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 .
Response to Arguments
Applicant's arguments filed 2-20-2026 have been fully considered but they are not persuasive.
In re pg. 9, applicant argues
PNG
media_image1.png
272
845
media_image1.png
Greyscale
In response, the Examiner respectfully disagrees.
Please see the rejection below that Ding does disclose the amended limitation.
Specifically, See C4L30-45 (33) FIG. 5 shows a graphical representation of an example restricted Boltzmann machine 500 with a reduced level of complication as compared to the Boltzmann model 100 of FIG. 1. The machine 500 contains a layer 504 of invisible/hidden units 508, 510, 512, and two layers 502, 506 of visible units 514, 516, 518; 520, 522, 524. The variables in the visible layers 502, 506 are denoted as x, y, respectively. In building the model, the variables x are the input, which will receive observed data, after the model is trained, and the variables y are the output. The variables in the hidden layer are labeled as s. The units within the same group do not interact with each other, and the units within different groups interact with each other. Examiner Note: “simulation” in view of spec [0013] “The term "simulation" herein indicates a mathematical formula or a program capable of obtaining results consistent with or close to the target observation results to be predicted based on observed data.” reads on the output of the trained model).
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 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)(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.
Claim(s) 1-4, 10-13 is/are rejected under 35 U.S.C. 102a2 as being anticipated by Ding et al (US 10339466 B1)
[Claim 1] (Currently Amended) An information processing device comprising:
at least one memory configured to store instructions (C10L60-C11L10 (95) Computers suitable for the execution of a computer program include, by way of example, can be based on general or special purpose microprocessors or both, or any other kind of central processing unit. Generally, a central processing unit will receive instructions and data from a read-only memory or a random access memory or both. The essential elements of a computer are a central processing unit for performing or executing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto-optical disks, or optical disks. However, a computer need not have such devices. Moreover, a computer can be embedded in another device, e.g., a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device, e.g., a universal serial bus (USB) flash drive, to name just a few); and
at least one processor configured to execute the instructions to (C10L10-25 (92) The term “data processing apparatus” refers to data processing hardware and encompasses all kinds of apparatus, devices, and machines for processing data, including by way of example a programmable digital processor, a digital computer, or multiple digital processors or computers. The apparatus can also be or further include special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit). The apparatus can optionally include, in addition to hardware, code that creates an execution environment for computer programs, e.g., code that constitutes processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more of them)
perform a machine learning of a model which represents the relation among observed data, an observation result, and a simulation parameter1 (C1L15-20 (2) The core of artificial intelligence and machine learning problems is to train a parametric model, e.g., a Boltzmann machine or an undirected graphical model, based on the observed training data, which may include labels and input features. The training process can benefit from using a quantum oracle for probabilistic inference. Input to the quantum oracle during training is derived from the training data and the model parameters, which maps at least part of the interactions of interconnected units of the model to the interactions of qubits in the quantum oracle. The output of the quantum oracle is used to determine values used to compute loss function values or loss function gradient values or both during a training process. The quantum oracle may also be used for inference during the prediction stage, i.e., at the performance stage, during which loss functions of different candidate output labels are compared. C6L55-65 (60) In particular, the computer 900 initializes (906) the parameters in the set Q. Using the initialized parameter set Q and the training data {circumflex over (x)}, ŷ, the computer derives and outputs (908) h.sub.0 and J.sub.0 for log p(s|{circumflex over (x)},ŷ)to the quantum oracle 902. The quantum oracle 902 then performs adiabatic annealing and outputs (910) s.sup.0 to the classical computer 900. The classable computer samples (912) x.sup.1 and y.sup.1, and based on s.sup.0 and Q, calculates and outputs (912) h.sub.1 and J.sub.1 for log p(s|x.sup.1, y.sup.1) to the quantum oracle 902. The quantum oracle again performs adiabatic annealing and outputs (914) s.sup.1 to the classical computer 900. Based on s.sup.0, s.sup.1, x.sup.1, y.sup.1, {circumflex over (x)}, ŷ, and Q, the classical computer computes the new set of parameters Q. If the computed new Q has converged, the process of training the TIBM is completed. Otherwise, the new set of parameters replaces the initialized Q and the training process continues with the step 908 until the parameters converge. Examiner Note: labels reads on “observation result”; model parameters reads on “simulation parameter” in view of spec [0013]),
