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. 18/048,807 has claims 1-13, and 20 pending based on the response filed on January 9, 2026 to withdraw claims 14-19.
Priority / Filing Date
Applicant’s claim for priority of provisional application No. 63/541,205 is acknowledged. The effective filing date for this application is September 28, 2023.
Abstract
The abstract of the disclosure is acceptable for examination purposes.
Drawings
The drawings filed on September 20, 2024 are acceptable for examination purposes.
Claim Interpretation under 35 U.S.C. 112(f)
Claim 20 is not interpreted as a 112(f) means-plus-function claim since it recites “one or more processing units configured to…” wherein “processing unit” here functions as concrete computer structure, not a nonce term. Processing units are commonly understood in the art as a structural term (e.g., CPU, GPU, processor) and not as purely functional placeholder like module or element or the like. The specification ([0130]-[0131]) expressly discloses a conventional hardware architecture with CPU, GPU, memory, and communications interface as the implementing structure for the neural network and associated processing. Further, the “configured to…” language is interpreted as stating the functional operation of known processor hardware running software and not as a means-plus limitation.
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 1-13, and 20 are directed to a judicial exception (i.e., an abstract idea) without significantly more.
Claims 1-13, and 20 pass step 1 of the 35 U.S.C. 101 analysis since each claim is either directed to a method, or a system comprising one or more processing units (i.e., CPU, GPU per [0130]-[0131] of specification) to perform the method.
Claims 1, and 20 recite each, in part, method using steps 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 limitation of identifying a plurality of data identifiers, identifying a plurality of data groups, applying at least a portion of the filter criteria, and searching the set of data groups. Each claim recites receiving an event prediction query; retrieving a plurality of documents…; generating an input vector…; processing the input vector..; and generating an event prediction outcome. The nature of the recited activities are:
Collecting data (questions, answers, news documents).
Classifying, ranking, summarizing, binning, and otherwise transforming the collected data into feature vectors.
Applying a trained neural network to predict outcomes.
Each of the claim does not tie these operations to a specific physical environment, sensor, machine or transformation of a physical article. They are purely informational operations on text and numerical values implemented on generic processors. Applicant is noted that mathematical models and statistical prediction, generic AI/ML pipelines operating on textual data, and/or mental processes (e.g., evaluation, classification, ranking) even when performed by a computer are abstract ideas per step 2A – prong 1 of the abstract idea analysis.
Under step 2A – prong 2 of the abstract idea analysis, 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) when not integrated into a specific technological improvement or practical application beyond generic computing. The claimed event prediction is a typical task for forecasting/predicting using known kinds of AI models (e.g., CPUs, GPUs, encoders/decoders, LLMs). Nothing in the claim recites a specific improvement to computer architecture (e.g., a specific technological improvement to a neural network training algorithms such as a new optimizer, new attention mechanism, or new memory structure) nor applies the predictions to control or transform a physical process (e.g., automatically adjusting machine parameters or controlling a device). The result is simply a better prediction of future events for human decision-makers which is still informational result.
Under step 2B of the abstract idea analysis, the additional elements are reevaluated to determining if each element is more than what is well-understood, routine, conventional activity (WURC) in the field. 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 WURC functions when claimed in a merely generic manner (as it is here). Each claim uses known components as building blocks in an expected way to achieve an expected informational result (e.g., better forecasts from news/questions/answers). The claims each does not recite any non-conventional hardware nor any specific novel training update rule, network topology, or numerical scheme. There is no technical problem in computer operation that is solved by a new solution in the ordered combination of claim elements. The combination describes a particular ML/LLM-based forecasting pipeline that appears to be a straightforward automation of human forecasters (i.e., collect news, read/summarize it, weigh relevance, bucket numerical answers, and predict an outcome) using known AI/ML tools. Such pipeline of tools does not supply an inventive concept without a technological improvement in the tools themselves. Accordingly, a conclusion that the claims are WURC activity is supported under Berkheimer Option 2. For these reasons, there is no inventive concept in each claim, thus, the claims are ineligible.
