DETAILED ACTION
1. This office action is in response to the preliminary amendment filed on 5/5/2023. Claims 1-14 are pending and are examined.
Notice of Pre-AIA or AIA Status
2. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement PTO-1449
3. The information Disclosure Statement submitted by applicant on 6/23/2023 has been considered. The submission is in compliance with the provisions of 37 CFR 1.97. Form PTO-1449 is signed and attached hereto.
Claim Rejections- 35 U.S.C. § 101
35 U.S.C. § 101 reads as follows:
Whoever invents or discovers any new or useful process, machine, manufacture, or combination of matter, or any new and useful improvements thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
4. Claim 1 is rejected under 35 U.S.C. § 101 because the claimed invention is directed to a judicial exception (an abstract idea) without significantly more.
Step 1: Statutory Category
Claim 1 is directed to a “method,” which is a statutory category of invention (process) under 35 U.S.C. § 101. However, even though it falls within a statutory category, it must also be directed to patent-eligible subject matter.
Step 2A – Prong One: Judicial Exception
The claim recites abstract ideas in the form of mathematical concepts and mental processes, including:
Mathematical concepts:
Transforming data records via a generator network.
Mapping data to outputs via a task network.
Creating “saliency records” that represent feature importance — a mathematical representation of data contributions.
Classifying records via a discriminator network.
Evaluating classification accuracy using a “transfer cost function.”
Optimizing network parameters based on evaluation metrics.These are all mathematical algorithms and statistical analyses.
Mental processes / organization of information:
Pooling saliency records, sampling them, and classifying them as belonging to one category or another is an abstract categorization process that could be performed mentally or by generic computing.
Accordingly, the claim is directed to abstract ideas — mathematical concepts and mental processes.
Step 2A – Prong Two: Integration into a Practical Application
The claim does not integrate the abstract idea into a practical application:
The recited “generator network,” “task network,” “discriminator network,” and “transfer cost function” are generic AI/ML components executing on unspecified processors.
The claim does not describe a specific technical improvement to computer functionality or a particular machine architecture.
The steps are described at a results-oriented level (“optimize parameters to worsen/improve evaluation”) without reciting how these are implemented in a non-conventional manner.
No specific hardware, data encoding, training pipeline constraints, or protocol modifications are claimed that tie the abstract idea to a concrete technological application.
Thus, the abstract idea is not integrated into a practical application as required under MPEP § 2106.04(d).
Step 2B: Inventive Concept
The additional elements in Claim 1, individually and in combination, do not amount to “significantly more” than the abstract idea:
The claim uses generic computing components to perform conventional machine learning tasks: data transformation, mapping, classification, evaluation, and parameter optimization.
“Saliency records” are a known data representation in AI/ML (feature attribution maps), and using them in classification is a conventional ML interpretability technique.
Applying a “predefined transfer cost function” to classification accuracy is a mathematical optimization step that is routine in ML training.
Adversarial optimization of generator and discriminator networks is a well-understood, routine, and conventional ML practice (GAN training), and the claim does not recite a non-conventional adaptation that changes the operation of the computer or network itself.
Therefore, the claim lacks an inventive concept sufficient to transform the abstract idea into patent-eligible subject matter.
5. Conclusion
Because Claim 1 is directed to abstract ideas (mathematical concepts and mental processes) and does not include additional elements that amount to significantly more than the abstract ideas, Claim 1 is ineligible subject matter under 35 U.S.C. § 101.
Examiner’s Note – Representative Case Law
Alice Corp. v. CLS Bank Int’l, 573 U.S. 208 (2014) — Implementing abstract ideas on generic computers is not patent-eligible.
Mayo Collaborative Servs. v. Prometheus Labs., Inc., 566 U.S. 66 (2012) — Applying natural laws or abstract concepts using conventional steps is not patent-eligible.
Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350 (Fed. Cir. 2016) — Collecting, analyzing, and displaying information using conventional techniques is abstract.
SAP America, Inc. v. InvestPic, LLC, 898 F.3d 1161 (Fed. Cir. 2018) — Mathematical calculations and data analysis are abstract absent a specific technical improvement.
BSG Tech LLC v. BuySeasons, Inc., 899 F.3d 1281 (Fed. Cir. 2018) — Conventional limitations do not supply an inventive concept.
Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016) — Claims improving computer functionality may be eligible; here, no such improvement is recited.
6. Independent claims 13 and 14 are similar in limitations to that of Claim 1 and are of consequence rejected. Dependent claims 2-12 do not further limit claim 1 to achieve consistency with the requirements of 35 U.S.C. § 101 and of consequence rejected.
7. Guidance to Applicant — Overcoming the 101 Rejection
a. High-level strategy
Show the claim recites a concrete, technical solution that is more than an abstract idea by: a. adding concrete technical limitations that define how saliency records are computed, represented, and processed (i.e., specific algorithms, data structures, and dataflows); and/orb. adding hardware or architectural features (e.g., specialized accelerator, memory architecture, on-chip pipelines) or showing how claimed operations materially improve computer/system performance; and/orc. providing evidence (specification passages, experimental results, declarations) that the claimed combination yields a technical improvement (e.g., better downstream task accuracy, reduced latency, lower memory footprint, improved training stability).
