DETAILED ACTION
This Office Action is in response to the remarks entered on 06/19/2026. Claims 1, 21-22 are amended. Claims 1-22 are pending.
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 .
Drawings
The drawings were received on 06/19/2026. These drawings are acceptable.
Response to Arguments
The Applicant’s arguments regarding the rejection of above claims have been fully considered.
In reference to Applicant’s arguments about:
101 rejections.
Examiner’s response:
Rejections under 35 USC 101 are withdrawn in view of the amendments and applicant’s arguments.
In reference to Applicant’s arguments about:
35 USC 103 rejections.
Examiner’s response:
Applicant arguments have been fully considered but are moot in view of new grounds of rejection.
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 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.
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 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, 6-11, 15-18, 20-22 are rejected under 35 U.S.C. 103 as being unpatentable over Lu et al (NPL submitted in IDS dated 12/09/2025 titled “The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery”- hereinafter Lu) in view of Ravid et al (US Patent No. 8,706,742 - hereinafter Ravid) and further in view of Ziegler et al (US Pub. No. 2023/0376409- hereinafter Ziegler).
Referring to Claim 1, Lu teaches a method, comprising:
determining that a document is relevant to a machine learning problem (see Lu at Abstract: “This paper presents the first comprehensive framework for fully automatic scientific discovery, enabling frontier large language models (LLMs) to perform research independently and communicate their findings”. Further at p. 2, 3rd paragraph: “The AI Scientist seamlessly performs ideation, a literature search, experiment planning, experiment iterations, manuscript writing, and peer reviewing to produce insightful papers”. Further, at p. 4 section 1. Idea Generation: “At each iteration, we prompt the language model to generate an interesting new research direction conditional on the existing archive, which can include the numerical review scores from completed previous ideas”. Therefore, this literature research for determining similar and previous existing archive of ideas is interpreted as determining a relevant document);
utilizing an autonomous research agent using a solution to the machine learning problem to conduct an experiment related to the machine learning problem based on information included in the document (see Lu at p. 2, 3rd paragraph: “The AI Scientist seamlessly performs ideation, a literature search, experiment planning, experiment iterations, manuscript writing, and peer reviewing to produce insightful papers”. Further, at p. 2, last paragraph: “Our contributions are summarized as follows: 1. We introduce the first end-to-end framework for fully automated scientific discovery in Machine Learning research, enabled by frontier LLMs (Section 3). This fully automated process includes idea generation, experiment design, execution, and visualizing and writing up the results into a full manuscript”);
evaluating the conducted experiment (see Lu at p. 4: “2. Experiment Iteration. Given an idea and a template, the second phase of The AI Scientist first executes the proposed experiments and then visualizes its results for the downstream write-up”); and
updating the solution to the machine learning problem based on the evaluation of the conducted experiment (see Lu at p. 4, 2. Experiment Iteration: “After the completion of each experiment, Aider is then given the results and told to take notes in the style of an experimental journal”. Further at p.5, 3. Paper Write-up (c)-(d): “(c) Refinement: After the previous two stages, The AI Scientist has a completed first draft, but can often be overly verbose and repetitive. To resolve this, we perform one final round of self-reflection section-by-section, aiming to remove any duplicated information and streamline the arguments of the paper. (d) Compilation: Once the LaTeX template has been filled in with all the appropriate results, this is fed into a LaTeX compiler. We use a LaTeX linter and pipe compilation errors back into Aider so that it can automatically correct any issues”. Therefore, since this Paper Write-up occurs after the Idea Generation and Experiment Iteration (as it can be seen at p. 4), it is interpreted as providing an updated solution after running the experiments).
