DETAILED ACTION
Notice of 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 .
Priority
Regarding Chinese Patent App. No. CN202410055105.0 (filed Jan. 12, 2024), receipt is acknowledged of certified copies of papers required by 37 CFR 1.55.
Information Disclosure Statement
The information disclosure statement submitted on 2/8/2024 has been considered.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Step 1 of the Alice/Mayo framework, Claims 1-9 are directed to a method (a process), Claims 10-18 are directed to an electronic device (a machine), Claim 19 is directed to a non-transitory computer-readable storage medium (an article of manufacture), and Claim 20 is directed to a computer program product being tangibly stored on a non-transitory computer-readable medium (an article of manufacture), which each fall within one of the four statutory categories of inventions.
Regarding Claim 1
Step 2A, prong 1 (Is the claim directed to a law of nature, a natural phenomenon or an abstract idea).
Claim 1 recites the following mental processes, that in each case under the broadest reasonable interpretation, covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components (e.g., “variational language model”).
determining a distribution of hidden variables in a variational language model of a question-answer system based on a query in a training data set; (under the broadest reasonable interpretation, a human such as a data scientist, can mentally review the latent space of a variational language model, such as a variational auto-encoder, and determine a distribution for such latent space, where such latent space has hidden variables that have been populated based on a query in the training data set going through the variational language model; the examiner notes that this claim limitation does not actually require the variational language model to be used)
generating a plurality of answers for the query ... based on a plurality of hidden variables randomly sampled from the distribution; (under the broadest reasonable interpretation, a human such as a data scientist, can randomly sample hidden variables from the distribution and use such variables to generate answers, such as by determining that a particular hidden variable pertains to negation, and negating potential answers correspondingly)
determining reward scores for the plurality of answers using a reward model; (under the broadest reasonable interpretation, a human such as a data scientist, could determine a reward score using a reward model, for example, a reward model that gives each answer a reward score based on the number of characters in the answer, where such reward model can be as simple as assigning 1 point to “a”, 2 points to “b”, ... 26 points to “z”)
Step 2A, prong 2 (Does the claim recite additional elements that integrate the judicial exception into a practical application?).
The judicial exception is not integrated into a practical application.
Regarding the “using the variational language model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. In particular, the claim only recites the additional element of a generic variational language model such as a VAE. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (a generic variational language model such as a VAE). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Regarding the “updating the variational language model based on the query and the best answer with the highest reward score among the plurality of answers” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. In particular, the claim only recites the additional element of updating or training/fine-tuning a model. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (updating or training/fine-tuning a model). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, after considering all claim elements individually and as an ordered combination, it is determined that the claims do not integrate the judicial exception into a practical application.
Step 2B (Does the claim recite additional elements that amount to significantly more than the judicial exception?)
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
Regarding the “using the variational language model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding the “updating the variational language model based on the query and the best answer with the highest reward score among the plurality of answers” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Accordingly, at Step 2B after considering all claim elements individually and as an ordered combination, it is determined that the claims do not integrate the judicial exception into a practical application.
Regarding Claim 2
Step 2A, Prong 1
wherein the hidden variables have a Gaussian distribution (under the broadest reasonable interpretation, a human such as a data scientist can mentally determine that the latent space of the variational language model has a Gaussian distribution)
determining distribution of hidden variables in the variational language model comprises: ... acquire a mean and a variance of the Gaussian distribution ... (under the broadest reasonable interpretation, a human such as a data scientist can mentally determine that the latent space of the variational language model has a Gaussian distribution and mentally (or using pencil and paper) calculate the mean and variance of such Gaussian distribution)
Step 2A, Prong 2
Regarding the “inputting the query to the variational language model to ... from the variational language model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. In particular, the claim only recites the additional element of a generic variational language model such as a VAE. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (a generic variational language model such as a VAE). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Step 2B
Regarding the “inputting the query to the variational language model to ... from the variational language model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding Claim 3
Step 2A, Prong 1
defining a prior distribution of the hidden variables relative to the query and a conditional distribution of the answers relative to the query and the hidden variables (under the broadest reasonable interpretation, a human such as a data scientist can mentally (or using pencil and paper), define such a prior distribution of the hidden variables and a conditional distribution as set forth in this limitation, such as by defining the prior distribution using the mathematical equation (2) of the instant specification at page 9, line 11 and by defining the conditional distribution using the mathematical equation (3) of the instant specification at page 9, line 21)
defining a joint distribution of the answers and the hidden variables relative to the query based on the prior distribution and the conditional distribution (under the broadest reasonable interpretation, a human such as a data scientist can mentally (or using pencil and paper), define such a joint distribution of the hidden variables as set forth in this limitation, such as by defining the joint distribution using the mathematical equation (1) of the instant specification at page 8, line 27)
determining a posterior probability of the hidden variables relative to the query and the answers (under the broadest reasonable interpretation, a human such as a data scientist can mentally (or using pencil and paper), determine a posterior probability of the hidden variables as set forth in this limitation, such as by using the mathematical equation (5) of the instant specification at page 10, line 15)
Step 2A, Prong 2
Regarding the “performing supervised pre-training on the variational language model, wherein the pre-training comprises:” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. In particular, the claim only recites the additional element of supervised training of a generic model. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (supervised training of a generic model). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Regarding the “training the variational language model by maximizing an evidence lower bound, the evidence lower bound being calculated based on the joint distribution, the prior distribution, the conditional distribution, and the posterior probability” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. In particular, the claim only recites the additional element of generic training of a generic model. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (generic training of a generic model). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Step 2B