the simulation parameter being required when performing a simulation for predicting the observation result based on observed data ((14) Useful models include Boltzmann machines, undirected graphical models, or stochastic recurrent neural networks. During training, model parameters are determined with the goal that the trained model optimally fits the labeled observed data from any of the problems to be solved. Part of or the entire training of the model can be computationally intractable, depending on the size or complexity of the model, or both. For example, the time in order required to collect equilibrium statistics for probabilistic inference can grow exponentially with the size (e.g., number of parameters or units in a network) of the model. Examiner Note: model parameters are needed/required during training/simulation in order to fits the labeled observed data);
the simulation parameter being a visible variable of the model (C4L30-45 (33) FIG. 5 shows a graphical representation of an example restricted Boltzmann machine 500 with a reduced level of complication as compared to the Boltzmann model 100 of FIG. 1. The machine 500 contains a layer 504 of invisible/hidden units 508, 510, 512, and two layers 502, 506 of visible units 514, 516, 518; 520, 522, 524. The variables in the visible layers 502, 506 are denoted as x, y, respectively. In building the model, the variables x are the input, which will receive observed data, after the model is trained, and the variables y are the output. The variables in the hidden layer are labeled as s. The units within the same group do not interact with each other, and the units within different groups interact with each other.) ,
the visible variable being an input variable provided to the model (C4L30-45 (33) FIG. 5 shows a graphical representation of an example restricted Boltzmann machine 500 with a reduced level of complication as compared to the Boltzmann model 100 of FIG. 1. The machine 500 contains a layer 504 of invisible/hidden units 508, 510, 512, and two layers 502, 506 of visible units 514, 516, 518; 520, 522, 524. The variables in the visible layers 502, 506 are denoted as x, y, respectively. In building the model, the variables x are the input, which will receive observed data, after the model is trained, and the variables y are the output. The variables in the hidden layer are labeled as s. The units within the same group do not interact with each other, and the units within different groups interact with each other.), and
the simulation parameter representing an uncertain factor used for executing a simulation to predict the observation result based on the observed data (C4L25-35 (32) A Boltzmann machine is trained in an iterative way: for the machine with parameters in a certain iteration, it generates samples from the equilibrium distribution of the machine, and then update the parameters based on the discrepancy of these samples and the observed training data. Examiner Note: the discrepancy implied an uncertain factor);
perform, based on a result of the machine learning, sampling of a simulation parameter to be used for machine learning (C4L25-35 (32) A Boltzmann machine is trained in an iterative way: for the machine with parameters in a certain iteration, it generates samples from the equilibrium distribution of the machine, and then update the parameters based on the discrepancy of these samples and the observed training data.); and
perform a simulation using the sampled simulation parameter (C4L25-35 (32) A Boltzmann machine is trained in an iterative way: for the machine with parameters in a certain iteration, it generates samples from the equilibrium distribution of the machine, and then update the parameters based on the discrepancy of these samples and the observed training data. Examiner Note: training means performing simulation using parameters (i.e. simulated parameter)),
wherein the at least one processor is configured to execute the instructions to perform the machine learning again based on an error evaluation on a simulation result which is a result of the simulation (C4L25-35 (32) A Boltzmann machine is trained in an iterative way: for the machine with parameters in a certain iteration, it generates samples from the equilibrium distribution of the machine, and then update the parameters based on the discrepancy of these samples and the observed training data. Examiner note: the discrepancy indicates error).
[Claim 2] The information processing device according to claim 1, wherein the at least one processor is configured to execute the instructions, after a completion of the machine learning, to extract a maximum likelihood estimate of the simulation parameter by sampling on an assumption that the simulation parameter is regarded as a variable in the model (C8L15-20 (74) Given training data z, at least two methods can be used to train the model. The first method is to maximize the log-likelihood log p(z)),.