Claim 2 further recites processing…the documents and the…query with a first large language model to classify… which merely applying the LLM to evaluate and update model parameters. The claim does not add a particular machine, transformation or other practical application beyond improving prediction quality. Further, the claim does not recite any specific new update rule, hardware or non-WURC implementation. Thus, the claim is ineligible.
Claim 3 merely defines the event prediction outcome is a numerical response or a non-numerical response… Thus, the claim is ineligible.
Claim 4 merely defines the event prediction query to comprise a…question; a plurality of outcomes; and an event prediction period… Thus, the claim is ineligible.
Claim 5 recites the event prediction outcome is discretized into one of a plurality of …groups of numerical values… which describes segmenting questions with numeric answers from those with non-numeric answers. This is mental process of categorizing data by type and/or mathematical organization of information. Thus, the claim is ineligible.
Claim 6 merely defines the relevance score. Thus, the claim is ineligible.
Claim 7 merely uses an LLM that processes a plurality of documents and the event prediction query over a number of iterations… Thus, the claim is ineligible.
Claim 8 further recites augmenting the relevance score with a recency score which is a process being implemented in a human mind as a mental augmentation of score. Thus, the claim is ineligible.
Claim 9 further recites the subset of relevant documents is selected based on a threshold… which is a process being implemented in a human mind as a mental selection based on a threshold. Thus, the claim is ineligible.
Claim 10 merely defines each of the…documents comprises a summary… Thus, the claim is ineligible.
Claim 11 further recites the neural network is tuned… which is used to perform the abstract idea of prediction. There is improvement to the computer’s hardware or transformation of the computer into a special purpose machine that solves a specific technical program outside of the abstract realm of data analysis. Thus, the claim is ineligible.
Claim 12 further recites the neural network is trained… which is used to perform the abstract idea of prediction. There is improvement to the computer’s hardware or transformation of the computer into a special purpose machine that solves a specific technical program outside of the abstract realm of data analysis. Thus, the claim is ineligible.
Claim 13 merely defines the plurality of documents are news articles. Thus, the claim is 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 1-4, 9, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Khosla et al. (Pub. No. US 2025/0005052, filed on June 29, 2023; hereinafter Khosla) in view of Kabeya et al. (Pub. No. US 2019/0122144, published on April 25, 2019; hereinafter Kabeya).
Regarding claims 1, and 20, Khosla clearly shows and discloses a method for performing event prediction (Abstract); and a system for performing event prediction, the system comprising one or more processing units configured to perform the method (Figure 1) comprising:
receiving a query (At (1), a customer computing device 122 provides a natural language question (or prompt) to the natural language question answering service 102 (e.g., via a user interface (UI) of the customer computing device 122). This question (or prompt) may be input by a user of the customer computing devices 122 , [0059]);
retrieving a plurality of documents comprising information pertaining to the query (The aggregator component 104 may utilize partial string techniques and DPR techniques to determine the meaning of the natural language question and also determine which network-based services or computing domains, passages should be retrieved from. At (3), the aggregator component 104 retrieves the passages from the search systems 124, [0060]), each of the plurality of the documents classified on a relevance thereof to the query (The aggregator component 104 may also use a similarity score to determine which passages retrieved are relevant (e.g., not out of scope). The retrieved passages may be sent through a dense encoder and to get their dense embeddings. Scores may be generated by the aggregator component 104 for each passage in relativity to the natural language question (e.g., how well the passage is related to the question), [0045]);
generating an input vector with the query and the plurality of documents classified on relevance (the aggregator component 104 may use DPR techniques to precompute dense vector representations of text and store them in a search index. For example, the aggregator component 104 may use DPR techniques for dense representations comprehended from deep neural networks to encode text passages and questions, [0046]. The aggregator component 104 may modify and supplement the question with the retrieved passages to form a prompt. The prompt may comprise selected passages and QA pairs for the LLM component 106 to provide an answer to, [0060]. Figure 3 shows that the prompt, comprising the retrieved passages and the question, is inputted into the LLM component 106. It is clear that such prompt input is in the form of input vector or embedding);
processing the input vector with a neural network trained to determine a response to the query (The LLM component 106 may receive the prompt from the aggregator component 104, the user context (optionally) from user context component 105, and generate one or more answers based on the prompt and the user context, [0024]. The LLM component 106 may utilize this evidence pool, current natural language question, and conversional context, to answer the current question, [0063]); and
generating response of the query with the neural network (At (7), the LLM component 106 sends the generated answer and retrieved passages to the verifier component 108, [0066]).