Regarding the “performing supervised pre-training on the variational language model, wherein the pre-training comprises:” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding the “training the variational language model by maximizing an evidence lower bound, the evidence lower bound being calculated based on the joint distribution, the prior distribution, the conditional distribution, and the posterior probability” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding Claim 4
Step 2A, Prong 1
wherein the posterior probability has a Gaussian distribution (under the broadest reasonable interpretation, a human such as a data scientist can mentally determine that a posterior probability has a Gaussian distribution)
the determining a posterior probability of the hidden variables relative to the query and the answers comprises: ... determine a mean and a variance of the Gaussian distribution of the posterior probability, (under the broadest reasonable interpretation, a human such as a data scientist can mentally determine that a posterior probability has a Gaussian distribution and can mentally (or using pencil and paper) determine the mean and variance of such distribution)
Step 2A, Prong 2
Regarding the “inputting the query and the answers to a variational reasoning network to...” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. In particular, the claim only recites the additional element of a generic variational reasoning network such as a VAE. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (a generic variational reasoning network such as a VAE). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Regarding the “the variational reasoning network being contained in the variational language model or being a separate network” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. In particular, the claim only recites the additional element of a generic variational reasoning network such as a VAE. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (a generic variational reasoning network such as a VAE). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Step 2B
Regarding the “inputting the query and the answers to a variational reasoning network to...” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding the “the variational reasoning network being contained in the variational language model or being a separate network” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding Claim 5
Step 2A, Prong 1
acquiring a hidden variable by randomly sampling from the distribution (under the broadest reasonable interpretation, a human such as a data scientist can mentally view the distribution of the latent space of the variational language model and randomly select a variable)
generating current answer characters ... based on the hidden variable, answer characters already generated by the variational language model, and the query, so as to acquire one of the plurality of answers. (under the broadest reasonable interpretation, a human such as a data scientist can mentally view the data points set forth in this limitation and generate a potential answer to the input query)
Step 2A, Prong 2
Regarding the “using the variational language model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. In particular, the claim only recites the additional element of a generic variational language model such as a VAE. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (a generic variational language model such as a VAE). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Step 2B
Regarding the “using the variational language model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding Claim 6
Step 2A, Prong 1
generating the plurality of answers for the query using a beam search with a diversity penalty function. (under the broadest reasonable interpretation, a human such as a data scientist can generate a graph-like structure of potential answers using pencil and paper, and then perform a beam search using a diversity penalty function to narrow down the potential answers, such as by pruning the graph-like structure)
Regarding Step 2A, Prong 2, the claim does not include any additional elements that integrate the judicial exception into a practical application and regarding Step 2B, there are no additional elements recited that amount to significantly more than the judicial exception.
Regarding Claim 7
Step 2A, Prong 2
Regarding the “wherein the reward model is trained by reinforcement learning based on human feedback, and the human feedback measures a quality of the answers” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. In particular, the claim only recites the additional element of generic model training. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (generic model training). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Step 2B
Regarding the “wherein the reward model is trained by reinforcement learning based on human feedback, and the human feedback measures a quality of the answers” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding Claim 8
Step 2A, Prong 2
Regarding the “performing the following actions at least once: updating the variational language model based on a first optimization objective, the first optimization objective being defined as an expectation of a reward function of the reward model for the query and the best answer” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. In particular, the claim only recites the additional element of generic model training using an optimization objective. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (generic model training using an optimization objective). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Regarding the “updating the variational language model based on a second optimization objective, the second optimization objective being defined as an expectation of a difference between the reward function and a reference function” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. In particular, the claim only recites the additional element of generic model training using an optimization objective. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (generic model training using an optimization objective). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Step 2B
Regarding the “performing the following actions at least once: updating the variational language model based on a first optimization objective, the first optimization objective being defined as an expectation of a reward function of the reward model for the query and the best answer” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding the “updating the variational language model based on a second optimization objective, the second optimization objective being defined as an expectation of a difference between the reward function and a reference function” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding Claim 9
Step 2A, Prong 2
Regarding the “wherein the reference function is used for estimating an expected reward score for the query, and the reference function is achieved by sharing parameters with the reward model or a separate network model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. In particular, the claim only recites the additional element of generic model training using an optimization objective. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (generic model training using an optimization objective). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Step 2B
Regarding the “wherein the reference function is used for estimating an expected reward score for the query, and the reference function is achieved by sharing parameters with the reward model or a separate network model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding Claim 10
Step 2A, Prong 1
Claim 10 recites an electronic device that corresponds to the method of claim 1, and therefore the analysis under Step 2A, Prong 1 with respect to claim 1 also applies to this claim 10. While claim 10 recites additional generic computing components (“electronic device”, “processor”, “memory”, and “instructions”), such additional generic computing components do not change the analysis under Step 2A, Prong 1.
