DETAILED ACTION
This Office Action is in response to the correspondence filed by the applicant on 4/4/2024.
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 .
Priority
Receipt is acknowledged of certified copies of papers submitted under 35 U.S.C. 119(a)-(d), which papers have been placed of record in the file.
Information Disclosure Statement
The Information Statements (IDS) filed on 4/4/2024 have been accepted and considered in this office action and are in compliance with the provisions of 37 CFR 1.97.
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.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 2 is 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 pre-AIA the applicant regards as the invention.
Claim 2 recites the limitation, “the code included in the prediction result data into the string representing the at least one of the plurality of words.” There is insufficient antecedent basis for the bolded limitation in the claim.
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 1-5 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The independent claims 1 4 and 5 recite “acquiring, by a processor, word dictionary data in which strings each representing one of a plurality of words used in a natural language are mapped to codes each identifying one of the plurality of words; converting, by the processor, text data written in the natural language into encoded text data by encoding words included in the text data based on the word dictionary data; initializing, by the processor, parameters included in a machine learning model based on the word dictionary data and the encoded text data; and running, by the processor, a learning process to train, based on the encoded text data, the machine learning model from which the word dictionary data has been detached after the initializing of the parameters.”
The limitations, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is, other than reciting “by a processor” and “a memory”, nothing in the claim element precludes the step from practically being performed in the mind. For example, a person can gather textual data, covert the data into numerical values, initialize a model parameter, and perform a model training by using the numerical values. The limitations, as drafted, are processes that, under its broadest reasonable interpretation, cover performance of the limitations in the mind.
If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. In particular, the claims only recite additional elements – “by a processor” and “a memory”. The additional elements in both steps is recited at a high-level of generality (i.e., as a generic processor performing a generic computer function of the recited steps) such that it amounts no more than mere instructions to apply the exception using a generic computer component. 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. The claim is directed to an abstract idea.
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using a processor and a memory to perform the recited steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding the dependent claims, claim 2 recites acquiring prediction results and converting the results into the string representations; and claim 3 recites setting distributed representation data for converting the codes into the distributed representation vectors.
Even though the disclosed invention is described in the specification as improving computer technology, the claim provides no meaningful limitations such that this improvement is realized. Therefore, the claim does not amount to significantly more than the abstract idea itself.
Accordingly, the limitations of the Claims, whether considered individually or as an ordered combination, are not sufficient to add significantly more to improve technological functionality. As such, claims 1-5 are rejected under 35 U.S.C. 101 as being directed to non-statutory subject matter.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 1 and 3-5 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by ELKIND (US 2019/0273509 A1).
REGARDING CLAIM 1, ELKIND discloses a machine learning method comprising:
acquiring, by a processor, word dictionary data in which strings each representing one of a plurality of words used in a natural language are mapped to codes each identifying one of the plurality of words (Par 76 – “An example RNN layer is shown in FIG. 4A. The RNN shown in FIG. 4A analyzes a value at time (sample or token) t. In the case of source data being vectorized, by, e.g., a Shannon Entropy calculation, the vector samples can be directly processed by the system. In the case of tokenized inputs (e.g., natural language or command line input), the tokens are first vectorized by a mapping function. One such mapping function is a lookup table mapping a token to a vectorized input.”; Par 138 – “In FIG. 10, the tokenized inputs 1005 may be represented to computational layers as real-valued vectors or initial “embeddings” representing the input. These initial embeddings can be created using word2vec, one-hot encoding, feature hashing, or latent semantic analysis, or other techniques.”);
converting, by the processor, text data written in the natural language into encoded text data by encoding words included in the text data based on the word dictionary data (Par 76 – “An example RNN layer is shown in FIG. 4A. The RNN shown in FIG. 4A analyzes a value at time (sample or token) t. In the case of source data being vectorized, by, e.g., a Shannon Entropy calculation, the vector samples can be directly processed by the system. In the case of tokenized inputs (e.g., natural language or command line input), the tokens are first vectorized by a mapping function. One such mapping function is a lookup table mapping a token to a vectorized input.”);
initializing, by the processor, parameters included in a machine learning model based on the word dictionary data and the encoded text data (Par 113 – “In one example, the initial input may be input that is applied to a feature extractor such as a Shannon Entropy window to generate the input at block 920. The weights of the network layers are adjusted based on the differences between the output of the fully connected layer and the input to the system. The comparison is then used in conjunction with a parameter adjusting algorithm such as a gradient descent algorithm to minimize the error in the reconstructed inputs calculated at block 980. The gradient descent algorithm adjusts the weights (parameters) of the convolutional filter (if present), the encoding RNN layers, the fully connected layer (if present), and the decoding RNN layers to minimize or reduce the error between input at block 920 and the output of decoding RNN layer at block 970 (e.g., the reconstructed inputs calculated at block 980).”); and