Kabeya then discloses:
the query is an event prediction query (The event prediction system 140 is configured to estimate a probability of the target event expected to be observed at a target time from an input set by using the regression model 160 that has already been trained by the learning system 110. The event prediction system 140 may be configured to receive a query for performing a prediction process from an operator 102. The query may include or specify test event sequence data and a target timestamp that represents the target time for analysis, [0049]-[0051]);
processing the input vector with a neural network trained to determine an event prediction outcome (The input vector generation module 142 is configured to receive test event sequence data 200T to obtain an input set of data records and a target positional value, [0071]) corresponding to a response to the event prediction query (the positional value is a geographical point (x, y) representing a location, each label represents an object, and the target outcome is estimated as a probability that a target result is obtained at a target location, [0091]-[0095]);
generating the event prediction outcome of the event prediction query with the neural network (The prediction system 340 reads the regression model 360 from the model store 330, inputs the give input set to the regression model 360 to estimate the probability of the target outcome expected for the target location and returns a result for the query to the operator 302, [0107]-[0108]).
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 Kabeya with the teachings of Khosla for the purpose of predicting an outcome expected for a particular set of input data using a learning model with a plurality of functions trained to estimate a target outcome that is grounded in relevant retrieved evidence.
Regarding claim 2, Khosla further discloses:
processing the plurality of documents and the event prediction query with a first large language model to classify the plurality of documents based on relevance by generating, with the first large language model, a relevance score for each document of the plurality of documents to determining the relevance of each document (The aggregator component 104 may also use a similarity score to determine which passages retrieved are relevant (e.g., not out of scope). The retrieved passages may be sent through a dense encoder and to get their dense embeddings. Scores may be generated by the aggregator component 104 for each passage in relativity to the natural language question (e.g., how well the passage is related to the question), [0045]);
wherein the plurality of documents included in the input vector comprises a subset of relevant documents selected from the plurality of documents based on the relevance score (The passages which pass a certain threshold (e.g., greater than 0.5) may be kept while passages under a certain threshold may not be kept (e.g., less than or equal to 0.5). The passages that are not kept may be deemed out of scope by the aggregator component 104, [0045]).
Regarding claim 3, Khosla further discloses wherein the event prediction outcome is a numerical response or a non-numerical response; and wherein the non-numerical response corresponds to a multiple-choice answer or a true or false answer (The memory 210 may also include a textual NLI module 236 for determining whether the answer generated from the LLM component 106 contradicts the retrieved passages from the aggregator component 104 by using a premise and hypothesis. The textual NLI interpretation module 236 may use NLI models to determine if there is a contraction. The textual NLI interpretation module 236 may utilize NLI models to take two text sequences as input (e.g., a hypothesis and a premise), and determine whether the hypothesis is true (entailment), false (contradiction), or undetermined (neutral) given the premise, [0053]).
Regarding claim 4, Kabeya further discloses the event prediction query comprises
a text-based event prediction question (the event prediction system 140 can answer a question like “What kind of event would occur at a particular time?” by entering the input set of the data records into the plural learning models 160, [0051]);
a plurality of possible event prediction outcomes (The label may be a value “bleeding” in the case where the value describes content of the event. The label may be a key-value pair “symptom=headache”, “symptom=slight fever”, etc., in the case where the key merely describes a type of an event and the key-value pair describes whole content of the event. Also the label may be a key “Blood glucose level” that is associated with a certain value that represents degree or quantity related to the key (e.g. “high” or “85 mg/dL” for “Blood glucose level”) in the case where the key describes content of the event together with the value, [0045]. The query may include or specify test event sequence data and a target timestamp that represents the target time for analysis. An input set of data records each having a label and a timestamp is prepared from the test event sequence data. The test event sequence data may also be obtained from the event record system 150 or the event collection database 120, [0049]-[0051]); and
an event prediction period corresponding to a start time and an end time defining a valid duration based on which the event prediction outcome is generated (predicting a target outcome expected for a target positional value in an event sequence analysis system, according to an exemplary embodiment of the present invention, will be described, in which the positional value is a timestamp representing a time, each label represents an event and the target outcome is estimated as a probability that a target event is observed at a target time, [0038]).
