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 .
DETAILED ACTION
The United States Patent & Trademark Office appreciates the application that is submitted by the inventor/assignee. The United States Patent & Trademark Office reviewed the following application and has made the following comments below.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 12/30/2024 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103(a) which forms the basis for all obviousness rejections set forth in this Office action:
(a) A patent may not be obtained though the invention is not identically disclosed or described as set forth in section 102 of this title, if the differences between the subject matter sought to be patented and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art to which said subject matter pertains. Patentability shall not be negatived by the manner in which the invention was made.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103(a) are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1, 2, 7, 9-10, and 13-19 are rejected under 35 U.S.C. 103(a) as being unpatentable over Rennie et al. (U.S. Patent Pub. No. 2026/0195332, hereafter referred to as Rennie) in view of Kandhari (U.S. Patent Pub No. 20260111682 A1, hereafter referred to as Kandhari).
Regarding Claim 1, Rennie teaches a method (Paragraph [0014], Rennie teaches a question answering method.) implemented by one or more processors (Paragraph [0015], Rennie teaches implementing a question answering method executable on one or more processors.), the method comprising: for each of a plurality of content items (Paragraph [0007], Rennie teaches generating answers to queries from a collection of data. The system receives a query and retrieves a plurality of document segments based on relevance to the query. One or more portions of the answer are associated with corresponding sub-segments of the document segments. Each subsegment may include a word, phrase, or sentence within a document segment. The Examiner interprets document segments and/or sub-segments to be a plurality of content items.), (Paragraph [0090], Rennie teaches for training the extractive QA module, providing a dataset of questions and corresponding document segments with identification (annotation) of whether an answer appears in the segment and the start and end of such an answer are provided.), as a training instance for training a generative model (Paragraph [0090], Rennie teaches training the module’s transformer makes use of conventional iterative updating of model parameters, for example, using gradient updates using small batches of the annotated question-segment pairs.).
Rennie does not explicitly disclose generating a respective plurality of labels each associated with one of a plurality of parts of said content item, each said label indicating whether a respective query is answerable based on information contained in said part of the content item.
Kandhari is in the same field of art of determining whether or not a query/question is answerable based on content items. Further, Kandhari teaches generating a respective plurality of labels (Paragraphs [0011], [0016], Kandhari teaches each user question is tagged with a feedback label indicating whether the question was answered, unanswered, requires revision, or any combination thereof.) each associated with one of a plurality of parts of said content item (Paragraph [0008], Kandhari teaches determining if the user question is answerable by an answer selected from the curated question and answer set based on comparing a threshold distance to a distance between the user question and/or each question from the curated question and answer set in the vector embedding space, using the LLM. The Examiner interprets the curated question and answer set to be a “content item” with each question from the curated question and answer set being the “parts” of said content item.), each said label indicating whether a respective query is answerable (Paragraphs [0011], [0016], Kandhari teaches each user question is tagged with a feedback label indicating whether the question was answered, unanswered, requires revision, or any combination thereof.) based on information contained in said part of the content item (Paragraphs [0008], [0013-14], Kandhari teaches the relevant content is determined based on the distance between a speculative content item and/or the selected best answer in the vector embedding space. The large language model (LLM) compares a first threshold distance to an actual distance between the user question and each question from a curated question and answer set in the vector embedding space. If the actual distance is within the first threshold distance, the LLM generates and outputs an answer. If the actual distance falls outside a threshold distance, the user’s question is tagged as unanswerable.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Rennie by labeling the input queries/questions to indicate whether the query was answered or unanswered that is taught by Kandhari, to make the invention that processes the tagged/labeled user questions to identify recurring patterns in unanswered questions; thus, one of ordinary skilled in the art would be motivated to combine the references to adjust the curated question and answer set to address identified gaps, improve coverage, or enhance the relevance of future responses (Kandhari, Paragraph [0016]). In addition, associating answers with content items (source attribution) encourages the LLM to be faithful to the passages it cites, thereby improving correctness and reducing hallucinations (Rennie, Paragraph [0036]).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