Even though Lu implicitly teaches determining that a document is relevant to a machine learning problem, as Lu performs a literature search, Ravid explicitly teaches it (see Ravid at Col. 10: lines 13-23: “Also provided, in accordance with certain embodiments of the present invention, is a method for electronic document analysis comprising using a text classifier to classify each document in a set of documents as relevant or irrelevant to an issue”. Further at Col. 32: lines 1-6: “A suitable method for generating a transparency report is the following: Run a text Classifier, which may be generated as shown and described herein, on a document d and obtain a rank x (a number between 0 and 100) representing the relevance of document d to a particular issue”. Therefore, Ravid’s classification of relevancy is interpreted as determining that a document is relevant, and Ravid’s particular issue is interpreted as the machine learning problem).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Lu with the above teachings of Ravid by conducting experiment using a solution to a machine learning problem, as taught by Lu, and wherein the solution is based on a document being relevant to the problem, as taught by Ravid. The modification would have been obvious because one of ordinary skill in the art would be motivated to generate a transparency report showing the relevance of a document to a particular issue and demonstrating the contributions that led to be classified as relevant (as suggested by Ravid at Col. 32: lines 1-6: “A suitable method for generating a transparency report is the following: Run a text Classifier, which may be generated as shown and described herein, on a document d and obtain a rank x (a number between 0 and 100) representing the relevance of document d to a particular issue”).
However, the combination of Lu and Ravid fails to teach:
wherein the conducted experiment is evaluated using an evaluation harness isolated from the machine learning codebase, the evaluation harness defining evaluation logic for the experiment, and the solution is updated responsive to the evaluation harness determining that the experiment satisfies the scoring threshold.
Ziegler teaches, in an analogous system, wherein the conducted experiment is evaluated using an evaluation harness isolated from the machine learning codebase, the evaluation harness defining evaluation logic for the experiment, and the solution is updated responsive to the evaluation harness determining that the experiment satisfies the scoring threshold (see Ziegler at [0017]: “In embodiments, the model evaluation harness 111 uses a containerization component 110 to install repository 108a in an isolated manner (e.g., within a virtual machine, within a container). Once repository 108a is installed, the model evaluation harness 111 “blanks-out” portion(s) of the code 119 of the repository 108a, fills in those blanked-out portion(s) of code with prediction(s) generated by the code synthesis model 109, and then determines if the code 119—when modified to be filled in with these prediction(s)—passes the tests 118 of the repository 108a. In embodiments, the model evaluation harness 111 performs code blanking, prediction, and testing using each of a plurality of repositories within software repositories 108. Thus, the model evaluation harness 111 harnesses code and tests available within the software repositories 108 to evaluate the performance of the code synthesis model 109”. Further, at [0035]: “the model evaluation component 117 uses the results of step 206 to evaluate the code synthesis model. In an example, the model evaluation component 117 analyzes the validation results from step 206, to evaluate the successes and failures of the subject code synthesis model for generating code that passes validation tests provided by the software repositories 108”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Lu and Ravid with the above teachings of Ziegler by conducting experiment using a solution to a machine learning problem wherein the solution is based on a document being ranked relevant to the problem by a classification model, as taught by Lu and Ravid, and using an evaluation harness isolated from the machine learning codebase, as taught by Ziegler. The modification would have been obvious because one of ordinary skill in the art would be motivated to evaluate the “real-world” performance of the model (as suggested by Ziegler at [0017]: “Thus, the model evaluation harness 111 harnesses code and tests available within the software repositories 108 to evaluate the performance of the code synthesis model 109. When the software repositories 108 comprise a corpus of computer source code that has been found from public sources, such as Github, PyPI, NPM, and the like, the model evaluation harness 111 evaluates the “real-world” performance of the code synthesis model 109”).
Referring to Claim 2, the combination of Lu, Ravid and Ziegler teaches the method of claim 1, wherein the solution is updated in a machine learning codebase related to the machine learning problem (see Lu at p. 3, Figure 1: “The AI Scientist first invents and assesses the novelty of a set of ideas. It then determines how to test the hypotheses, including writing the necessary code by editing a codebase powered by recent advances in automated code generation”).