Frame arguments to show the claimed elements are not well-understood, routine, or conventional in the art and that their combination produces a technical effect.
b. Types of claim amendments that strengthen eligibility (examples)
Tie saliency computation to a concrete method and structure:
Example limitation to add: “wherein creating a respective saliency record comprises computing, for each data record, a feature-attribution vector using integrated gradients computed with respect to the task network’s output layer, the saliency record comprising normalized attribution values per input feature.”
Specify the format and structure of saliency records and pooling operations:
Example: “the saliency records are two-dimensional matrices of attribution scores aligned to input tokens, and bringing together comprises concatenating the matrices into a fixed-size saliency buffer with a predefined axis order and normalization procedure.”
Specify the transfer cost function concretely:
Example: “the predefined transfer cost function comprises a Wasserstein distance computed between distributions of real and synthetic saliency records, wherein the discriminator’s loss is the empirical estimate of the Wasserstein distance between sampled saliency records.”
Specify concrete optimization routines and dataflow (how gradients propagate):
Example: “optimizing parameters comprises backpropagating the gradient of the transfer cost function through the discriminator, through the saliency computation module, and through the generator to update generator weights, wherein saliency computation is implemented as a differentiable module enabling joint gradient flow.”
Add hardware/architecture or performance constraints:
Example: “the method is performed on a computing apparatus comprising a neural-network accelerator with a dedicated saliency-computation pipeline and on-chip saliency buffer, the pipeline performing saliency computation in O(n) time per record and reducing host memory transfers by at least 40% compared to conventional off-device saliency computation.”
Add concrete timing/resource/throughput limits:
Example: “the generator is updated in real time with an update latency of less than X ms per batch due to pipelined saliency computation and gradient accumulation in the on-chip buffer.”
c. Argument points to present in rebuttal
Emphasize that the claim does not merely recite “train generator to fool discriminator” but recites a specific training signal (task-aware saliency records) and a particular optimization target (transfer cost over saliency space) that materially alters the training objective in a non-routine way.
Argue technical improvement: show how training on saliency records improves downstream task performance on target-domain real data, reduces labeling needs, stabilizes adversarial training, reduces compute/memory overhead (if supported), or otherwise improves how the computing system functions.
Distinguish conventional art: identify prior art that uses raw-data discriminators or feature embeddings and explain how saliency-based discriminators differ qualitatively (task-centric attribution vs. statistical similarity).
If available, cite specification passages describing concrete saliency computation, differentiability, data structures, implementation details, or experimental results.
If the specification lacks details, amend claims to include the concrete details listed above.
d. Evidence to submit
Point to specification paragraphs (identify exact FIGs, pseudocode, or embodiments) that disclose: saliency computation method, saliency record structure, transfer cost definition, joint backpropagation through saliency computation, hardware/acceleration, timing/latency results, and experimental benchmarks.
Provide empirical test data or benchmark comparisons showing improved downstream accuracy, reduced domain gap, reduced labeling, training stability, latency or memory improvements attributable to the claimed technique.
If possible, include a declarant declaration (expert or inventor) explaining why the claimed combination is unconventional and describing technical advantages.
If the claim is amended to recite hardware or accelerators, include descriptive detail or figures showing architecture and how it implements claimed steps.
Sample compact applicant amendment (one-sentence example)
Original: “creating a respective saliency record for each …”
Amended example: “creating a respective saliency record for each …, wherein creating the respective saliency record comprises computing a normalized, fixed-length feature-attribution vector for the record using integrated gradients with respect to the task network’s output, the attribution vector being stored in a fixed-size saliency buffer accessible to the discriminator and to the generator during joint optimization.”
e. If applicant prefers to argue instead of amend:
Provide detailed claim construction arguments explaining that the claim limitations, read in view of the specification, require a concrete, specific saliency computation that is implemented as a differentiable module enabling backpropagation and thus effecting a technological improvement; and
Submit supporting evidence (spec, declarations, benchmarks) showing nonconventionality and technical effects.
f. Draft language for prosecution reply
“Applicant respectfully traverses the Section 101 rejection. Claim 1 recites a specific, unconventional technical solution — namely [insert concise recitation of added specifics: saliency computation method, data structure, transfer cost, hardware optimization]. The claimed method transforms how training signals are computed and used by, inter alia, (i) computing task-aware saliency records using [method], (ii) using a discriminative loss computed in saliency space (the predefined transfer cost), and (iii) performing differentiable backpropagation through saliency computation to update the generator. These specific features materially improve training transferability and system performance by [cite results/figures]. Accordingly, Claim 1 is not directed to an abstract idea alone and recites significantly more under Alice Step 2B.”
Conclusion
8. The prior art made of record and not relied upon is considered pertinent to applicant's disclosure: See IDS
9. Any inquiry concerning this communication or earlier communications from the examiner should be directed to Joseph P. Hirl whose telephone number is (571)272-3685. The examiner can normally be reached Monday - Thursday 5:30 am to 3:30 pm. 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.
10. If attempts to reach the examiner by telephone are unsuccessful, the examiner's Director, Amy C. Johnson can be reached on 571-272-2238. 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/patentcenter 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.
/JOSEPH P HIRL/Supervisory Patent Examiner, Art Unit 2435