Step 2A, Prong 2
Claim 10 recites an electronic device that corresponds to the method of claim 1, and therefore the analysis under Step 2A, Prong 2 with respect to claim 1 also applies to this claim 10. While claim 10 recites additional generic computing components (“electronic device”, “processor”, “memory”, and “instructions”), such additional generic computing components do not change the analysis under Step 2A, Prong 2 because these additional elements are recited at a high-level of generality and amount to no more than mere instructions to apply the exception using a generic computer component. Accordingly, these additional elements do not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Step 2B
Claim 10 recites an electronic device that corresponds to the method of claim 1, and therefore the analysis under Step 2B with respect to claim 1 also applies to this claim 10. While claim 10 recites additional generic computing components (“electronic device”, “processor”, “memory”, and “instructions”), such additional generic computing components do not change the analysis under Step 2B because these additional elements are recited at a high-level of generality and amount to no more than mere instructions to apply the exception using a generic computer component. Accordingly, these additional elements do not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Claims 11-18 depend from claim 10, and recite electronic devices that correspond to the methods of claims 2-9, and are therefore each rejected for the same reasons explained above with respect to claim 10 and claims 2-9, respectively.
Claim 19 depends from claim 1, and recites a non-transitory computer-readable storage medium that performs the method of claim 1, and is therefore rejected for the same reasons explained above with respect to claim 1.
Regarding Claim 20
Step 2A, Prong 1
Claim 20 recites a computer program product being tangibly stored on a non-transitory computer-readable medium that corresponds to the method of claim 1, and therefore the analysis under Step 2A, Prong 1 with respect to claim 1 also applies to this claim 20. While claim 20 recites additional generic computing components (“computer program product”, “non-transitory computer-readable medium”, “machine”, and “instructions”), such additional generic computing components do not change the analysis under Step 2A, Prong 1.
Step 2A, Prong 2
Claim 20 recites a computer program product being tangibly stored on a non-transitory computer-readable medium that corresponds to the method of claim 1, and therefore the analysis under Step 2A, Prong 2 with respect to claim 1 also applies to this claim 20. While claim 20 recites additional generic computing components (“computer program product”, “non-transitory computer-readable medium”, “machine”, and “instructions”), such additional generic computing components do not change the analysis under Step 2A, Prong 2 because these additional elements are recited at a high-level of generality and amount to no more than mere instructions to apply the exception using a generic computer component. Accordingly, these additional elements do not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Step 2B
Claim 20 recites a computer program product being tangibly stored on a non-transitory computer-readable medium that corresponds to the method of claim 1, and therefore the analysis under Step 2B with respect to claim 1 also applies to this claim 20. While claim 20 recites additional generic computing components (“computer program product”, “non-transitory computer-readable medium”, “machine”, and “instructions”), such additional generic computing components do not change the analysis under Step 2B because these additional elements are recited at a high-level of generality and amount to no more than mere instructions to apply the exception using a generic computer component. Accordingly, these additional elements do not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
EXAMINER’S SUGGESTION TO PROGRESS THE CLAIMS TOWARDS SUBJECT MATTER ELIGIBILTY UNDER 35 U.S.C. 101: The examiner suggests amending the independent claims to specifically recite a “chatbot” or “virtual assistant” (see instant specification at p. 1, line 17), and providing an explanation as set forth in MPEP 2106.04(d)(1) (supported by citations to the instant specification), explaining why the addition of a variational learning model is an improvement in the technical field of “chatbots” or “virtual assistants” that would be recognized by one of ordinary skill in the art.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries 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, 5, 7, 10, 14, 16, and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over US 20200226475 A1, hereinafter referenced as MA, in view of US 11995803 B1, hereinafter referenced as KARPMAN, and further in view of US 20230367976 A1, hereinafter referenced as LI.