In regards to Claim 2, Rennie in view of Kandhari discloses the method of claim 1, wherein each said label takes one of a plurality of values (Paragraphs [0011], [0033], Kandhari discloses each user question is tagged with a feedback label indicating whether the question was answered, unanswered, requires revision, or any combination thereof. The method includes determining if the user question is answerable by an answer selected from the curated question and answer set, based on comparing a threshold distance to an actual distance between the user question and each question in the curated question and answer set in the vector embedding space using the LLM.), the plurality of values comprising: a first value indicating that the query cannot be answered based on said information contained in the respective part of said content item (Paragraph [0014], Kandhari teaches tagging a user’s question as unanswerable if its actual distance falls outside a first threshold distance. The Examiner interprets the “distance” between the user question and each question in the curated question/answer set to be the “value” indicating that the query cannot be answered.); and a second value indicating that the query can be answered based on said information contained in the respective part of said content item (Paragraph [0013], Kandhari teaches the LLM compares a first threshold distance to an actual distance between the user question and each question from a curated question and answer set in the vector embedding space. The method includes determining whether the actual distance is within the first threshold distance. If the actual distance is determined to be within the first threshold distance, the method further includes automatically generating a best answer as a responsive communication to the user as an output of the LLM. The Examiner interprets the “distance” between the user question and each question in the curated question/answer set to be the “value” indicating that the query can be answered.).
In regards to Claim 7, Rennie in view of Kandhari discloses the method of claim 1, wherein providing at least the plurality of parts of the content item as the training instance comprises providing the complete content item (Paragraphs [0090], [0009], Rennie teaches the dataset of questions and corresponding document segments with identification of whether an answer appears are provided for training. The Examiner interprets a document segment to be a “complete content item” since the claim is silent to the meaning of “complete content item.” In addition, Rennie discusses breaking the segmented document into “sub-segments” (words, phrases, sentences) and therefore the “segmented document” is being interpreted as the complete content item.)
In regards to Claim 9, Rennie in view of Kandhari discloses the method of claim 1, wherein for one or more of the plurality of content items the method comprises: for one or more of the plurality of parts of said one or more of the plurality of content items, generating contextual information (Paragraph [0015], Kandhari teaches the contextual appropriateness is determined by considering the context of the user’s specific data.) indicative of an event occurring in said part of said content item (Paragraph [0126], Kandhari teaches the large language model analyzes the user question to detect trigger words indicative of an adverse event.).
In regards to Claim 10, Rennie in view of Kandhari discloses the method of claim 9, wherein the query relates to said event (Paragraph [016], Kandhari teaches detecting trigger words indicative of an adverse event by analyzing the user’s question. The Examiner interprets the question (query) relates to an “event” since it contains words indicative of an adverse event.).
In regards to Claim 13, Rennie in view of Kandhari discloses the method of claim 1, wherein the generative model is a large language model (LLM) (Abstract, Paragraph [0072], Kandhari teaches a large language model (LLM).).
In regards to Claim 14, Rennie in view of Kandhari discloses the method of claim 1, wherein the generative model is a multimodal generative model configured to process input in a plurality of formats (Paragraphs [0078], [0123], Kandhari teaches the voice/text input 206 may be a multimodal data capture mechanism configured to enable the user to submit queries through spoken and/or typed natural language. The voice/text input 206 may transmit the tokenized inputs to the content delivery platform 102, where the AI engine 120 may communicate with the large language model 146 to process the user question 124.).
In regards to Claim 15, Rennie in view of Kandhari discloses the method of claim 14, wherein the plurality of formats comprise one or more video formats and/or one or more audio formats (Paragraph [0078], Kandhari teaches the voice/text input may include a speech-to-text converter and/or a natural language preprocessing module adapted to convert audio-based inputs into tokenized linguistic units. Under Broadest Reasonable Interpretation (BRI), the Examiner interprets “and/or” to mean only one of either video format or audio format is required to meet the claim limitation.).