running, by the processor, a learning process to train, based on the encoded text data, the machine learning model from which the word dictionary data has been detached after the initializing of the parameters (Par 113 – “The samples are repetitively applied to the network, and the parameters are adjusted until the network achieves an acceptable level of input reconstruction. Once trained, fully connected layer at block 960, decoding RNN layers at block 970 and the input reconstruction layer at block 980 can be removed from the system. In some examples, fully connected layer at block 960 is not removed after training and instead its output is used as the output of the system after training. Additionally, in some examples some layers from fully connected layer at block 960 are removed and the remaining layers are used as output. For example, the final layer can be removed and the penultimate layer of fully connected layer at block 960 is used as output.”; Par 79 – “In some examples, samples are vectors derived from strings which are mapped to vectors of numerical values. In one example, this mapping from string tokens to vectors of numerical values can be learned during training of the RNN. In an example, the mapping from string tokens to vectors of numerical values may not be learned during training. In an example, the mappings from string tokens to vectors of numerical values need not be learned during training. In some examples, the mapping is learned separately from training the RNN using a separate training process. In other examples, the sample can represent a character in a sequence of characters (e.g., a token) which may be mapped to vectors of numerical values. This mapping from a sequence of characters to tokens to vectors of numerical values can be learned during training of the RNN or not, and may be learned or not. In some examples, the mapping is learned separately from training the RNN using a separate training process.”; In other words, the dictionary is only used for vectorizing the input data so that the vectorized data can be trained. The initialization step includes the vectorization of the input data and obtaining the initial parameters. The training step includes re-running the vectorized data to adjust the parameters until an acceptable level is achieved (e.g., minimum error). Since the learning process (i.e., training) does not use the lookup table (i.e., word dictionary data), the lookup table is detached after the initialization of the parameters.).
REGARDING CLAIM 3, ELKIND discloses the machine learning method according to claim 1, wherein:
the initializing of the parameters includes setting, in the machine learning model, distributed representation data in which distributed representation vectors each assigned to one of the plurality of words are mapped to the codes (Par 115 –"After the network is adequately trained to reconstruct the inputs, the last output and state at block 950 extracted from the one or more encoding RNN layers 940 may represent the source data input, albeit at the same or a reduced dimensionality.”; Par 127 – “In some other examples, the output of encoder RNN 725 is further processed before being used as input to a supervised or unsupervised machine learning technique, for example using principal component analysis, t-distributed stochastic neighbor embedding, random projections, or other techniques.”), and the machine learning model includes an embedding layer for converting the codes included in the encoded text data into the distributed representation vectors (Par 115 – “After the network is adequately trained to reconstruct the inputs, the last output and state at block 950 extracted from the one or more encoding RNN layers 940 may represent the source data input, albeit at the same or a reduced dimensionality. … The output and state of the trained encoder RNN layers is a reduced dimensional representation of the input. In this fashion, the system shown in FIG. 9 produces a reduced dimensionality or embedding of the input to the system. In other examples, the dimensionality of the output is not reduced, and the representation of the encoder RNN has more desirable properties such as reduced noise, increased signal, and/or more sparsity. In some examples, the output of one of the one or more encoder RNNs represent an embedding of source data. In other examples, the output of the optional one or more fully connected layers may represent an embedding of the source data.”).
REGARDING CLAIM 4, ELKIND discloses a non-transitory computer-readable recording medium storing therein a computer program that causes a computer to execute a process comprising: performing the steps of claim 1; thus, it is rejected under the same rationale.
REGARDING CLAIM 5, ELKIND discloses an information processing apparatus comprising:
a memory (Fig. 7 memory 702) configured to store word dictionary data in which strings each representing one of a plurality of words used in a natural language are mapped to codes each identifying one of the plurality of words and text data written in the natural language (Par 76 – “An example RNN layer is shown in FIG. 4A. The RNN shown in FIG. 4A analyzes a value at time (sample or token) t. In the case of source data being vectorized, by, e.g., a Shannon Entropy calculation, the vector samples can be directly processed by the system. In the case of tokenized inputs (e.g., natural language or command line input), the tokens are first vectorized by a mapping function. One such mapping function is a lookup table mapping a token to a vectorized input.”; Par 138 – “In FIG. 10, the tokenized inputs 1005 may be represented to computational layers as real-valued vectors or initial “embeddings” representing the input. These initial embeddings can be created using word2vec, one-hot encoding, feature hashing, or latent semantic analysis, or other techniques.”); and a
processor coupled to the memory and the processor (Fig. 7 processor) configured to: perform the steps of claim 1; thus, it is rejected under the same rationale.