Referring to Claim 6, the combination of Lu, Ravid and Ziegler teaches the method of claim 1, wherein the machine learning problem is associated with a chosen metric for evaluating experiments (see Lu at p. 9 Table 1: “Performance metrics for different experimental runs across datasets”).
Referring to Claim 7, the combination of Lu, Ravid and Ziegler teaches the method of claim 6, wherein the solution to the machine learning problem is evaluated with respect to a scoring function (see Lu at p. 3: Figure 1: “The AI Scientist first invents and assesses the novelty of a set of ideas. It then determines how to test the hypotheses, including writing the necessary code by editing a codebase powered by recent advances in automated code generation. Afterward, the experiments are automatically executed to collect a set of results consisting of both numerical scores and visual summaries (e.g. plots or tables). The results are motivated, explained, and summarized in a LaTeX report”).
Referring to Claim 8, the combination of Lu, Ravid and Ziegler teaches the method of claim 6, wherein the chosen evaluation metric is testing accuracy with respect to a dataset (see Lu at p. 6 Table 1: Accuracy Column and text below “Evaluating the Automated Reviewer. To evaluate the LLM-based reviewer’s performance, we compared the artificially generated decisions with ground truth data for 500 ICLR 2022 papers extracted from the publicly available OpenReview dataset”).
Referring to Claim 9, the combination of Lu, Ravid and Ziegler teaches the method of claim 6, wherein the chosen evaluation metric is cost with respect to available compute resources (see Lu at p. 11, end of script: “"Limitations": ["The paper should address the high computational cost and explore ways to optimize it”. Further at p. 13, first paragraph: “Additionally, we provide the mean and max reviewer scores of the generated papers and the total cost of the run”).
Referring to Claim 10, the combination of Lu, Ravid and Ziegler teaches the method of claim 1, further comprising utilizing a large language model to generate one or more additional experiments based on the document and the machine learning problem (see Lu at Abstract: “We introduce The AI Scientist, which generates novel research ideas, writes code, executes experiments, visualizes results, describes its findings by writing a full scientific paper, and then runs a simulated review process for evaluation”. Lu teaches the execution of multiple experiments, which is interpreted as “additional experiments”. See also Fig. 1 below:
PNG
media_image1.png
280
588
media_image1.png
Greyscale
).
Referring to Claim 11, the combination of Lu, Ravid and Ziegler teaches the method of claim 1, wherein determining that the document is relevant to the machine learning problem includes performing a scientific literature search (see Lu at Abstract: “This paper presents the first comprehensive framework for fully automatic scientific discovery, enabling frontier large language models (LLMs) to perform research independently and communicate their findings”. Further at p. 2, 3rd paragraph: “The AI Scientist seamlessly performs ideation, a literature search, experiment planning, experiment iterations, manuscript writing, and peer reviewing to produce insightful papers”).
Referring to Claim 15, the combination of Lu, Ravid and Ziegler teaches the method of claim 1, further comprising adding the document to a corpus of relevant documents (see Ravid at Col. 29: lines 36-40: “The final results are stored in the Relevance database 140 of FIG. 1. For each document, the database 140 stores the metadata imported from the data collection database 150, together with the relevance score computed for each document”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Lu with the above teachings of Ravid by conducting experiment using a solution to a machine learning problem, as taught by Lu, and wherein the solution is based on a document being relevant to the problem, as taught by Ravid. The modification would have been obvious because one of ordinary skill in the art would be motivated to generate a transparency report showing the relevance of a document in a corpus of documents to a particular issue and demonstrating the contributions that led to be classified as relevant (as suggested by Ravid at Col. 32: lines 1-6: “A suitable method for generating a transparency report is the following: Run a text Classifier, which may be generated as shown and described herein, on a document d and obtain a rank x (a number between 0 and 100) representing the relevance of document d to a particular issue”).
