a transformer model which merely narrows the model to a well-documented and conventional class of neural network architecture in the art. Limiting the application of an abstract idea to a specific and known computer architecture does not transform the claim in to patent-eligible subject matter. Thus, the claims are ineligible.
Claims 5, 11, and 17 further recite in each claim …training parameters…comprises masking attention between…query points… which is a standard mathematical step within transformer-based processing. This represents a specific mathematical instruction for how the model processes the abstract data and not a technical solution to a computer hardware problem. Thus, the claims are ineligible.
Claims 6, 12, and 18 further recite in each claim …selecting the context…comprises randomly assigning the subset… which is a conventional statistical method used to prevent bias in data modeling. Implementing a fundamental statistical technique on a computer is not “significantly more” than the underlying abstract idea. Thus, the claims are ineligible.
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 1-5, 7-11, and 13-17 are rejected under 35 U.S.C. 103 as being unpatentable over Barel et al. (Pub. No. US 2017/0161271, published on June 8, 2017; hereinafter Barel) in view of Moon et al. (Pub. No. US 2024/0242024, published on July 18, 2024; hereinafter Moon).
Regarding claims 1, 7, and 13, Barel clearly shows and discloses a method for training a data model with localized content (Abstract); a computing system for training a data model with localized context, comprising: one or more processors configured to execute instructions; and a non-transitory computer-readable storage medium containing instructions executable by the one or more processors for implementing the method; and a non-transitory computer-readable medium for training a data model with localized content, the non-transitory computer-readable medium comprising instructions executable by a processor for implementing the method (Figures 14-15), comprising:
selecting a training data point from a set of training data points (traversing the approximate nearest neighbor search tree generated at operation 1201 based on an individual input query of the training set of queries to determine a resultant leaf node associated with the individual input query, [0086]. It is clear that an individual query is selected from the training set of queries) for a domain (training set 815 may include a large training set of real life inquiries, [0063]);
identifying a subset of data points in the set of training data (it may be desirable to find one or more nearest neighbors in a database based on a query point, [0025]. The hashing table, based on the key, may provide pointers to one or more candidate entries for further evaluation to determine nearest neighbors for the input query, [0027]. The priority queue may maintain a list of a particular number of closest database entries to the input query. As the approximate nearest neighbor search tree is traversed, if a node provides a closer entry than any of the entries in the priority queue, the entry associated with the node may be kept and the entry associated with a farthest distance from the input query may be discarded, [0028]) that form a neighborhood around the training data point (Resultant leaf node 812, in this context, may be considered an actual nearest neighbor guess for the input query from training set 815, [0063]. It is noted that the leaf node is reached specifically by evaluating the distance of the input query against node thresholds. Because the search tree is a metric structure, the points associated with the leaf node are spatially proximate to the query wherein the points form a neighborhood around the training point);
selecting a context and a plurality of query points from the subset of data points that form the neighborhood around the training data point (for a particular leaf node, multiple candidate entries (e.g., a particular number of candidate entries such as M candidate entries) from the database that are the highest frequency entries of the frequency distribution table for the particular leaf node may be provided as candidate entries, [0087]. It is clear that, for each leaf node defining the neighborhood, only candidate entries with highest frequencies are selected); and
training parameters of a data model with a training batch including the context and the plurality of query points (Based on resultant leaf node 812 and actual nearest neighbor 813, frequency distribution table generation module 803 may populate the frequency distribution table for the current input query from training set 815 by, for example, incrementing a table entry (e.g., by 1) of the frequency distribution table associated with resultant leaf node 812 and actual nearest neighbor 813, [0063]. Incrementing a value of the table location associated with the resultant leaf node and the actual nearest neighbor, [0086], [0138]. It is noted that the system updates its parameters by incrementing a value in a frequency distribution table based on the co-occurrence of the query and its localized neighborhood entries. This iterative update using a group of queries and their local contexts is the training of parameters using a localized batch).
Moon then discloses:
the domain of tabular data (domain-specific written reports given tabular data (e.g., a financial report, an academic report, etc.), [0030]. Contextual table embeddings may be generated for any tabular data, and are not limited to a specific dimension of the table or structural format of the table (e.g., column headers, merged cells, etc.). Thus, embodiments may use a single model to generate features for any type of tabular data, [0034]); and
training parameters of the tabular data model with a training batch (The data contained in the cells of the table may be embedded by the computer program. Specifically, the computer program may be trained to represent data in each cell as a vector representation using neural network layers and pre-trained large language models, [0050]. It is noted that training, using a transformer-style model, involves batches of data where attention mechanisms related subject cells to others cells in the table).
It would have been obvious to an ordinary person skilled in the art at the time of the invention was effectively filed to incorporate the teachings of Moon with the teachings of Barel for the purpose of utilizing tabular transform to apply on proximity-based neighborhood to create a localized structure for scaling in large data tables to achieve a more resource-efficient model.