[Claim 3] The information processing device according to claim1, wherein the model is a restricted Boltzmann machine (C4L30-45 (33) FIG. 5 shows a graphical representation of an example restricted Boltzmann machine 500 with a reduced level of complication as compared to the Boltzmann model 100 of FIG. 1. The machine 500 contains a layer 504 of invisible/hidden units 508, 510, 512, and two layers 502, 506 of visible units 514, 516, 518; 520, 522, 524. The variables in the visible layers 502, 506 are denoted as x, y, respectively. In building the model, the variables x are the input, which will receive observed data, after the model is trained, and the variables y are the output. The variables in the hidden layer are labeled as s. The units within the same group do not interact with each other, and the units within different groups interact with each other).
[Claim 4] The information processing device according to claim1, wherein the model is a neural network including three or more layers (C2L15-25 (14) Useful models include Boltzmann machines, undirected graphical models, or stochastic recurrent neural networks. During training, model parameters are determined with the goal that the trained model optimally fits the labeled observed data from any of the problems to be solved. Part of or the entire training of the model can be computationally intractable, depending on the size or complexity of the model, or both. For example, the time in order required to collect equilibrium statistics for probabilistic inference can grow exponentially with the size (e.g., number of parameters or units in a network) of the model. (C4L30-45 (33) FIG. 5 shows a graphical representation of an example restricted Boltzmann machine 500 with a reduced level of complication as compared to the Boltzmann model 100 of FIG. 1. The machine 500 contains a layer 504 of invisible/hidden units 508, 510, 512, and two layers 502, 506 of visible units 514, 516, 518; 520, 522, 524. The variables in the visible layers 502, 506 are denoted as x, y, respectively. In building the model, the variables x are the input, which will receive observed data, after the model is trained, and the variables y are the output. The variables in the hidden layer are labeled as s. The units within the same group do not interact with each other, and the units within different groups interact with each other).
[Claim 10] The information processing device according to claim1, wherein the at least one processor is configured to execute the instructions to determine that the machine learning is completed when an error indicated by the error evaluation is equal to or less than a threshold value (C2L15-25 (14) Useful models include Boltzmann machines, undirected graphical models, or stochastic recurrent neural networks. During training, model parameters are determined with the goal that the trained model optimally fits the labeled observed data from any of the problems to be solved. Part of or the entire training of the model can be computationally intractable, depending on the size or complexity of the model, or both. For example, the time in order required to collect equilibrium statistics for probabilistic inference can grow exponentially with the size (e.g., number of parameters or units in a network) of the model. C2L65-67(17) In some cases, the quantum oracle is used in minimizing a loss function of the model to be trained.).
[Claim 11] The information processing device according to claim 1, wherein the machine learning is a machine learning of binary classification based on a label of an inputted simulation parameter, the label indicating either a positive example or a negative example (page 2 right column Neven et al. “Training a Binary Classifier with the Quantum Adiabatic Algorithm,” Nov. 4, 2008. Examiner Note: label in binary classifier is either positive or negative), and wherein the at least one processor is configured to execute the instructions to generate the label based on the error estimation and thereby to perform the machine learning again based on the sampled simulation parameter (C2L15-25 (14) Useful models include Boltzmann machines, undirected graphical models, or stochastic recurrent neural networks. During training, model parameters are determined with the goal that the trained model optimally fits the labeled observed data from any of the problems to be solved. Part of or the entire training of the model can be computationally intractable, depending on the size or complexity of the model, or both. For example, the time in order required to collect equilibrium statistics for probabilistic inference can grow exponentially with the size (e.g., number of parameters or units in a network) of the model. C2L65-67 (17) In some cases, the quantum oracle is used in minimizing a loss function of the model to be trained.).
Claims 12 and 13 are drawn to claim 1 and are rejected under same rationale. Note for claim 13 (Currently Amended) A non-transitory computer readable storage medium storing a program executed by a computer, the program causing the computer to (C11L20-30 (97) Control of the various systems described in this specification, or portions of them, can be implemented in a computer program product that includes instructions that are stored on one or more non-transitory machine-readable storage media, and that are executable on one or more processing devices. The systems described in this specification, or portions of them, can be implemented as an apparatus, method, or electronic system that may include one or more processing devices and memory to store executable instructions to perform the operations described in this specification.).