Regarding claim 9, Khosla further discloses the subset of relevant documents is selected based on a threshold relevance score (The passages which pass a certain threshold (e.g., greater than 0.5) may be kept while passages under a certain threshold may not be kept (e.g., less than or equal to 0.5). The passages that are not kept may be deemed out of scope by the aggregator component 104, [0045]).
Claims 5-6 are rejected under 35 U.S.C. 103 as being unpatentable over Khosla in view of Kabeya in view of Hargras et al. (Pub. No. US 2022/0036221, published on February 3, 2022; hereinafter Hargras).
Regarding claim 5, Hargras then discloses the event prediction outcome is discretized into to one of a plurality of binned groups of numerical values and wherein the event prediction outcome corresponds to one of a plurality of midpoints of the plurality of binned groups (Continuous features: These are entirely numerical features, containing no blanks or any other distinct category. By default, these features will be associated with 3 linguistic labels (a higher number is configurable), [0070]-[0071]. If feature k is continuous, then the original distribution is grouped in 10 bins; then, a given bin is chosen according to the binned distribution, and the midpoint of such a bin is assigned, [0199]-[0202]).
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 Hargras with the teachings of Khosla, as modified by Kabeya, for the purpose of improving the efficiency and reliability of the neural network classifier based on a mathematical frameworks for data processing and enhancing interpretable event prediction.
Regarding claim 6, Hargras further discloses the relevance score corresponds to one of a plurality of integer bins representing the relevance of each document (The decision can be presented in a comprehensive format as shown in FIG. 11 or compact view format as shown in FIG. 12. Such formats allow the user to easily understand the drivers and their combination using linguistic labels such as Low, Medium and High, where such drivers are combined using IF-Then rules and the given decision has Pros and Cons, [0252]).
Claims 7, and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Khosla in view of Kabeya in view of Gao et al. (Pat. No. US 12073188, filed on August October 12, 2021; hereinafter Gao).
Regarding claim 7, Gao then discloses the first large language model processes the plurality of documents and the event prediction query over a number of iterations to generate a plurality of relevance scores for each document and wherein the relevance score of each document is based on the plurality of relevance scores (In the iterative self-training process, passages retrieved by the retriever are labelled as positive instances of the query description if and only if the keywords associated with the retrieved passages are used in the generation of the low-resource language keywords by the generator. This is determined through a comparison of the keywords generated using a baseline model without the particular retrieved passage against the keywords generated using the model having the input of the retrieved passage. In this manner, passages may be labelled as positive retrievals when they positively impact the keyword generation, thereby self-training the algorithm through iterations, [Column 3, [Lines 22-43]).
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 Gao with the teachings of Khosla, as modified by Kabeya, for the purpose of classifying and searching for the matching results through a natural language processing algorithm to identify similar passages across based on similar keywords amongst the items through an iterative self-tunning scheme ensure matching results continuously updated.
Regarding claim 13, Gao further discloses the plurality of documents are news articles (The passage 202 may include an item description, may be a short paragraph, a few lines, several pages, or longer. The passage 202 may include abstracts for scientific papers, scholarly articles, news articles, or other written text that may be searched for and therefore benefit from association with a string of keywords that aid users in finding the passage 202, [Column 3, [Lines 22-43], [Column 9, Line 63 – Column 10, Line 2]).
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Khosla in view of Kabeya in view of Kulkarni (Pub. No. US 2019/0340256, published on November 7, 2019).
Regarding claim 8, Kulkarni then discloses augmenting the relevance score with a recency score, wherein the recency score is determined based on a time associated with each document (Each feature of each determined search result and each candidate object is scored according to its relevance to search criteria of the search request, then those scores are weighted and combined to create a relevance score for each search result and each candidate object. In an embodiment, the relevance score for a candidate object also factors in a most recent use timestamp, [0060]).