In regards to Claim 16, Rennie in view of Kandhari discloses the method of claim 14, wherein the plurality of content items comprise video and/or audio content (Paragraph [0008], Kandhari teaches the approved content includes a video, an audio, etc. Under Broadest Reasonable Interpretation (BRI), the Examiner interprets “and/or” to mean only one of either video content or audio content is required to meet the claim limitation.).
In regards to Claim 17, Rennie in view of Kandhari discloses the method of claim 1, wherein the generative model supports bidirectional streaming (Paragraphs [0100], [0015], [0127], Fig. 1, Kandhari teaches a feedback loop to support dynamic feedback. The feedback loop may help update the curated topic set and the curated question and answer set based on user engagement, content results, and input from the regulatory body.).
In regards to Claim 18, Rennie discloses a system (Abstract, Rennie teaches a question answering system.) comprising: one or more processors (Paragraph [0015], Rennie teaches one or more processors.); and memory (Paragraph [0015], Rennie teaches at least one data storage system (including volatile and/or non-volatile memory and/or storage elements).) storing computer readable instructions that, when executed by the one or more processors, cause the one or more processor to be operable to (Paragraph [0167], Rennie teaches a computing system executing suitable software instructions. Software may include one or more computer programs that execute on one or more programmed or programmable computing system.): for each of a plurality of content items (Paragraph [0007], Rennie teaches generating answers to queries from a collection of data. The system receives a query and retrieves a plurality of document segments based on relevance to the query. One or more portions of the answer are associated with corresponding sub-segments of the document segments. Each subsegment may include a word, phrase, or sentence within a document segment. The Examiner interprets document segments and/or sub-segments to be the plurality of content items.), (Paragraph [0090], Rennie teaches for training the extractive QA module, providing a dataset of questions and corresponding document segments with identification (annotation) of whether an answer appears in the segment and the start and end of such an answer are provided.), as a training instance for training a generative model (Paragraph [0090], Rennie teaches training the module’s transformer makes use of conventional iterative updating of model parameters, for example, using gradient updates using small batches of the annotated question-segment pairs.).
Rennie does not explicitly disclose generat
Kandhari is in the same field of art of determining whether or not a query/question is answerable based on content items. Further, Kandhari teaches generat(Paragraphs [0011], [0016], Kandhari teaches each user question is tagged with a feedback label indicating whether the question was answered, unanswered, requires revision, or any combination thereof.) each associated with one of a plurality of parts of said content item (Paragraph [0008], Kandhari teaches determining if the user question is answerable by an answer selected from the curated question and answer set based on comparing a threshold distance to a distance between the user question and/or each question from the curated question and answer set in the vector embedding space, using the LLM. The Examiner interprets the “curated question and answer set” to be the “content item” with each question from the set being a “part” of said content item.), each said label indicating whether a respective query is answerable (Paragraphs [0011], [0016], Kandhari teaches each user question is tagged with a feedback label indicating whether the question was answered, unanswered, requires revision, or any combination thereof.) based on information contained in said part of the content item (Paragraphs [0008], [0013-14], Kandhari teaches the relevant content is determined based on the distance between a speculative content item and/or the selected best answer in the vector embedding space. The large language model (LLM) compares a first threshold distance to an actual distance between the user question and each question from a curated question and answer set in the vector embedding space. If the actual distance is within the first threshold distance, the LLM generates and outputs an answer. If the actual distance falls outside a threshold distance, the user’s question is tagged as unanswerable.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Rennie by labeling the input queries/questions to indicate whether the query was answered or unanswered that is taught by Kandhari, to make the invention that processes the tagged/labeled user questions to identify recurring patterns in unanswered questions; thus, one of ordinary skilled in the art would be motivated to combine the references to adjust the curated question and answer set to address identified gaps, improve coverage, or enhance the relevance of future responses (Kandhari, Paragraph [0016]). In addition, associating answers with content items (source attribution) encourages the LLM to be faithful to the passages it cites, thereby improving correctness and reducing hallucinations (Rennie, Paragraph [0036]).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
In regards to Claim 19, Rennie discloses a non-transitory computer readable medium containing computer-readable instructions that, when executed by a computer, cause the computer to (Paragraph [0015], Rennie teaches a non-transitory computer readable media programmed with instructions, executable on one or more processors of a computing system to implement a method.): for each of a plurality of content items (Paragraph [0007], Rennie teaches generating answers to queries from a collection of data. The system receives a query and retrieves a plurality of document segments based on relevance to the query. One or more portions of the answer are associated with corresponding sub-segments of the document segments. Each subsegment may include a word, phrase, or sentence within a document segment. The Examiner interprets document segments and/or sub-segments to be the plurality of content items.), respective query and the plurality of labels (Paragraph [0090], Rennie teaches for training the extractive QA module, providing a dataset of questions and corresponding document segments with identification (annotation) of whether an answer appears in the segment and the start and end of such an answer are provided.), as a training instance for training a generative mode (Paragraph [0090], Rennie teaches training the module’s transformer makes use of conventional iterative updating of model parameters, for example, using gradient updates using small batches of the annotated question-segment pairs.).