Regarding Claim 1
MA teaches:
A method comprising: (MA, para. 0007: “a deep learning-based NLG technique comprising a neural network-based method for dialog generation in a chatbot.”)
determining a distribution of hidden variables in a variational language model of a question-answer system based on a query in a training data set; (MA, para. 0007: “This allows the chatbot to handle different types of inputs, which are processed by a system including two components: a variational autoencoder (VAE) and a generative adversarial network (GAN). The VAE can encode and decode input in a prediction modeling which is trained by the GAN, an unsupervised generative model, by minimizing the distance between the input query and a predicted response. In this way, the VAE and the GAN may be used in combination to generate improved responses in a supervised learning manner. Furthermore, the combined VAE and GAN may be adapted to a Seq2Seq model dialog generator to improve syntactical or semantical completeness of responses.”;
MA, para. 0065: “The framework 401 of the VSDG 400 comprises an initial testing process 402 that includes training processes (indicated by arrows), followed by additional training steps. The testing process 402 begins with the entry of queries (seed questions) into the VAE encoder 404 after basic processing of the queries such as standardizing cases, word embedding lookup, sentence padding, etc. The most important feature of the VAE is a variational layer that is trained by KL-Divergence loss to unit-Gaussian vectors. The trained variational layer attempts to regularize the embeddings abstracted by the encoder to be closer to unit-Gaussian distribution. It is known that regularizing embedding vectors to unit-Gaussian distribution improves performance of the VAE in image classification tasks, due to a higher load of information, and thus may have a similar effect when applied to dialog generation.”;
MA, para. 0075: “The language model used is trained on a combination of a twitter text8 dataset and a portion of the Maluuba training set.”
Examiner’s Note: As shown in Fig. 4, element 400 corresponds to the recited “question-answer system” and includes a VAE and GAN that collectively correspond to the recited “variational language model”, where the vector space of embedding vectors in a unit-Gaussian distribution corresponds to recited “distribution of hidden variables”)
generating a plurality of answers for the query using the variational language model based on a plurality of hidden variables (MA, para. 0018: “The VAE may convert queries received by the chatbot to vectors that are then used by the neural network to generate a response. The responses generated by the neural network are sent to the GAN where a competitive process evaluates a difference between the generated answer and a seed answer. The difference is used to train the neural network.”;
MA, para. 0070: “The model takes a training set, consisting of samples drawn from a distribution and learns to represent an estimate of the distribution. The result is a probability distribution which may be estimated explicitly or the model may generate samples from the probability distribution.”;
MA, para. 0072: “The generated answers provided by the GAN-G 408 is the final component of the testing process 402. The framework 401 of the VSDG 400 continues training steps including seeding a GAN discriminator (GAN-D) 410 with answers that are representative of expected responses from humans.”;
Examiner’s Note: the VAE+GAN output generated answers (plural) corresponding to the seed question, and such output is based on samples drawn from a distribution)
However, MA fails to explicitly teach:
hidden variables randomly sampled from the distribution
determining reward scores for the plurality of answers using a reward model; and
updating the variational language model based on the query and the best answer with the highest reward score among the plurality of answers.
However, in a related field of endeavor (generative models, see col. 1, lines 24-26), KARPMAN teaches and makes obvious:
determining reward scores for the plurality of answers using a reward model; and (KARPMAN, col. 9, lines 14-35: “Once assembled, the system 100 can then execute the (untrained) reward model 114 on the preference training set, enabling the reward model 114 to automatically derive correlations between visual features of images generated by the text-to-image diffusion model 112 and preference scores produced by human annotators. ... Once trained, the reward model 114 can then analyze an image (e.g., an image generated by the text-to-image diffusion model 112) and compute a reward score that aligns with (e.g., is similar to, highly correlated with) human aesthetic preferences for generated images. As described in more detail below, the system can therefore leverage the reward model 114 (e.g., preference scores generated by the reward model 114) to update, optimize, and/or fine-tune parameters of the text-to-image diffusion model 112 during the training stage(s) to incorporate (simulated) human feedback into the text-to-image diffusion model 112's training objectives.”;
Examiner’s Note: the MA-KARPMAN combination now modifies the VAE+GAN of MA to use a modified rewards model 114 of KARPMAN to determine reward scores for each of the plurality of generated answers of MA)
updating the variational language model based on the query and the best answer with the highest reward score among the plurality of answers. (KARPMAN, col. 9, lines 14-35: “Once assembled, the system 100 can then execute the (untrained) reward model 114 on the preference training set, enabling the reward model 114 to automatically derive correlations between visual features of images generated by the text-to-image diffusion model 112 and preference scores produced by human annotators. ... Once trained, the reward model 114 can then analyze an image (e.g., an image generated by the text-to-image diffusion model 112) and compute a reward score that aligns with (e.g., is similar to, highly correlated with) human aesthetic preferences for generated images. As described in more detail below, the system can therefore leverage the reward model 114 (e.g., preference scores generated by the reward model 114) to update, optimize, and/or fine-tune parameters of the text-to-image diffusion model 112 during the training stage(s) to incorporate (simulated) human feedback into the text-to-image diffusion model 112's training objectives.”;
KARPMAN, col. 15, lines 23-30: “The system can therefore leverage the reward model 114 when fine-tuning some or all parameters of the base image diffusion model 120 (and the high-resolution diffusion models 116) in order to bias the model towards generating images that are more likely to align with human aesthetic preferences and/or human expectations of alignment between the text prompt and the generated image.”;
Examiner’s Note: the MA-KARPMAN combination now modifies the VAE+GAN of MA to use a modified rewards model 114 of KARPMAN to determine reward scores for each of the plurality of generated answers of MA, and then using such scores (including the highest score) to update the VAE+GAN of MA)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill to combine the teachings of MA with KARPMAN as explained above. As disclosed by KARPMAN, one of ordinary skill would have been motivated to do so in order to improve “human expectations” with respect to the generated answers. (col. 15, lines 23-30). As disclosed by KARPMAN, one of ordinary skill would have been motivated to do so in order to simulate human feedback, (col. 3, lines 32-36) “without relying on prohibitively costly human sourcing and annotation.” (col. 9, lines 62-63).