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 Kulkarni with the teachings of Khosla, as modified by Kabeya, for the purpose of managing relevancy associated with a set of search results based on one or more recency attributes associated with past search queries to deliver the most relevant results corresponding to a user query.
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Khosla in view of Kabeya in view of Qin (Pub. No. US 2024/0346566, filed on April 12, 2023).
Regarding claim 10, Qin then discloses each of the plurality of documents comprises a summary thereof generated with a second large language model (The LLM may generate a summary of each recommended testimonial along with a link to the recommended testimonial. The summaries of the recommended testimonials along with the corresponding links are displayed to the user on the current webpage, for example, as recommended testimonials related to the current webpage, [0024]).
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 Qin with the teachings of Khosla, as modified by Kabeya, for the purpose of using retrieval augmented artificial intelligence based on the information related to a user query and associated retrieved content items to provide content recommendations to the user query.
Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Khosla in view of Kabeya in view of Portisch et al. (Pub. No. US 2025/0094707, filed on September 15, 2023; hereinafter Portisch).
Regarding claim 11, Portisch then discloses the neural network is tuned using low-rank adaptation of large language models architecture (training the language model using the subset of the directed graph may involve fine tuning the language model (e.g., by applying low-rank adaptation) to optimize the language model for a task or a domain, e.g., the domain of the subset of the knowledge graph, [0243.]).
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 Portisch with the teachings of Khosla, as modified by Kabeya, for the purpose of automatically and dynamically supplementing user prompts to large language models with information to be used by the large language model in formulating a response based on a semantic framework to provide a modified user prompt.
Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over Khosla in view of Kabeya in view of McCarson (Pub. No. US 2023/0342392, filed on June 27, 2023).
Regarding claim 12, McCarson then discloses the neural network is trained using a loss function comprising a decoder loss corresponding to accuracy of the event prediction outcome and an alignment loss corresponding to confidence in human temporal prediction (RLHF 902 is a mechanism by which reinforcement learning can be provided to one or more modules of engine 830. That is, as is known in the art, human feedback may be provided to “align” the output of a LLM (e.g., ChatGPT) to human preferences. Such RLHF 902 procedures may be used to improve performance, optimization of prompts, and overall training accuracy. That is, it is typical for LLMs to be trained using a loss function based on next-token prediction, which does not always conform to human expectations. This training may take place via actual physical humans interacting with engine 830 (and judging results in real-time), RLHF 902 might also use known question/answer forums (such as Reddit channels (subreddits), Quora, Stack Overflow, and the like), in which users have already ranked the value of answers in various ways, [0099]).
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 EEE with the teachings of Khosla, as modified by Kabeya, for the purpose of continuously monitoring correlations and adding additional relevant data sources over time to learn from historical activities and improve predictions over time autonomously or with the aid human reinforcement.
Relevant Prior Art
The following references are considered relevant to the claims:
Sharp et al. (Pub. No. US 2025/0310301) teaches AI controller may configure RAG agent to communicate with entities outside private network to leverage hybrid AI functionality. Client devices may request AI controller to send client devices daily news updates. In response, AI controller may leverage an existing, or deploy an LLM within private network. In order to retrieve the news updates, AI controller may instantiate RAG agent in connection with the Internet. This will allow RAG agent to retrieve news from various publicly accessible news sites without compromising entities or data on private network. RAG agent may send the retrieved news to the LLM at AI training center. The LLM may then include the news retrieved by RAG agent within its responses.
Rogynskyy et al. (Pub. No. US 2025/0045313) teaches a query engine can generate and/or use queries (e.g., hive queries or SQL queries) to retrieve data from data sources (e.g., data from record objects, such as object field-value pairs), a dataset of text strings regarding the record object (e.g., containing enriched and/or unenriched text strings), and/or recently generated text strings (e.g., text strings generated within a defined time period, such as within the past week). The query engine can use the queries to retrieve the data from the different sources. The query engine can aggregate or otherwise combine the retrieved data together, in some cases with the natural language query that initiated the retrieval, into a prompt.
Contact Information
Any inquiry concerning this communication or earlier communications from the Examiner should be directed to Son T. 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 March 16, 2026