Referring to Claim 16, the combination of Lu, Ravid and Ziegler teaches the method of claim 1, wherein determining that the document is relevant to the machine learning includes using a classification model to determine that the document is relevant to the problem (see Ravid at Col. 5: lines 35-49: “Certain embodiments of the present invention seek to provide automated prioritization of documents and keywords including an expert-guided system performing some or all of the following steps, suitably ordered e.g. as follows: a. An expert reviews a sample of documents, ranking them as relevant or not. b. Based on the results, the system "learns" how to score documents for relevance. c. In an iterative, self-correcting process, the system feeds additional samples to the expert. These statistically generated samples allow the system to progressively improve the accuracy of its relevance scoring. d. Once a threshold level of accuracy is achieved, the system ranks the entire collection, computing a graduated relevance score for each document”. The system that learns is interpreted as the classification model. Further, at Col. 25: lines 51-53: “Any suitable conventional method for keyword generation may be employed, e.g. as follows: In a support vector machine, the system is trained with R (relevant) and NR (not relevant) examples”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Lu with the above teachings of Ravid by conducting experiment using a solution to a machine learning problem, as taught by Lu, and wherein the solution is based on a document being ranked relevant to the problem by a classification model, as taught by Ravid. The modification would have been obvious because one of ordinary skill in the art would be motivated to generate a transparency report showing the ranked relevance of a document to a particular issue and demonstrating the contributions that led to be classified as relevant (as suggested by Ravid at Col. 32: lines 1-6: “A suitable method for generating a transparency report is the following: Run a text Classifier, which may be generated as shown and described herein, on a document d and obtain a rank x (a number between 0 and 100) representing the relevance of document d to a particular issue”).
Referring to Claim 17, the combination of Lu, Ravid and Ziegler teaches the method of claim 16, wherein the classification model is trained on a plurality of documents and their relevance to a plurality of machine learning problems (see Ravid at Col. 5: lines 35-49: “Certain embodiments of the present invention seek to provide automated prioritization of documents and keywords including an expert-guided system performing some or all of the following steps, suitably ordered e.g. as follows: a. An expert reviews a sample of documents, ranking them as relevant or not. b. Based on the results, the system "learns" how to score documents for relevance. c. In an iterative, self-correcting process, the system feeds additional samples to the expert. These statistically generated samples allow the system to progressively improve the accuracy of its relevance scoring. d. Once a threshold level of accuracy is achieved, the system ranks the entire collection, computing a graduated relevance score for each document”. The system that learns is interpreted as the classification model. Further, at Col. 25: lines 51-53: “Any suitable conventional method for keyword generation may be employed, e.g. as follows: In a support vector machine, the system is trained with R (relevant) and NR (not relevant) examples”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Lu with the above teachings of Ravid by conducting experiment using a solution to a machine learning problem, as taught by Lu, and wherein the solution is based on a document being ranked relevant to the problem by a classification model, as taught by Ravid. The modification would have been obvious because one of ordinary skill in the art would be motivated to generate a transparency report showing the ranked relevance of a document to a particular issue and demonstrating the contributions that led to be classified as relevant (as suggested by Ravid at Col. 32: lines 1-6: “A suitable method for generating a transparency report is the following: Run a text Classifier, which may be generated as shown and described herein, on a document d and obtain a rank x (a number between 0 and 100) representing the relevance of document d to a particular issue”).