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 of this title, 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.
Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over ELKIND (US 2019/0273509 A1), and in further view of AGHA (US 2020/0210523 A1).
REGARDING CLAIM 2, ELKIND discloses the machine learning method according to claim 1, further comprising:
acquiring, by the processor, prediction result data including the code of at least one of the plurality of words by entering input data to the machine learning model after the learning process (Par 115 – “After the network is adequately trained to reconstruct the inputs, the last output and state at block 950 extracted from the one or more encoding RNN layers 940 may represent the source data input, albeit at the same or a reduced dimensionality.”; Par 115 – “The output and state of the trained encoder RNN layers is a reduced dimensional representation of the input. In this fashion, the system shown in FIG. 9 produces a reduced dimensionality or embedding of the input to the system. In other examples, the dimensionality of the output is not reduced, and the representation of the encoder RNN has more desirable properties such as reduced noise, increased signal, and/or more sparsity. In some examples, the output of one of the one or more encoder RNNs represent an embedding of source data. In other examples, the output of the optional one or more fully connected layers may represent an embedding of the source data.”) and
converting, [based on the word dictionary data], the code included in the prediction result data into the string representing the at least one of the plurality of words (Par 113 – “The output or reconstructed input at block 980 may be compared to the input to the system. In some examples, the comparison of the output is to the input of encoding RNN layer at block 940. In other examples, the output of decoding RNN layer at block 970 is compared against the input of convolutional filter at block 930.”).
ELKIND does not explicitly teach the [square-bracketed] limitation. In other words, ELKIND teach comparing the reconstructed input with the input; thus, ELKIND teaches the reconstructed input is in the same format as the input, i.e., string representing the words. However, ELKIND does not explicitly teach the converting is based on the word dictionary data (e.g., the lookup table).
AGHA discloses the [square-bracketed] limitation. AGHA discloses a method/system for encoding/decoding natural language data comprising: converting, by the processor, text data written in the natural language into encoded text data by encoding words included in the text data based on the word dictionary data (AGHA Par 37 – “In one implementation, each language-specific input component first maps the words in the input training example into embeddings using a pre-generated lookup table. Each language-specific input component then maps the embeddings into a language-specific representation of the training example. It performs this task using a machine-trained model. A language-agnostic encoder component (LAEC) 120 then maps the language-specific representations provided by the language-specific encoder components (114, 116, 118) into language-agnostic representations of the set of input training examples. The LAEC 120 performs this task using another machine-trained model.”);
acquiring, by the processor, prediction result data including the code of at least one of the plurality of words by entering input data to the machine learning model after the learning process (AGHA Par 40 –“ In a final stage, a set of language-specific output components (126, 128, 130) perform the predictive function of the language model component 104. They do this by converting the different language-specific decoded representations provided by the LSDC 122 into a set of predictive output results in the respective natural languages (e.g., English, French, Chinese, etc.). For example, each language-specific output component can generate an output result which provides a prediction of a next word that will follow the corresponding input training example. In one implementation, each language-specific output component uses a machine-trained model to map a language-specific decoded representation into an intermediary output result. It then multiplies the intermediary output result by the transpose of the embedding table used by the counterpart language-specific encoder component. This yields the output result of the language-specific output component.”) and converting, [based on the word dictionary data] (AGHA Par 40 –“In one implementation, each language-specific output component uses a machine-trained model to map a language-specific decoded representation into an intermediary output result. It then multiplies the intermediary output result by the transpose of the embedding table used by the counterpart language-specific encoder component. This yields the output result of the language-specific output component.”), the code included in the prediction result data into the string representing the at least one of the plurality of words (AGHA Par 40 –“In one implementation, each language-specific output component uses a machine-trained model to map a language-specific decoded representation into an intermediary output result. It then multiplies the intermediary output result by the transpose of the embedding table used by the counterpart language-specific encoder component. This yields the output result of the language-specific output component.”; Par 79 –“In block 1010, for each natural language associated with the input training examples, the training system 102 uses a language-specific output component (e.g., language-specific output component 126) to convert the language-specific decoded representation into an output result expressed in the natural language.”).
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 method/system of ELKIND to include converting back into the string representation based on the word mapping table, as taught by AGHA.
One of ordinary skill would have been motivated to include converting back into the string representation based on the word mapping table, in order to accurately provide the machine-learning results to the user in a human-readable format.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JONATHAN C KIM whose telephone number is (571)272-3327. The examiner can normally be reached Monday to Friday 8:00 AM thru 4:00 PM EST.
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, Andrew C Flanders can be reached at 571-272-7516. 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.
/JONATHAN C KIM/Primary Examiner, Art Unit 2655