However, MA and KARPMAN fail to explicitly teach:
hidden variables randomly sampled from the distribution
However, in a related field of endeavor (natural language processing using variational autoencoders, see paras. 0020 and 0022), LI teaches and makes obvious:
hidden variables randomly sampled from the distribution (LI, para. 0056: “The latent variable can be a numerical value or a vector. Random sampling can be performed on the first semantic distribution to obtain the latent variable.”;
Examiner’s Note: the MA-KARPMAN-LI combination now modifies the VAE+GAN of MA so that the embedding space of MA, which is a latent vector space, is now randomly sampled as in LI)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill to combine the teachings of MA with KARPMAN and LI as explained above. One of ordinary skill would have been motivated to do so in order to randomly select from the latent space, and therefore minimize selection bias and prevent human unconscious biases from influencing the results.
Regarding Claim 5
MA, KARPMAN, and LI teach the method of claim 1. MA further teaches:
generating current answer characters using the variational language model based on the hidden variable, answer characters already generated by the variational language model, and the query, so as to acquire one of the plurality of answers. (MA, para. 0046: “The example of FIG. 2 shows a reply that provides an answer directly correlated to the entry question. Alternatively, if the response does not satisfy the user's question, as determined by the chatbot by continued queries from the user, the chatbot may continue attempting to match the query to data from a VAE model of the VSDG and send further responses to the user interface 200. When a response is found acceptable by the user, as detected by a user entry such as “Ok, thank you”, or “Yes, that's what I need”, a GAN of the VSDG may then train the VAE to use the response determined to be appropriate for future responses to the same or similar queries.”;
MA, para. 0065: “The framework 401 of the VSDG 400 comprises an initial testing process 402 that includes training processes (indicated by arrows), followed by additional training steps. The testing process 402 begins with the entry of queries (seed questions) into the VAE encoder 404 after basic processing of the queries such as standardizing cases, word embedding lookup, sentence padding, etc. The most important feature of the VAE is a variational layer that is trained by KL-Divergence loss to unit-Gaussian vectors. The trained variational layer attempts to regularize the embeddings abstracted by the encoder to be closer to unit-Gaussian distribution. It is known that regularizing embedding vectors to unit-Gaussian distribution improves performance of the VAE in image classification tasks, due to a higher load of information, and thus may have a similar effect when applied to dialog generation.”;
Examiner’s Note: MA teaches that the vector embeddings (corresponding to recited “hidden variable”), previous responses that are deemed to be good (corresponding to recited “answer characters already generated by the variational language model”) and the input question itself are all utilized by the VAE+GAN when generating answers)
However, MA and KARPMAN fail to explicitly teach:
acquiring a hidden variable by randomly sampling from the distribution;
However, in a related field of endeavor (natural language processing using variational autoencoders, see paras. 0020 and 0022), LI teaches and makes obvious:
acquiring a hidden variable by randomly sampling from the distribution (LI, para. 0056: “The latent variable can be a numerical value or a vector. Random sampling can be performed on the first semantic distribution to obtain the latent variable.”;
Examiner’s Note: the MA-KARPMAN-LI combination now modifies the VAE+GAN of MA so that the embedding space of MA, which is a latent vector space, is now randomly sampled as in LI)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill to combine the teachings of MA with KARPMAN and LI as explained above. One of ordinary skill would have been motivated to do so in order to randomly select from the latent space, and therefore minimize selection bias and prevent human unconscious biases from influencing the results.
Regarding Claim 7
MA, KARPMAN, and LI teach the method of claim 1. However, MA fails to explicitly teach:
wherein the reward model is trained by reinforcement learning based on human feedback, and the human feedback measures a quality of the answers.