Referring to Claim 18, the combination of Lu, Ravid and Ziegler teaches the method of claim 1, wherein determining that the document is relevant includes:
extracting key contributions from the document (see Ravid at Col. 10: lines 13-23: “Also provided, in accordance with certain embodiments of the present invention, is a method for electronic document analysis comprising using a text classifier to classify each document in a set of documents as relevant or irrelevant to an issue; and generating a computer display of at least one user-selected document within the set of documents, wherein at least some words in the user-selected document are differentially presented depending on their contribution to the classification of the document as relevant or irrelevant by the text classifier”. Further, at Col. 32: lines 1-20 where Ravid explain the use of a text classifier for analyzing and highlighting in different colors the words that most contribute to the document being relevant or not, which is interpreted as extracting the key contributions);
prompting the large language model to determine whether the document is relevant to the machine learning problem based in part on the extracted key contributions (see Ravid at Col. 10: lines 13-23: “Also provided, in accordance with certain embodiments of the present invention, is a method for electronic document analysis comprising using a text classifier to classify each document in a set of documents as relevant or irrelevant to an issue; and generating a computer display of at least one user-selected document within the set of documents, wherein at least some words in the user-selected document are differentially presented depending on their contribution to the classification of the document as relevant or irrelevant by the text classifier”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Lu with the above teachings of Ravid by conducting experiment using a solution to a machine learning problem, as taught by Lu, and wherein the solution is based on a document being ranked relevant to the problem by a classification model, as taught by Ravid. The modification would have been obvious because one of ordinary skill in the art would be motivated to generate a transparency report showing the ranked relevance of a document to a particular issue and demonstrating the contributions that led to be classified as relevant (as suggested by Ravid at Col. 32: lines 1-6: “A suitable method for generating a transparency report is the following: Run a text Classifier, which may be generated as shown and described herein, on a document d and obtain a rank x (a number between 0 and 100) representing the relevance of document d to a particular issue”).
Referring to Claim 20, the combination of Lu, Ravid and Ziegler teaches the method of claim 1, further comprising monitoring a plurality of document sources for one or more documents relevant to the machine learning problem (see Lu at p.5: “In a similar vein to idea generation, The AI Scientist is allowed 20 rounds to poll the Semantic Scholar API looking for the most relevant sources to compare and contrast the near-completed paper against for the related work section. This process also allows The AI Scientist to select any papers it would like to discuss and additionally fill in any citations that are missing from other sections of the paper”).
Referring to independent Claim 21 and Claim 22, they are rejected on the same basis as independent claim 1, mutatis mutandis, since they are analogous claims.
Claim 3 is rejected under 35 U.S.C. 103 as being unpatentable over Lu et al (NPL submitted in IDS dated 12/09/2025 titled “The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery”- hereinafter Lu) in view of Ravid et al (US Patent No. 8,706,742 - hereinafter Ravid), in view of Ziegler et al (US Pub. No. 2023/0376409- hereinafter Ziegler), and further in view of Gschwind (US Pub. No. 2015/0278110- hereinafter Gschwind).
Referring to Claim 3, the combination of Lu, Ravid and Ziegler teaches the method of claim 2, however, fails to teach wherein the autonomous research agent modifies a portion of the machine learning codebase while maintaining other portions of the machine learning codebase unchanged.
Gschwind teaches, in an analogous system, wherein the autonomous research agent modifies a portion of the machine learning codebase while maintaining other portions of the machine learning codebase unchanged (see Gschwind at [0277]: “As described above, an ability is provided to reuse existing binary code such that only code that is actually a candidate for optimization or modification is to be translated and stored”. Further at [0278]: “In accordance with one aspect, only a portion of the code is copied and modified instead of the entire code. Thus, in accordance with one aspect, the existing binary remains unchanged and those portions that are to be modified are copied and then modified, instead of direct code modification which may be problematic for self-referential code, as an example”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Lu, Ravid and Ziegler with the above teachings of Gschwind by conducting experiment using a solution to a machine learning problem wherein the solution is based on a document being ranked relevant to the problem by a classification model, as taught by Lu, Ravid and Ziegler, and modifying only a portion of the machine learning codebase while maintaining other portions of the machine learning codebase unchanged, as taught by Gschwind. The modification would have been obvious because one of ordinary skill in the art would be motivated to transparently patch code while maintaining correct execution in the presence of self-referential behavior, i.e., preserving the appearance that code has not been modified (as suggested by Gschwind at [0278]: “Therefore, in accordance with one aspect, a technique is provided to transparently patch code while maintaining correct execution in the presence of self-referential behavior, i.e., preserving the appearance that code has not been modified”).