Rennie does not explicitly disclose generate a respective plurality of labels each associated with one of a plurality of parts of said content item, each said label indicating whether a respective query is answerable based on information contained in said part of the content item.
Kandhari is in the same field of art of determining whether or not a query/question is answerable based on content items. Further, Kandhari teaches generate a respective plurality of labels (Paragraphs [0011], [0016], Kandhari teaches each user question is tagged with a feedback label indicating whether the question was answered, unanswered, requires revision, or any combination thereof.) each associated with one of a plurality of parts of said content item (Paragraph [0008], Kandhari teaches determining if the user question is answerable by an answer selected from the curated question and answer set based on comparing a threshold distance to a distance between the user question and/or each question from the curated question and answer set in the vector embedding space, using the LLM.), each said label indicating whether a respective query is answerable (Paragraphs [0011], [0016], Kandhari teaches each user question is tagged with a feedback label indicating whether the question was answered, unanswered, requires revision, or any combination thereof.) based on information contained in said part of the content item (Paragraphs [0008], [0013-14], Kandhari teaches the relevant content is determined based on the distance between a speculative content item and/or the selected best answer in the vector embedding space. The large language model (LLM) compares a first threshold distance to an actual distance between the user question and each question from a curated question and answer set in the vector embedding space. If the actual distance is within the first threshold distance, the LLM generates and outputs an answer. If the actual distance falls outside a threshold distance, the user’s question is tagged as unanswerable.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Rennie by labeling the input queries/questions to indicate whether the query was answered or unanswered that is taught by Kandhari, to make the invention that processes the tagged/labeled user questions to identify recurring patterns in unanswered questions; thus, one of ordinary skilled in the art would be motivated to combine the references to adjust the curated question and answer set to address identified gaps, improve coverage, or enhance the relevance of future responses (Kandhari, Paragraph [0016]). In addition, associating answers with content items (source attribution) encourages the LLM to be faithful to the passages it cites, thereby improving correctness and reducing hallucinations (Rennie, Paragraph [0036]).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claims 3 and 11-12 are rejected under 35 U.S.C. 103(a) as being unpatentable over Rennie et al. (U.S. Patent Pub. No. 2026/0195332, hereafter referred to as Rennie) in view of Kandhari (U.S. Patent Pub No. 2026/0111682, hereafter referred to as Kandhari) in further view of Jedema et al. (U.S. Patent No. 12,579,174, hereafter referred to as Jedema).
Regarding Claim 3, Rennie in view of Kandhari discloses the method of claim 2.
Rennie in view of Kandhari does not explicitly disclose wherein the plurality of values comprises: a third value indicating that the query cannot be answered based on said information contained in the respective part of said content item, but can potentially be answerable based on information contained in another one of the plurality of parts of said content item.