However, in a related field of endeavor (generative models, see col. 1, lines 24-26), KARPMAN teaches and makes obvious:
wherein the reward model is trained by reinforcement learning based on human feedback, and the human feedback measures a quality of the answers. (KARPMAN, col. 3, lines 28-36: “Additionally, the system 102 can fine-tune the text-to-image diffusion model 112 using outputs of a human visual preference model (e.g., a reward model 114) trained on human input judgments of aesthetic quality and/or text-image alignment, thereby incorporating (simulated) human feedback on images generated by the text-to-image diffusion model 112 to further improve the model's performance on image generation tasks during operation.”;
KARPMAN, col. 15, lines 31-39: “During the fine-tuning phase, the system 100 can initialize a copy of the text-to-image diffusion model 112 with the initial set of image generation parameters (e.g., a “policy”) that describe and/or control means and variances of a noise level at each pixel at each diffusion step. In some implementations, for each text prompt in a fine-tuning set, the system 100 can execute a reinforcement learning algorithm to: input the text prompt to the (pre-trained) text-to-image diffusion model 112 to generate a first image”
Examiner’s Note: the MA-KARPMAN-LI combination now modifies the VAE+GAN of MA to use a modified rewards model 114 of KARPMAN, where such rewards model 114 uses human annotation with respect to quality)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill to combine the teachings of MA with KARPMAN and LI as explained above. As disclosed by KARPMAN, one of ordinary skill would have been motivated to do so in order to improve “human expectations” with respect to the generated answers. (col. 15, lines 23-30). As disclosed by KARPMAN, one of ordinary skill would have been motivated to do so in order to simulate human feedback, (col. 3, lines 32-36) “without relying on prohibitively costly human sourcing and annotation.” (col. 9, lines 62-63).
Regarding Claim 10
MA teaches:
An electronic device, comprising: at least one processor; and a memory coupled to the at least one processor and having instructions stored therein, wherein the instructions, when executed by the at least one processor, cause the electronic device to perform actions comprising: (MA, para. 0090: “a system includes a user interface device, and a processor communicatively coupled to the user interface device, the processor configured with a variational autoencoder (VAE) combined with a generative adversarial network (GAN) stored in non-transitory memory, the processor further configured with instructions stored in the non-transitory memory that, when executed, cause the processor to”)
The remaining limitations correspond to the method of claim 1, and therefore this claim 10 is rejected for the same reasons explained above with respect to claim 1.
Claim 14 depends from claim 10 and claims an electronic device that corresponds to the method of claim 5, and is therefore rejected for the same reasons explained above with respect to claims 5 and 10.
Claim 16 depends from claim 10 and claims an electronic device that corresponds to the method of claim 7, and is therefore rejected for the same reasons explained above with respect to claims 7 and 10.
Regarding Claim 19
MA teaches: A non-transitory computer-readable storage medium having machine-executable instructions stored thereon, wherein the machine-executable instructions, when executed by a machine, cause the machine to perform the method of claim 1. (MA, para. 0090: “a system includes a user interface device, and a processor communicatively coupled to the user interface device, the processor configured with a variational autoencoder (VAE) combined with a generative adversarial network (GAN) stored in non-transitory memory, the processor further configured with instructions stored in the non-transitory memory that, when executed, cause the processor to”)
The remaining limitations correspond to the method of claim 1, and therefore this claim 19 is rejected for the same reasons explained above with respect to claim 1.
Regarding Claim 20
MA teaches:
A computer program product, the computer program product being tangibly stored on a non-transitory computer-readable medium and comprising machine-executable instructions, wherein the machine-executable instructions, when executed by a machine, cause the machine to perform actions comprising: (MA, para. 0090: “a system includes a user interface device, and a processor communicatively coupled to the user interface device, the processor configured with a variational autoencoder (VAE) combined with a generative adversarial network (GAN) stored in non-transitory memory, the processor further configured with instructions stored in the non-transitory memory that, when executed, cause the processor to”)
The remaining limitations correspond to the method of claim 1, and therefore this claim 20 is rejected for the same reasons explained above with respect to claim 1.
Claims 2 and 11 are rejected under 35 U.S.C. 103 as being unpatentable over MA in view of KARPMAN and LI and further in view of Japa, Sai Sharath, et al. "Question Answering over Knowledge Base with Variational Auto-Encoder." 2022 IEEE Eighth International Conference on Multimedia Big Data (BigMM). IEEE, 2022, hereinafter referenced as JAPA.
Regarding Claim 2
MA, KARPMAN, and LI teach the method of claim 1. However, MA further teaches:
wherein the hidden variables have a Gaussian distribution, (MA, para. 0065: “The framework 401 of the VSDG 400 comprises an initial testing process 402 that includes training processes (indicated by arrows), followed by additional training steps. The testing process 402 begins with the entry of queries (seed questions) into the VAE encoder 404 after basic processing of the queries such as standardizing cases, word embedding lookup, sentence padding, etc. The most important feature of the VAE is a variational layer that is trained by KL-Divergence loss to unit-Gaussian vectors. The trained variational layer attempts to regularize the embeddings abstracted by the encoder to be closer to unit-Gaussian distribution. It is known that regularizing embedding vectors to unit-Gaussian distribution improves performance of the VAE in image classification tasks, due to a higher load of information, and thus may have a similar effect when applied to dialog generation.”)