Claims 4 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Lu et al (NPL submitted in IDS dated 12/09/2025 titled “The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery”- hereinafter Lu) in view of Ravid et al (US Patent No. 8,706,742 - hereinafter Ravid), in view of Ziegler et al (US Pub. No. 2023/0376409- hereinafter Ziegler), and further in view of Marinescu et al (US Pub. No. 2023/0409957- hereinafter Marinescu).
Referring to Claim 4, the combination of Lu, Ravid and Ziegler teaches the method of claim 1, further comprising:
implementing the solution to the machine learning problem by inputting the one or more code repositories into a large language model (see Lu at Figure 1: “Code via LLM & Aider”:
PNG
media_image2.png
272
594
media_image2.png
Greyscale
. Further, at p. 4: 2. Experiment Iteration: “2. Experiment Iteration. Given an idea and a template, the second phase of The AI Scientist first executes the proposed experiments and then visualizes its results for the downstream write-up. The AI Scientist uses Aider to first plan a list of experiments to run and then executes them in order”).
However, it fails to teach downloading one or more code repositories based on information included in the document.
Marinescu teaches, in an analogous system, downloading one or more code repositories based on information included in the document (see Marinescu at [0019]: “In some embodiments of the invention, the offline dataset may be a training dataset for a reinforcement learning algorithm” and “The offline dataset may comprise data from previous experiments and/or human demonstrations, and may not be updated with further environmental interaction, in real time or otherwise”. Further, at [0052]: “The agent 108 and the offline learning program 110A in the client computing device 102 and the offline learning program 110B in the server 112 can be downloaded to the client computing device 102 and the server 112 from an external computer via a network (for example, the Internet, a local area network or other, wide area network) and respective network adapters or interfaces”. Therefore, the offline dataset downloaded is interpreted as the code repositories for the agent to be able to work offline).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Lu, Ravid and Ziegler with the above teachings of Marinescu by conducting experiment using a solution to a machine learning problem wherein the solution is based on a document being ranked relevant to the problem by a classification model, as taught by Lu, Ravid and Ziegler, and downloading one or more code repositories based on information included in the document, as taught by Marinescu. The modification would have been obvious because one of ordinary skill in the art would be motivated to improve the technical field of machine learning by providing an offline method of creating a policy for a training an agent (as suggested by Marinescu at [0012]: “Therefore, the present embodiment has the capacity to improve the technical field of machine learning by providing an offline method of creating a policy for a training an agent through reinforcement learning that is robust even with unreliable reward signals, and which can tackle effectively any sequential decision-making problem with multiple objectives and user tradeoffs”).
Referring to Claim 19, the combination of Lu, Ravid and Ziegler teaches the method of claim 1, however, fails to teach further comprising updating the autonomous research agent using reinforcement learning based on outcomes of previously conducted experiments.
Marinescu teaches, in an analogous system, further comprising updating the autonomous research agent using reinforcement learning based on outcomes of previously conducted experiments (see Marinescu at [0019]: “In some embodiments of the invention, the offline dataset may be a training dataset for a reinforcement learning algorithm. The training data comprising the offline dataset may comprise static datasets of previously collected interactions of the agent”, and “The offline dataset may comprise data from previous experiments and/or human demonstrations, and may not be updated with further environmental interaction, in real time or otherwise”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Lu, Ravid and Ziegler with the above teachings of Marinescu by conducting experiment using a solution to a machine learning problem wherein the solution is based on a document being ranked relevant to the problem by a classification model, as taught by Lu, Ravid and Ziegler, wherein the classification model utilized reinforcement learning, as taught by Marinescu. The modification would have been obvious because one of ordinary skill in the art would be motivated to improve the technical field of machine learning by providing an offline method of creating a policy for a training an agent through reinforcement learning that is robust even with unreliable reward signals, and which can tackle effectively any sequential decision-making problem with multiple objectives and user tradeoffs (as suggested by Marinescu at [0012]: “Therefore, the present embodiment has the capacity to improve the technical field of machine learning by providing an offline method of creating a policy for a training an agent through reinforcement learning that is robust even with unreliable reward signals, and which can tackle effectively any sequential decision-making problem with multiple objectives and user tradeoffs”).