Jedema is in the same field of art of determining whether a content item(s) is/are sufficient for a language model to generate a response to an input (user request or query). Further, Jedema teaches wherein the plurality of values comprises: a third value indicating that the query cannot be answered based on said information contained in the respective part of said content item (Col. 9, lines 33-43, Jedema teaches determining a completeness metric representing that the answer data corresponding to the question data, can be determined using two or more of the document results. For example, a document result may be partially relevant to the question, yet contain insufficient information to generate the correct answer.), but can potentially be answerable based on information contained in another one of the plurality of parts of said content item (Col. 9, lines 33-55, Jedema teaches the completeness component may determine that the information required to answer the question data (or generate the answer data) is spread among different documents in the set of document results.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Rennie in view of Kandhari by determining a “completeness metric” for the query and respective content item(s) that is taught by Jedema, to make the invention that estimates the degree that the set of retrieved documents (content items) contain/s sufficient information to answer the question asked by the user; thus, one of ordinary skilled in the art would be motivated to combine the references to increase the accuracy while decreasing cost and latency of a generative model (Jedema, Col. 5, lines 11-18).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
In regards to Claim 11, Rennie in view of Kandhari disclose the method of claim 1, further comprising: using said training instance (Paragraph [0090], Rennie teaches for training of the extractive question-answering (QA) module, a dataset of questions and corresponding document segments with identification (annotation) of whether an answer appears in the segment and the start and end of such answer are provided.), training the generative model to be able to generate, responsive to an input query and an input content item (Abstract, Claim 1, Rennie teaches the extraction stage receives a query and retrieves a plurality of document segments from the collection of data based on relevance to the query and the generative stage generates an answer based on the retrieved document segments. The generative stage associates one or more portions of the answer with the corresponding sub-segments of the document segments.), generative output (Paragraph [0159], Rennie teaches a generative LLM is trained to output iteratively optimized generative output.)
Rennie in view of Kandhari does not explicitly disclose (generative output) indicative of whether the input query is answerable based on at least part of the input content item.
Jedema is in the same field of art of determining whether a content item(s) is/are sufficient for a language model to generate a response to an input (user request or query). Further, Jedema teaches (generative output) indicative of whether the input query is answerable based on at least part of the input content item (Col. 8, lines 32-44, Jedema teaches if the probability of a document result being sufficient for answering the question data satisfies the condition (e.g., exceeds threshold answerability value), then the answerability metric may be “yes”, “true” or other value indicating that the question data is answerable using the corresponding document result. If the probability of a document result being sufficient for answering the question data fails to satisfy the condition (e.g., below a threshold answerability value), then the answerability metric may be “no”, “false” or other value indicating that the question data is not answerable using the corresponding document result.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Rennie in view of Kandhari by providing an indication of answerability of a user query using a specific content item, such as yes/true and no/false that is taught by Jedema, to make the invention that uses a condition such as a threshold answerability value to determine whether or not a question is answerable and indicate/output the metric; thus, one of ordinary skilled in the art would be motivated to combine the references to mitigate LLM hallucinations by encouraging the LLM to be faithful to the passage it cites, thereby improving correctness and reducing hallucinations (Rennie, Paragraph [0036]).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
In regards to Claim 12, Rennie in view of Kandhari in further view of Jedema discloses the method of claim 11, wherein training the generative model comprises: training the generative model to be able to update the generative output (Paragraph [0090], Rennie teaches training the QA module’s transformer using gradient updates (e.g., using backpropagation) using small batches of the annotated question segment-pairs.) responsive to the generative model processing a part of the input content item containing information that enables the input query to be answered (Paragraphs [0104], [0105], Rennie teaches determining a first subset of matching document segments for a question, and then determining a second subset that is a subset of the first subset (i.e., further narrowing down the number of candidate segments.) The result of the Detailed Match may be a single document segment and subsequence identified as the answer based on the output of the Detailed Match transformer.).
Claims 4 and 5 are rejected under 35 U.S.C. 103(a) as being unpatentable over Rennie et al. (U.S. Patent Pub. No. 2026/0195332, hereafter referred to as Rennie) in view of Kandhari (U.S. Patent Pub No. 2026/0111682, hereafter referred to as Kandhari) in further view of Kozkov et al. (U.S. Patent Pub. No. 2026/0094442, hereafter referred to as Kozkov).