However, MA, KARPMAN, and LI fail to explicitly teach:
the determining distribution of hidden variables in the variational language model comprises: inputting the query to the variational language model to acquire a mean and a variance of the Gaussian distribution from the variational language model.
However, in a related field of endeavor (question answering using a variational auto-encoder), JAPA teaches and makes obvious:
the determining distribution of hidden variables in the variational language model comprises: inputting the query to the variational language model to acquire a mean and a variance of the Gaussian distribution from the variational language model. (JAPA, p. 32, section IV.F: “In Gaussian embedding, we learn both mean and variance for the representations. Projecting the representations as densities over a latent space instead of lowdimensional vectors. These densities provide uncertainty among the objects. Our method adopts KL divergence between Gaussian distributions to measure the relationship between objects, which is straightforward to calculate”;
Examiner’s Note: the MA-KARPMAN-LI-JAPA combination now modifies the VAE+GAN of MA to learn the mean and variance for the Gaussian distribution (as in JAPA) of the latent space of MA)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill to combine the teachings of MA with KARPMAN, LI, and JAPA as explained above. As disclosed by JAPA, one of ordinary skill would have been motivated to do so in order to “better captur[e] uncertainty about a representation and its relationships, providing asymmetric comparisons between objects, which is more effective than dot product or cosine similarity, and which enables more expressive parameterization of decision boundaries.” (JAPA, p. 32, section IV.F)
Claim 11 depends from claim 10 and claims an electronic device that corresponds to the method of claim 2, and is therefore rejected for the same reasons explained above with respect to claims 2 and 10.
Claims 6 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over MA in view of KARPMAN and LI and further in view of US 20240274123 A1, hereinafter referenced as ZHANG.
Regarding Claim 6
MA, KARPMAN, and LI teach the method of claim 1. However, MA, KARPMAN, and LI fail to explicitly teach:
generating the plurality of answers for the query using a beam search with a diversity penalty function.
However, in a related field of endeavor (natural language processing, see paras. 0013, 0016), ZHANG teaches and makes obvious:
generating the plurality of answers for the query using a beam search with a diversity penalty function. (ZHANG, para. 0076: “During the decoding step ..., for each candidate monophone p, diversity penalty (A) may be added to the original beam search loss:”;
Examiner’s Note: the MA-KARPMAN-LI-ZHANG combination now modifies MA to generate answers for a query using a beam search with a diversity penalty as taught by ZHANG)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill to combine the teachings of MA with KARPMAN, LI, and ZHANG as explained above. One of ordinary skill would have been motivated to apply the beam search of ZHANG in order to efficiently narrow options in a graph-like structure
Claim 15 depends from claim 10 and claims an electronic device that corresponds to the method of claim 6, and is therefore rejected for the same reasons explained above with respect to claims 6 and 10.
Allowable Subject Matter
Claims 3-4, 8-9, 12-13, and 17-18 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims, and provided that the rejections under 35 U.S.C. 101 are overcome.
The following is a statement of reasons for the indication of allowable subject matter:
Claims 3 and 12 would be considered allowable, if rewritten in independent form including all of the limitations of the base claim and any intervening claims, and provided that the rejections under 35 U.S.C. 101 are overcome, because none of the references of record either alone or in combination fairly disclose or suggest the combination of limitations specified in claims 3 and 12, including at least:
defining a prior distribution of the hidden variables relative to the query and a conditional distribution of the answers relative to the query and the hidden variables;
defining a joint distribution of the answers and the hidden variables relative to the query based on the prior distribution and the conditional distribution;
determining a posterior probability of the hidden variables relative to the query and the answers; and
training the variational language model by maximizing an evidence lower bound, the evidence lower bound being calculated based on the joint distribution, the prior distribution, the conditional distribution, and the posterior probability.
The closest prior art of record discloses:
20200226475 A1, hereinafter referenced as MA, teaches a variational autoencoder used to generate responses to queries. (paras. 0007, 0018, 0065).
US 11995803 B1, hereinafter referenced as KARPMAN, teaches determining reward scores for generative AI models using a model to simulate human feedback. (col. 9, lines 14-35, col. 15, lines 23-30)
US 20230367976 A1, hereinafter referenced as LI, teaches random sampling from a distribution. (para. 0056).
US 20210057108 A1, hereinafter referenced as FISHER, teaches, with respect to generative AI models, the concepts of defining a “prior distribution” (para. 0012, 0041), a “conditional distribution” (para. 0042), a “joint distribution” (para. 0042), and a “posterior possibility” (para. 0054).
Lee, Dong Bok, et al. "Generating diverse and consistent QA pairs from contexts with information-maximizing hierarchical conditional VAEs." Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. 2020, hereinafter referenced as LEE, teaches, with respect to a variational auto-encoder for generating question-answering pairs, maximizing the Evidence Lower Bound (ELBO) based on a variational posterior. (p. 211, section 3.1).