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Lu et al (NPL submitted in IDS dated 12/09/2025 titled “The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery”- hereinafter Lu) in view of Ravid et al (US Patent No. 8,706,742 - hereinafter Ravid), in view of Ziegler et al (US Pub. No. 2023/0376409- hereinafter Ziegler), and further in view of Unsal et al (US Pub. No. 2018/0018582- hereinafter Unsal).
Referring to Claim 5, the combination of Lu, Ravid and Ziegler teaches the method of claim 1, however, fails to teach further comprising utilizing a genetic algorithm to select one or more solutions to the machine learning problem.
Unsal teaches, in an analogous system, further comprising utilizing a genetic algorithm to select one or more solutions to the machine learning problem (see Unsal at [0007]: “Embodiments of the present disclosure provide a technical solution to the longstanding problems discussed herein, and thus solve some of the shortcomings associated with traditional electronic document preparation systems by providing methods and systems for automatically incorporating new or updated forms by utilizing machine learning using genetic programming and genetic algorithms in conjunction with training set data. In particular, embodiments of the present disclosure receive form data of or related to a new or updated form that includes data fields to be completed”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Lu, Ravid and Ziegler with the above teachings of Unsal by conducting experiment using a solution to a machine learning problem wherein the solution is based on a document being ranked relevant to the problem by a classification model, as taught by Lu, Ravid and Ziegler, and using a genetic algorithm, as taught by Unsal. The modification would have been obvious because one of ordinary skill in the art would be motivated to quickly and accurately determine an acceptable function needed to resolve a problem (as suggested by Unsal at [0007]: “Embodiments of the present disclosure utilize machine learning and genetic algorithms to quickly and accurately determine an acceptable function needed to complete the one or more data fields).
Claims 12-14 are rejected under 35 U.S.C. 103 as being unpatentable over Lu et al (NPL submitted in IDS dated 12/09/2025 titled “The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery”- hereinafter Lu) in view of Ravid et al (US Patent No. 8,706,742 - hereinafter Ravid), in view of Ziegler et al (US Pub. No. 2023/0376409- hereinafter Ziegler), and further in view of Brambila et al (US Pub. No. 2023/0140791- hereinafter Brambila).
Referring to Claim 12, the combination of Lu, Ravid and Ziegler teaches the method of claim 11, wherein determining that the document is relevant to the machine learning problem further includes: extracting key contributions based on results from the scientific literature search, and performing a semantic search on the extracted key contributions (see Lu at Abstract: “This paper presents the first comprehensive framework for fully automatic scientific discovery, enabling frontier large language models (LLMs) to perform research independently and communicate their findings”. Further at p. 2, 3rd paragraph: “The AI Scientist seamlessly performs ideation, a literature search, experiment planning, experiment iterations, manuscript writing, and peer reviewing to produce insightful papers”. Further, see Ravid at Col. 10: lines 13-23: “Also provided, in accordance with certain embodiments of the present invention, is a method for electronic document analysis comprising using a text classifier to classify each document in a set of documents as relevant or irrelevant to an issue; and generating a computer display of at least one user-selected document within the set of documents, wherein at least some words in the user-selected document are differentially presented depending on their contribution to the classification of the document as relevant or irrelevant by the text classifier”).
However, fails to teach extracting code snippets, and information from associated figures based on results from the scientific literature search, and performing a semantic search on the extracted code snippets, and information from associated figures.