Regarding Claim 4, Rennie in view of Kandhari discloses the method of claim 1, wherein at least one of the plurality of content items comprises a video (Paragraph [0008], Kandhari teaches the approved content (received by the LLM) includes a video. The LLM extracts one or more topics from the approved content).).
Rennie in view of Kandhari does not explicitly disclose each one of the plurality of parts of said video comprising a plurality of frames of video.
Kozkov is in the same field of art of identifying relevant media content in response to a received prompt or query. Further, Kozkov discloses each one of the plurality of parts of said video comprising a plurality of frames of video (Paragraphs [0034-35], Kozkov teaches segmenting the video into a sequence of video segments or “chunks.” The language model can treat the video segment as a sequence of images (e.g., video frames).).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Rennie in view of Kandhari by segmenting the content item (video) into chunks of frames that is taught by Kozkov, to make the invention that breaks the content item into segments that can each be analyzed by a vision language model; thus, one of ordinary skilled in the art would be motivated to combine the references to generate vector representations for the individual video segments (chunks) which may be stored to allow for more accurate searching (when the user wants to locate content relevant to a specific topic), particularly when a video file includes many segments with different content (Kozkov, Paragraph [0036]).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
In regards to Claim 5, Rennie in view of Kandhari discloses the method of claim 1, wherein at least one of the plurality of content items comprises a video (Paragraph [0008], Kandhari teaches the approved content (received by the LLM) includes a video. The LLM extracts tone or more topics from the approved content.).
Rennie in view of Kandhari does not explicitly disclose each one of the plurality of parts of said video comprising a plurality of frames of video.
Kozkov is in the same field of art of identifying relevant media content in response to a received prompt or query. Further, Kozkov discloses each one of the plurality of parts of said video comprising a respective frame of said video (Paragraph [0042], Kozkov teaches a user can capture an image (or video) of a part and provide the image (or video) along with a prompt, with the search vector being generated based on both the text of the prompt and the image features extracted from the image (or video), which can help locate relevant content that may include images or video of that part, potentially being user for (or in) the task or issue at hand.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Rennie in view of Kandhari by segmenting the content item (video) into single frame “segments” or images that is taught by Kozkov, to make the invention that identifies a single representative video frame from the content item to provide to the user; thus, one of ordinary skilled in the art would be motivated to combine the references to quickly and easily locate relevant content that is highly specific to the user prompt, rather than providing a large chunk (longer video segment) (Kozkov, Paragraphs [0042-43]) .
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claim 6 is rejected under 35 U.S.C. 103(a) as being unpatentable over Rennie et al. (U.S. Patent Pub. No. 2026/0195332, hereafter referred to as Rennie) in view of Kandhari (U.S. Patent Pub No. 2026/0111682, hereafter referred to as Kandhari) in view of Kozkov et al. (U.S. Patent Pub. No. 2026/0094442, hereafter referred to as Kozkov) in further view of Lakhotia et al. (U.S. Patent Pub. No. 2026/0011149, hereafter referred to as Lakhotia).
Regarding Claim 6, Rennie in view of Kandhari in further view of Kozkov discloses the method of claim 5.
Rennie in view of Kandhari in further view of Kozkov does not explicitly disclose wherein consecutive ones of the plurality of parts are spaced apart in time in said video, such that each said respective frame is separated from respective frames of other ones of the plurality of parts by one or more other frames of said video.