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries 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.
Claim(s) 5-8 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ding et al (US 10339466 B1) in view of Herbster et al (US 2020/0005154 A1)
[Claim 5] Ding disclose wherein the at least one processor is configured to execute the instructions to perform machine learning of the first neural network based on the observed data and the observation result (C1L20-40 (2) The core of artificial intelligence and machine learning problems is to train a parametric model, e.g., a Boltzmann machine or an undirected graphical model, based on the observed training data, which may include labels and input features. The training process can benefit from using a quantum oracle for probabilistic inference…), but fails to disclose the idea of using the first neural networks to train a second neural network.
However, Hebster disclose neural networking training (thereby in the same field of endeavor) and further disclose wherein the model includes a first neural network configured in a former part of the model and a second neural network configured in a latter part of the model and perform a machine learning of the second neural network based on the output from the first neural network and the simulation parameter ([0008] Accordingly there may be provided a method of training a computer system for use in classification of an image by processing image data representing the image, comprising the steps of: compressing the image data; loading the compressed image data onto a programmable quantum annealing device comprising a Restricted Boltzmann Machine; training the Restricted Boltzmann Machine to act as a classifier of image data, thereby providing a trained Restricted Boltzmann Machine; and, using the trained Restricted Boltzmann Machine to initialize a neural network for image classification thereby providing a trained computer system for use in classification of an image. [0087] Quantum pre-training may be performed to compute a set of parameters/weight on small networks that can be used to initialize a larger network. This may employ a standard approach to make training faster and less computationally intensive. The pre-training phase is well suited to quantum computing as it is possible to construct smaller networks that use sampling. [0091] The loss is preferably computed using a contrastive divergence algorithm (CD-1). To have a more complex representation of the data, a series of RBM may be stacked to construct a Deep Belief Network (DBN). This model matches the architecture of the deep neural network herein. This enables the pre-trained parameters to be used to initialize the deep neural network. Examiner Note: the Restricted Boltzmann Machine is the first neural network and the initialized neural network is the second neural network), wherein the at least one processor is configured to execute the instructions to perform, based on a result of the machine learning of the second neural network, the sampling of the simulation parameter to be used in the machine learning of the second neural network ([0009] A programmable quantum annealing device may comprise a quantum annealer such as those produced by D-Wave, Inc. These are able to solve Quadratic Unconstrained Binary Optimization (QUBO) problems or can be used as samplers to draw samples from Boltzmann distributions. [0034] In the following non-limiting examples of illustrative embodiments of the invention, an auto-encoder is described employing e.g. a Convolutional Neural Network (CNN) structure, and a Restricted Boltzmann Machines (RBM) employing quantum sampling from a quantum annealing machine is also described. As an aid to understanding the general principles of operation of these components, and in order to make evident the advantages of their use, the following brief description of these components is provided. In summary, the use of quantum sampling overcomes computational barriers that quickly prevent the effective use of wholly classical computational techniques, and the use of an auto-encoder permits real-world data sets/problems to be adjustably compressed to adjustably suit the changing (ever-improving) capabilities of quantum computing technology. In this way, a system is provided which may dynamically respond to the improving capabilities of quantum computing technology.), and wherein the at least one processor is configured to execute the instructions to perform the machine learning of the second neural network again based on error evaluation on the simulation result obtained by using the sampled simulation parameter ([0068] The encoder part (5) is arranged to output the feature vector generated by it as an input to the input layer of the decoder part (6) via a first switching unit (7). The first switching unit is switchable between a first state (Al) which places the encoder part and the decoder part in communication in this way. Furthermore, the decoder part is arranged to calculate an appropriate loss of function (6B) which measures a difference between the decompressed image generated at the output layer of the decoder and the original uncompressed image input to the input layer of the encoder, and to input the value of that loss function into the encoder for use by the encoder in adjusting the biases and weights applied to the notes of the encoder in such a way as to minimize the value of the loss function thereby optimizing the accuracy of the compressed representation of the input image produced by the decoder. [0069] A method for calculating a loss function may be used such as would be readily apparent and available to the person skilled in the art. [0070] When the value of the loss function falls below a predetermined threshold, the encoder part is deemed to be trained. The encoder part (5) is arranged to respond to this condition by issuing a switch control signal (8) to which the switch unit (7) is responsive to switch from its initial switch state A1 to a subsequent switch state B1 which disconnects the output layer of the encoder part from the input layer of the decoder part, and places the encoder part (5) of the auto-encoder in communication with an input port of the classifier unit (3). The input port of the classifier unit comprises a second switching unit operable to switch between a first state A2 which places the input port of the classifier unit in communication with a quantum pre-training unit (11) arranged to implement quantum annealing, and a second state B2 which places the input port of the classifier unit in communication with a classical classifier unit (10).).