However, the examiner has found that the distinct feature of the Applicant's claimed invention over the prior art is the explicit claiming of the aforementioned limitations in combination with all the other limitations as specified in claims 3 and 12. In particular, one of ordinary skill would not have been motivated to maximize an evidence lower bound, where such evidence lower bound is calculated based on the joint distribution, the prior distribution, the conditional distribution, and the posterior probability as recited in claims 3 and 12, without the hindsight aid of Applicant’s disclosure. Therefore, because the prior art of record does not anticipate nor make obvious the limitations recited in claims 3 and 12, claims 3 and 12 would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims, and provided that the rejections under 35 U.S.C. 101 are overcome.
Claim 4 depends from claim 3 and would be allowable for the same reasons explained with respect to claim 3 if rewritten in independent form including all of the limitations of the base claim and any intervening claims, and provided that the rejections under 35 U.S.C. 101 are overcome.
Claim 13 depends from claim 12 and would be allowable for the same reasons explained with respect to claim 12 if rewritten in independent form including all of the limitations of the base claim and any intervening claims, and provided that the rejections under 35 U.S.C. 101 are overcome.
Claims 8 and 17 would be considered allowable, if rewritten in independent form including all of the limitations of the base claim and any intervening claims, and provided that the rejections under 35 U.S.C. 101 are overcome, because none of the references of record either alone or in combination fairly disclose or suggest the combination of limitations specified in claims 8 and 17, including at least:
updating the variational language model based on a first optimization objective, the first optimization objective being defined as an expectation of a reward function of the reward model for the query and the best answer; and
updating the variational language model based on a second optimization objective, the second optimization objective being defined as an expectation of a difference between the reward function and a reference function.
The closest prior art of record discloses:
20200226475 A1, hereinafter referenced as MA, teaches a variational autoencoder used to generate responses to queries. (paras. 0007, 0018, 0065).
US 11995803 B1, hereinafter referenced as KARPMAN, teaches determining reward scores for generative AI models using a model to simulate human feedback. (col. 9, lines 14-35, col. 15, lines 23-30)
US 20230367976 A1, hereinafter referenced as LI, teaches random sampling from a distribution. (para. 0056).
US 20240281891 A1, hereinafter referenced as FIELDS, teaches with respect to reinforcement learning, determining an award based on an expected response to a prompt. (para. 0281).
US 20230214453 A1, hereinafter referenced as SANTHAR, teaches, with respect to reinforcement learning for generative adversarial networks, determining a difference between actual and expected reward scores. (para. 0061).
However, the examiner has found that the distinct feature of the Applicant's claimed invention over the prior art is the explicit claiming of the aforementioned limitations in combination with all the other limitations as specified in claims 8 and 17. In particular, one of ordinary skill would not have been motivated to utilize the two recited optimization objectives with respect to reinforcement learning without the hindsight aid of Applicant’s disclosure. Therefore, because the prior art of record does not anticipate nor make obvious the limitations recited in claims 8 and 17, claims 8 and 17 would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims, and provided that the rejections under 35 U.S.C. 101 are overcome.
Claim 9 depends from claim 8 and would be allowable for the same reasons explained with respect to claim 8 if rewritten in independent form including all of the limitations of the base claim and any intervening claims, and provided that the rejections under 35 U.S.C. 101 are overcome.
Claim 18 depends from claim 17 and would be allowable for the same reasons explained with respect to claim 17 if rewritten in independent form including all of the limitations of the base claim and any intervening claims, and provided that the rejections under 35 U.S.C. 101 are overcome.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
US 20210201192 A1 (Kim). As depicted in Fig. 1B, discloses an auto-encoder model that generates questions and answers from passages. (paras. 0004, 0070).
US 20190087728 A1 (Agarwal). “The VAE is modeled using RNNs comprising of LSTM units. The LSTM-VAE can be used to automatically generate linguistically novel questions, which, (a) corrects classifier bias when augmented to the training data, (b) uncovers incompleteness in the set of answers and (c) improves the accuracy and generalization abilities of the base LSTM classifier, enabling it to learn from smaller training data. The novel questions sometimes belonged to completely new classes not present in the original training data.” (para. 0019).
US 20190065948 A1 (Henry). “In some implementations, only the selection decisions having the highest reward scores may be used to improve the selection model. Using only selection decisions with the highest reward scores may provide the selection model with positive feedback to reinforce good decisions made by the selection model. Where Q selection decisions with the highest reward scores are used to update the selection model” (para. 0057).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MICHAEL C LEE whose telephone number is (571)272-4933. The examiner can normally be reached M-F 12:00 pm - 8:00 pm ET.
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, Omar Fernandez Rivas can be reached at 571-272-2589. 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.
/MICHAEL C. LEE/Examiner, Art Unit 2128
Prosecution Insights