Lakhotia is in the same field of art of receiving a query related to an event in a video, selecting relevant portions of the video, and applying the selected video data and related sub-query to a video model. Further, Lakhotia discloses wherein consecutive ones of the plurality of parts are spaced apart in time in said video (Paragraph [0035], Lakhotia teaches the system can segment the video data into segments and/or chunks based at least on segmentation parameters and/or a frame sampling parameter (e.g., uniform frame intervals).), such that each said respective frame is separated from respective frames of other ones of the plurality of parts by one or more other frames of said video (Paragraphs [0035], [0062], [0064], Lakhotia teaches segmenting video data based on a frame sampling parameter such as uniform frame intervals. The frame sampling parameter can correspond to how frames are selected (e.g., every N frames, keyframes only, etc.).).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Rennie in view of Kandhari in further view of Kozkov by sampling the frames of the video based on a frame sampling parameter (e.g., every N frames) that is taught by Lakhotia, to make the invention that selects for example, every Nth frame within each video segment; thus, one of ordinary skilled in the art would be motivated to combine the references to reduce the redundant data in static scenes by eliminating frames with little motion/density (Lakhotia, Paragraph [0063]).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claim 8 is rejected under 35 U.S.C. 103(a) as being unpatentable over Rennie et al. (U.S. Patent Pub. No. 2026/0195332, hereafter referred to as Rennie) in view of Kandhari (U.S. Patent Pub No. 2026/0111682, hereafter referred to as Kandhari) in further view of Lakhotia et al. (U.S. Patent Pub. No. 2026/0011149, hereafter referred to as Lakhotia).
Regarding Claim 8, Rennie in view of Kandhari discloses the method of claim 1.
Rennie in view of Kandhari does not explicitly disclose wherein consecutive ones of the plurality of parts are spaced apart in time in said content item, such that the plurality of parts comprise discontinuous parts of said content item, and wherein providing at least the plurality of parts of the content item as the training instance comprises providing said discontinuous parts of said content item.
Lakhotia is in the same field of art of receiving a query related to an event in a video, selecting relevant portions of the video, and applying the selected video data and related sub-query to a video model. Further, Lakhotia discloses wherein consecutive ones of the plurality of parts are spaced apart in time in said content item (Paragraph [0035], Lakhotia teaches the system can segment the video data into segments and/or chunks based at least on segmentation parameters and/or a frame sampling parameter (e.g., uniform frame intervals).), such that the plurality of parts comprise discontinuous parts of said content item (Paragraphs [0035], [0063-64], Lakhotia teaches adaptive segmentation based at least on activity levels detected within the video segment. For example, the sampling rate may be increased in segments with high motion or high density and the rate may be reduced in segments with minimal activity.), and wherein providing at least the plurality of parts of the content item as the training instance comprises providing said discontinuous parts of said content item (Paragraphs [0103-104], Lakhotia teaches training/updating the model by modifying or updating one or more parameters (weights and/or biases) responsive to evaluating estimated outputs of the model (e.g., generated in response to receiving training examples in a training dataset, such as a training dataset including annotated queries, video-label pairs, etc.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Rennie in view of Kandhari by adjusting the sampling rate of the content item (video) depending on the activity levels in the video segment that is taught by Lakhotia, to make the invention that increases the sampling rate in segments with high motion or event density and reduces the sampling rate in segments with minimal activity; thus, one of ordinary skilled in the art would be motivated to combine the references since the dynamic adjustment can improve temporal resolution where needed (high density segments) while reducing redundant data in static scenes (Lakhotia, Paragraph [0063]).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Pertinent Prior Art
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Jansen (U.S. Patent No. 2026/0093722) teaches a system and method using retrieval-augmented generation (RAG) in a LLM having additional content to enhance relevance of LLM generated content. The LLM outputs a “true”/”false” indication of its ability to confidently answer the user question and a probability score, indicating how sure the model is in its output. Based on this output, the system either lets the LLM answer the question normally or the system prompts the LLM to answer with several alternate answers based on the given content. The incorporation of self-evaluation helps to reduce hallucinations and provide more trustworthy and accurate responses.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SYDNEY L BLACKSTEN whose telephone number is (571)272-7120. The examiner can normally be reached 8:30am-4:30pm.
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, Oneal Mistry can be reached at 313-446-4912. 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.
/SYDNEY L BLACKSTEN/Examiner, Art Unit 2674
/ONEAL R MISTRY/Supervisory Patent Examiner, Art Unit 2674
Prosecution Insights