It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify training method of Ding to incorporate training method of Herbster.
Given the advantage of quantum pre-training (Herbster [0011] One of the restrictions of current and near term quantum computers is the amount of information that can be loaded and processed. The present invention provides a hybrid quantum/classical framework for image classification that is capable of processing large, non-binary, multichannel images. The framework is designed to be generic and scalable. It can process a variety of sizes and types of images (e.g. size, RGB, greyscale) and it is capable of scaling to larger quantum hardware as they becomes available. The present invention provides a method that does not require manual image downsampling or binarization. [0074] Once the quantum pre-training unit (11) is deemed to be optimized, such as if the loss function in question falls below a predetermined threshold value, then the quantum pre-training unit is arranged to transfer the weights and biases of the nodes of the RBM to the classical classifier unit (10). The classical classifier unit is arranged to initialize the nodes of the layers of the neural network within it, using the received weights and biases. Once the neural network within the classical classifier has been initialized (or pre-trained), the classical classifier unit is arranged to issue a switch control signal (9) to the second such unit to cause the second switch unit to switch from its first state A2 to its second state B2 thereby to disconnect the encoder unit (5) from the quantum pre-training unit (11), and to connect the encoder unit (5) to the pre-trained classical classifier unit (10). In this way, a trained classical encoder is placed in connection with a pre-trained classical classifier. FIG. 6 schematically illustrates the resulting arrangement. The consequence is that an input image may be received by the trained encoder and compressed into a feature vector of suitably small size to be received by the pre-trained classical classifier unit (10) containing a neural network weighted with weights generated efficiently using a quantum annealer.), one having ordinary skill in the art would have been motivated to make this obvious modification.
[Claim 6] Herbster disclose the information processing device according to claim 5, wherein the first neural network is a network categorized as an auto-encoder ([0013] The training of the artificial neural network may comprise training an auto-encoder.).
[Claim 7] Herbster disclose the information processing device according to claim 5, wherein the first neural network is a deep neural network ([0080] FIG. 3 illustrates an example of a pipeline of the framework: the input image is compressed using an encoder, and then inputted into a deep neural network that is initialized by training an RBM on a Quantum Computer.).
[Claim 8] Herbster disclose the information processing device according to claim 5, wherein the at least one processor is configured to execute the instructions, after a completion of the machine learning, to extract a maximum likelihood estimate of the simulation parameter by sampling on an assumption that the simulation parameter is regarded as a variable in the second neural network ([0035] An RBM comprises stochastic binary variables which are arranged in the manner of a neural network. It comprises a ‘visible layer’ and a subsequent ‘hidden layer’. An RBM is ‘restricted’ in the sense that only nodal connections between layers are allowed, and connections between nodes within a layer are forbidden. [0040] When training the RBM one may aim to find the values of the biases and the weights of the nodes of the RBM, which maximize the value of a log-likelihood function in respect of the training data applied to the RBM during its training. For example, given a fixed vector V of the training data (e.g. a so-called ‘feature vector’), the gradient of a log-likelihood function with respect to the nodal biases (c.sub.j and b.sub.i) may be:).