Brambila teaches, in an analogous system, extracting code snippets, and information from associated figures based on results from the scientific literature search, and performing a semantic search on the extracted code snippets, and information from associated figures (see Brambila at [0048]: “After finding suitable supporting documents, i.e., those documents that have a threshold similarity with the programming task to be completed based on some similarity comparison such as natural language processing, are found, the document extractor (212) collects, or extracts, the supporting documentation associated with the completed task that is determined to be similar to the programing task to be analyzed. That is, the document extractor (212) may collect the code snippets, examples of configuration files, commits, etc. from a completed task that is deemed similar to the programming task to be completed. The document extractor (212) may collect other supporting documents as well such as user guides, manuals, etc. that relate to the completed task”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Lu, Ravid and Ziegler with the above teachings of Brambila by conducting experiment using a solution to a machine learning problem wherein the solution is based on a document being ranked relevant to the problem by a classification model, as taught by Lu, Ravid and Ziegler, and extracting code snippets, and information from associated figures based on results from the scientific literature search, and performing a semantic search on the extracted code snippets, and information from associated figures, as taught by Brambila. The modification would have been obvious because one of ordinary skill in the art would be motivated to develop cross-industry supporting material for similar tasks to be executed by extracting code snippets, files and commits that could be reusable for a similar task (as suggested by Brambila at [0001] and [0048]: “The present invention relates to programming task execution, and more specifically to generating cross-industry supporting material for a programming task to be executed”).
Referring to Claim 13, the combination of Lu, Ravid, Ziegler and Brambilla teaches the method of claim 12, further comprising utilizing a large language model to rank the results from the scientific literature search based on a predicted evaluation with respect to the scoring threshold (see Ravid at Col. 5: lines 35-49: “Certain embodiments of the present invention seek to provide automated prioritization of documents and keywords including an expert-guided system performing some or all of the following steps, suitably ordered e.g. as follows: a. An expert reviews a sample of documents, ranking them as relevant or not. b. Based on the results, the system "learns" how to score documents for relevance. c. In an iterative, self-correcting process, the system feeds additional samples to the expert. These statistically generated samples allow the system to progressively improve the accuracy of its relevance scoring. d. Once a threshold level of accuracy is achieved, the system ranks the entire collection, computing a graduated relevance score for each document”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Lu with the above teachings of Ravid by conducting experiment using a solution to a machine learning problem, as taught by Lu, and wherein the solution is based on a document being ranked relevant to the problem, as taught by Ravid. The modification would have been obvious because one of ordinary skill in the art would be motivated to generate a transparency report showing the ranked relevance of a document to a particular issue and demonstrating the contributions that led to be classified as relevant (as suggested by Ravid at Col. 32: lines 1-6: “A suitable method for generating a transparency report is the following: Run a text Classifier, which may be generated as shown and described herein, on a document d and obtain a rank x (a number between 0 and 100) representing the relevance of document d to a particular issue”).
Referring to Claim 14, the combination of Lu, Ravid, Ziegler and Brambilla teaches the method of claim 13, further comprising adding the ranked results to a queue (see Ravid at Col. 46: lines 64-67: “c. Document Issue table: a table with three columns: docID, issueID, and rank. Each row represents an individual docID which belongs to a certain issue, issueID, and has a particular rank as indicated by the corresponding classifier”. Therefore, this table with the ranked documents with their ID and ranking is interpreted as a queue).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Lu with the above teachings of Ravid by conducting experiment using a solution to a machine learning problem, as taught by Lu, and wherein the solution is based on a document being ranked relevant to the problem, as taught by Ravid. The modification would have been obvious because one of ordinary skill in the art would be motivated to generate a transparency report showing the ranked relevance of a document to a particular issue and demonstrating the contributions that led to be classified as relevant (as suggested by Ravid at Col. 32: lines 1-6: “A suitable method for generating a transparency report is the following: Run a text Classifier, which may be generated as shown and described herein, on a document d and obtain a rank x (a number between 0 and 100) representing the relevance of document d to a particular issue”).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LUIS A SITIRICHE whose telephone number is (571)270-1316. The examiner can normally be reached M-F 9am-6pm.
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, David Yi can be reached at (571) 270-7519. 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.
/LUIS A SITIRICHE/Primary Examiner, Art Unit 2126
Prosecution Insights