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 .
Claim Objections
Claims 1, 13 and 20 are objected to because of the following informalities: applicant claims, “one or more selected candidate architecture”. Architecture should be plural. Appropriate correction is required.
Claim Rejections - 35 USC § 101
Claim 20 is rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claim(s) does/do not fall within at least one of the four categories of patent eligible subject matter because applicant claims computer readable memory with instructions, and Spec. 126 makes it clear that includes signals. Signals per se are not a statutory class of patent eligible subject matter. Amend to claim non-transitory computer storage medium.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
Claims 5, 10, 11 and 17 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claims 5 and 17 recite the limitation "the training". There is insufficient antecedent basis for this limitation in the claim.
Claim 10 recites the limitation "the global historical population data". There is insufficient antecedent basis for this limitation in the claim.
Claim 11 recites the limitation "the historical population of candidate architectures having the best fitness measure". There is insufficient antecedent basis for this limitation in the claim.
Allowable Subject Matter
Claims 1-4, 6-9, 12-16, 18 and 19 are allowed.
There is no prior art rejection for claims 5, 10, 11, 17 and 20.
The following is a statement of reasons for the indication of allowable subject matter: Novelty Driven Evolutionary Neural Architecture Search by Sinha et al teaches element from claims 1, 13 and 20. A computer-implemented method comprising:
receiving training data for training a neural network to perform a machine learning task, the training data comprising a plurality of training examples and a respective target output for each of the training examples; (Sinha sec. 3.4 “In each generation, the supernet is trained on the training data…. the fitness of each individual architecture in the population, Facc, is calculated using the supernet.” The fitness is a measure of the how close the inference is to the target output of the training data. Sinha sec. 4.4.2 shows that NNs are trained, “method provides a set of neural architecture solutions in a single run.”)
searching for a final architecture of the neural network for performing the machine learning task using a (Sinha abs “the NAS problem is posed as a multi objective problem with 2 objectives: (i) maximize architecture novelty, (ii) maximize architecture fitness/accuracy. The novelty search is used for maintaining a diverse set of solutions at each generation which helps avoiding local optimum traps while the architecture fitness is calculated using supernet.”)
maintaining current population data that includes, for each candidate architecture in a current population of candidate architectures, (Sinha algorithm 1 p. 4 “P ← Initialize population for NSGA-II algorithm;” The architecture fitness measure for each architecture candidate is calle “Facc”. The source code definining the candidate architecture is the
repeatedly performing evolutionary architecture search operations comprising: (Sinha algorithm 1 loops through G, going from the first generation 1 to the Gth generation G. The part of this loop that makes it evolutionary is the “Apply NSGA-II to get next generation population” step. Algorithm 1 is a search because it is a “NEvoNAS” which is a novelty driven evolutionary neural architecture search. Sinha Title.)
selecting one or more candidate architectures from the current population of candidate architectures defined by the (Sinha’s algorithm 1 selects candidate models for evaluation in the evaluation steps in lines 8-10.)
using the plurality of new candidate architectures defined by the output (Sinha algorithm 1 update archive step in line 11.)
Sinha doesn’t teach storing the source code of the neural network architecture candidate or the LLM for evolving the architecture.
Evolution through Large Models by Lehman et al teaches language model neural network, wherein the searching comprises: (Lehman abs “ELM combined with MAP Elites generates hundreds of thousands of functional examples of Python programs that output working ambulating robots in the Sodarace domain, which the original LLM had never seen in pre-training. These examples then help to bootstrap training a new conditional language model that can output the right walker for a particular terrain.” Lehman p. 3 “(2) a technique for improving ELM’s ability to search over time by fine-tuning its LLM-based mutation operator…”)
maintaining current population data that includes, for each candidate architecture in a current population of candidate architectures, (i) source code defining the candidate architecture, (Lehman teaches this MAP of code defining candidate architecture in fig. 3, see below.)
PNG
media_image1.png
390
572
media_image1.png
Greyscale
generating an input prompt that comprises (i) the source code defining the one or more selected candidate architectures and (ii) a set of performance metrics of the one or more selected candidate architecture; (Lehman fig. 3 p. 11 “In each iteration, an existing Python solution is sampled from the map of solutions for each independent replica of a diff model. Each replica generates a batch of diffs that are applied to the sampled solution to generate modified candidate solutions.” Lehman p. 10 teaches the performance metric for each solution as a behaviour characterization outperforming another niche’s current inhabitant, “a single pre-existing (hand-designed in this paper) solution is evaluated and placed into the map. In each iteration thereafter, an inhabited niche is randomly chosen and the solution within that niche is perturbed by the diff model and evaluated. The new candidate solution is assigned its niche from its behavior characterization, and if that niche is unfilled or the new solution outperforms the niche’s current inhabitant, it becomes the champion of that niche…” Lehman p. 2 shows that the mutation is from a generated prompt, “simply by prompting the LLM to generate changes the LLM can serve as a highly sophisticated mutation operator embedded within an overarching evolutionary algorithm.”)
processing the input prompt using the language model neural network in accordance with current parameter values of the language model neural network to generate output source code that defines a plurality of new candidate architectures; and (Lehman p. 2 “the LLM in concert with evolution can steer each other towards the right region of the solution space even though neither evolution with a conventional mutation operator nor the LLM on its own could generate anything close.” Lehman fig. 3, below.)
PNG
media_image1.png
390
572
media_image1.png
Greyscale
using the plurality of new candidate architectures defined by the output source code to update the current population data. (Lehman p. 10 “The new candidate solution is assigned its niche from its behavior characterization, and if that niche is unfilled or the new solution outperforms the niche’s current inhabitant, it becomes the champion of that niche…”)
Lehman, the claims and Sinha are all evolutionary search algorithms. However, it’s not clear that a person would be motivated to switch the model parameters of Sinha with source code taught by Lehman while keeping the source code directed to Sinha’s neural networks. Therefore, claims 1-20 are not taught or made obvious by the prior art of record.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Austin Hicks whose telephone number is (571)270-3377. The examiner can normally be reached Monday - Thursday 8-4 PST.
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, Mariela Reyes can be reached at (571) 270-1006. 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.
/AUSTIN HICKS/Primary Examiner, Art Unit 2142
Prosecution Insights