Claim(s) 9 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ding et al (US 10339466 B1) in view of Herbster et al (US 2020/0005154 A1) and further in view of Rose et al (US 2016/0321559 A1)
[Claim 9] Ding disclose Quadratic Unconstrained Binary Optimization (C2L45-65 (16) Generally, input to the quantum machine is derived from the model to be trained and the training data, and the quantum machine outputs the desired equilibrium statistics for the trained model. The derivation of the input takes into consideration the structure of the model and the physical structure of the quantum machine. For the D-Wave system, the input has a QUBO (Quadratic Unconstrained Binary Optimization) format. In other words, during training, one or more computational tasks of a model are mapped into a QUBO problem for the quantum machine to solve. For example, when an undirected graphical model has the same structure as the hardware connectivity of the quantum machine (e.g., the D-Wave system), exact inference is made. In other implementations, e.g., when the model is a densely connected graph such as a Boltzmann machine, the quantum machine (e.g., the D-Wave system) is applied in part of the inference to improve the power of the model by adding more interactions between variables than are provided traditional methods), but fails to explicitly disclose tabu search or genetic algorithm.
However, Rose disclose Quadratic Unconstrained Binary Optimization (thereby in the same field of endeavor) and further disclose using tabu search ([0154] A variant of tabu search was used to solve these QUBOs. Following are two hardware-focused strategies for improvement over tabu)
It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify training method of Ding to incorporate training method of Rose.
Given the advantage of tabu search (Rose [0162] So one way to use the 512 qubit quantum processor to solve the sparse coding QUBOs is to restrict K to be 33 or less and embed these problems. However this is unsatisfactory for two (related) reasons. The first is that 33 dictionary atoms typically is not enough for sparse coding on big data sets. The second is that QUBOs generated by the aforementioned procedure are really easy for tabu search at that scale. For problems this small, tabu gives excellent performance with a per problem timeout of about 10 milliseconds (about the same as the runtime for a single problem on the 512 qubit quantum processor), and since it can be run in the cloud, the tabu approach can take advantage of massive parallelism as well. So even though on a problem by problem basis, the 512 qubit quantum processor is competitive at this scale, when you gang up for instance 1,000 cores against it, the 512 qubit quantum processor loses, because there are not a thousand quantum processors available for processing.), one having ordinary skill in the art would have been motivated to make this obvious modification with predictable result of the information processing device according to claim 5, wherein the at least one processor is further configured to execute the instructions, after a completion of the machine learning, to extract an estimate of the simulation parameter by applying a tabu search or a genetic algorithm on an assumption that the simulation parameter is regarded as a variable in the second neural network.
Pertinent Prior Art
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Sarkar et al (US 20180349788 A1) disclose an introspection network, a machine-learned neural network trained to predict a weight (i.e., a parameter) value at a future training step in the training of another neural network, i.e. the target neural network, given a history of the variance of the weight in previous training steps of the target neural network. See [0005].
Amin et al (US 20180308007 A1) disclose creating and using quantum Bolzman machine. See abstract.
Upadhya et al (“An Overview of Restricted Boltzmann Machines” 2019) disclose An Overview of Restricted Boltzmann Machines and the concept of RBMs can be stacked to form deep Boltzmann machine. See abstract and section 4.
Conclusion
Claim 14 is 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.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LUT WONG whose telephone number is (571)270-1123. The examiner can normally be reached M-F 10am-6pm 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, Abdullah Al Kawsar can be reached at 5712703169. 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.
/LUT WONG/Primary Examiner, Art Unit 2127
1 [0013] The information processing device 1 calculates an optimal solution of one or more parameters (also referred to as "simulation parameters") required when simulating by modeling a causal relation between observed data and observation results. The term "simulation" herein indicates a mathematical formula or a program capable of obtaining results consistent with or close to the target observation results to be predicted based on observed data. In this example embodiment, the information processing device 1 uses the method of a restricted Boltzmann machine, which is one of machine learning, and uses sampling according to the quantum annealing. Thus, the information processing device 1 obtains the optimal simulation parameters even with a small number of training data.
Prosecution Insights