Regarding claims 2, 8, and 14, Barel further discloses identifying the subset of data points comprises selecting nearest-neighbors of the identified training data point as the neighborhood (it may be desirable to find one or more nearest neighbors in a database based on a query point, [0025]. The hashing table, based on the key, may provide pointers to one or more candidate entries for further evaluation to determine nearest neighbors for the input query, [0027]. The priority queue may maintain a list of a particular number of closest database entries to the input query. As the approximate nearest neighbor search tree is traversed, if a node provides a closer entry than any of the entries in the priority queue, the entry associated with the node may be kept and the entry associated with a farthest distance from the input query may be discarded, [0028]. Resultant leaf node 812, in this context, may be considered an actual nearest neighbor guess for the input query from training set 815, [0063]. It is noted that the leaf node is reached specifically by evaluating the distance of the input query against node thresholds. Because the search tree is a metric structure, the points associated with the leaf node are spatially proximate to the query wherein the points form a neighborhood around the training point).
Regarding claims 3, 9, and 15, Barel further discloses a number of the subset of data points varies based on the distance of data points to the training data point (As the approximate nearest neighbor search tree is traversed, if a node provides a closer entry than any of the entries in the priority queue, the entry associated with the node may be kept and the entry associated with a farthest distance from the input query may be discarded, [0028]. The database entries at traversed nodes having a minimum-distance with respect to the input query), may be maintained and updated at every node along the traversal based on input query 111. The distance between input query 111 and the database entry associated with a particular node may be any suitable distance function, metric, or distance such as a Euclidian distance, [0034]).
Regarding claims 4, 10, and 16, Moon further discloses the tabular data model is a transformer model (In step 230, the computer program may obtain contextual embeddings using a table transformer. In one embodiment, the table transformer may include an encoder that learns to generate a contextual representation of sequential input (e.g., sequence of cells in a table) and a decoder that can interpret the encoded input to generate another meaningful sequence (e.g., a sequence of words describing the table), [0051]).
Regarding claims 5, 11, and 17, Moon further discloses training parameters of the tabular data model comprises masking attention between the plurality of query points during application of the tabular data model (the computer program may be trained to obtain contextual embeddings of the table by solving a masked-cell prediction task during the pre-training step. In doing so, the computer program (e.g., the neural network layers) may learn to embed a table with masked cells. This embedding may then be used to reconstruct the table, and the computer program may evaluate how well the masked cells are reconstructed. By learning to accurately reconstruct the masked cells, the computer program learns to generate meaningful contextual tabular embeddings, [0058]).
Claims 6, 12, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Barel in view of Moon and further in view of Rangan (Pub. No. US 2012/0296891, published on November 22, 2012).
Regarding claims 6, 12, and 18, Rangan then discloses selecting the context and the plurality of query points comprises randomly assigning the subset of data points to the context or the plurality of query points (each context (e.g. each document or each word) in a corpus of information is assigned a unique and randomly generated representation called an index vector. These index vectors are sparse, high-dimensional, and ternary, which means that their dimensionality (d) is on the order of thousands, and that they consist of a small number of randomly distributed +1s and -1s, with the rest of the elements of the vectors set to 0, [0116]).
It would have been obvious to an ordinary person skilled in the art at the time of the invention was effectively filed to incorporate the teachings of Rangan with the teachings of Barel, as modified by Moon, for the purpose of analyzing documents belonging to specific concepts in a training dataset and building a model to determine whether other documents belong to one or more several concepts that are present in the training dataset to enhance document review and similarity analysis.
Relevant Prior Art
The following references are considered relevant to the claims:
Patel (Pub. No. US 2020/0311345) teaches enabling a user to extract relevant information from character-based contextual embedding of entities in the document, thus overcoming language barrier during interpretation of the document. Specifically, character-based embedding of information is performed in the document so as to derive a language-independent interpretation of the document. Such language-independent interpretation of the document further enables programs equipped with artificial intelligence to perform a variety of operations such as named entity recognition, information extraction, information retrieval, machine translation, sentiment analysis, feedback analysis, link prediction, comparison, summarization and so forth.
Dunning et al. (Pub. No. US 2008/0189232) teaches using data structured according to an indicator-based recommendation paradigm such that items to be considered for recommendation are stored in a text retrieval system, along with associated meta-data such as title and description. To these conventional characteristics are added additional characteristics known as indicators which are derived from an analysis of the usage of the system by users. This indicator-based system provides a more robust recommendation system that is able to capture a greater depth and variety of real-world relationships among items, and is able to handle data of higher relations.
Contact Information
Any inquiry concerning this communication or earlier communications from the Examiner should be directed to Son Hoang whose telephone number is (571) 270-1752. The Examiner can normally be reached on Monday – Friday (7:00 AM – 4:00 PM).
If attempts to reach the Examiner by telephone are unsuccessful, the Examiner’s supervisor, Sherief Badawi can be reached on (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 an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/SON T HOANG/Primary Examiner, Art Unit 2169
February 6, 2026