DETAILED ACTION
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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 01/18/2023 and 04/05/2024 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
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 1-9 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1
According to the first part of the analysis, in the instant case, claims 1-4 and 9 are directed to a method and claim 5-8 are directed to a device comprising at least hardware circuitry. Thus, each of the claims falls within one of the four statutory categories (i.e. process, machine, manufacture, or composition of matter).
Claim 1 recites:
Step 2A, Prong 1
“a first conversion step of converting an output from an intermediate layer using a bounded nonlinear function in a final layer of a deep neural network having the intermediate layer and the final layer” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
“a second conversion step of converting a value obtained by conversion in the first conversion step using an activation function” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
Step 2A, Prong 2
“An inference method executed by an inference device” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
This judicial exception is not integrated into a practical application.
Step 2B
“An inference method executed by an inference device” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 2 recites:
Step 2A, Prong 1
“wherein conversion in the first conversion step is performed using a nonlinear function in which a maximum value of an absolute value is not infinite and a value of an argument when taking a maximum value is not infinite” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 3 recites:
Step 2A, Prong 1
“wherein conversion in the first conversion step is performed by multiplying an output of the nonlinear function by a parameter y (where 0 < y <
∞
) determined by trial and error” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 4 recites:
Step 2A, Prong 1
“a first conversion step of converting an output from an intermediate layer using a bounded nonlinear function in a final layer of a deep neural network having the intermediate layer and the final layer” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
“a second conversion step of converting a value obtained by conversion in the first conversion step using an activation function” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
“an update step of updating a parameter of the deep neural network so that an objective function based on a value obtained by conversion in the second conversion step is optimized” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
Step 2A, Prong 2
“A learning method executed by a learning device” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
This judicial exception is not integrated into a practical application.
Step 2B
“A learning method executed by a learning device” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 5 recites:
Step 2A, Prong 1
“a first conversion to convert an output from an intermediate layer using a bounded nonlinear function in a final layer of a deep neural network having the intermediate layer and the final laver” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
“a second conversion to convert a value obtained by conversion in the first conversion using an activation function” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
Step 2A, Prong 2
“An inference device comprising conversion circuitry configured to perform” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
This judicial exception is not integrated into a practical application.
Step 2B
“An inference device comprising conversion circuitry configured to perform” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 6 recites:
Step 2A, Prong 1
“a first conversion to convert an output from an intermediate layer using a bounded nonlinear function in a final layer of a deep neural network having the intermediate layer and the final layer. and a second conversion to convert a value obtained by conversion in the first conversion step using an activation function” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
“update a parameter of the deep neural network so that an objective function based on a value obtained by conversion in the second conversion is optimized” (This step is directed to a mathematical concept. See MPEP § 2106.04(a)(2), subsection I.)
Step 2A, Prong 2
“A learning device comprising: conversion circuitry configured to perform” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
“update circuitry configured to update” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
This judicial exception is not integrated into a practical application.
Step 2B
“A learning device comprising: conversion circuitry configured to perform” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
“update circuitry configured to update” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 7 recites:
Step 2A, Prong 1
This claim recites at least the abstract idea identified above in claim 5.
Step 2A, Prong 2
“A non-transitory computer readable medium including a program for causing a computer to function as the inference device according to claim 5” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
This judicial exception is not integrated into a practical application.
Step 2B
“A non-transitory computer readable medium including a program for causing a computer to function as the inference device according to claim 5” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 8 recites:
Step 2A, Prong 1
This claim recites at least the abstract idea identified above in claim 6.
Step 2A, Prong 2
“A non-transitory computer readable medium including a program for causing a computer to function as the leaming device according to claim 6” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
This judicial exception is not integrated into a practical application.
Step 2B
“A non-transitory computer readable medium including a program for causing a computer to function as the leaming device according to claim 6” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 9 recites:
Step 2A, Prong 1
This claim recites at least the abstract idea identified above in claim 4.
Step 2A, Prong 2
“A non-transitory computer readable medium including a program for causing a computer to perform the method of Claim 4” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
This judicial exception is not integrated into a practical application.
Step 2B
“A non-transitory computer readable medium including a program for causing a computer to perform the method of Claim 4” (Mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
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.
Claims 1 and 4-9 are rejected under 35 U.S.C. 103 as being unpatentable over Kalchbrenner et al. (US-20200401874-A1) in view of Chai et al. (US-20220083839-A1).
Regarding Claim 1,
Kalchbrenner (US 20200401874 A1) teaches an inference method executed by an inference device, the method comprising:
a first conversion step of converting an output from an intermediate layer (para [0045] “Generally, the single recurrent neural network layer 210 maintains a hidden state h that is updated at each generation time step. The hidden state is a vector of numeric values.” Recurrent layer 210 (i.e. intermediate layer).) using a nonlinear function in a final layer of a deep neural network having the intermediate layer and the final layer (para [0061] “In particular, in the example of FIG. 2, the fine output layers 230 are configured to apply a weight matrix O.sub.3 to y.sub.f to generate a second projected updated hidden state, apply an element-wise non-linear activation function (e.g., the rectified liner unit (“relu”) function) to the second projected updated hidden state to generate a second activation vector” layer 230 (i.e., final layer).); and
a second conversion step of converting a value obtained by conversion in the first conversion step using an activation function (para [0061] “apply a weight matrix O.sub.4 to the second activation vector to generate second logits, and apply a softmax function (“softmax”) to the second logits to generate the second score distribution.” Softmax function (i.e., activation function).).
Kalchbrenner does not explicitly disclose
using a bounded nonlinear function in a final layer of a deep neural network
However, Chai (US 20220083839 A1) teaches
using a bounded nonlinear function (para [0059] “σ is the activation function sigmoid, and tanh is the activation function”) in a final layer of a deep neural network (para [0020] “Using a double-layer LSTM (Long Short-Term Memory) network for the deep learning algorithm to establish the accuracy compensation model of the caustic alkali concentration measuring device.”)
Kalchbrenner and Chai are analogous because they are both directed towards recurrent neural networks implementing softmax functions.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the recurrent neural network of Kalchbrenner with the error function of Chai.
Doing so would allow for improving the accuracy and reliability of the model (Chai para [0070]).
Regarding Claim 4,
Kalchbrenner (US 20200401874 A1) teaches a learning method executed by a learning device, the method comprising:
a first conversion step of converting an output from an intermediate layer (para [0045] “Generally, the single recurrent neural network layer 210 maintains a hidden state h that is updated at each generation time step. The hidden state is a vector of numeric values.” Recurrent layer 210 (i.e. intermediate layer).) using a nonlinear function in a final layer of a deep neural network having the intermediate layer and the final layer (para [0061] “In particular, in the example of FIG. 2, the fine output layers 230 are configured to apply a weight matrix O.sub.3 to y.sub.f to generate a second projected updated hidden state, apply an element-wise non-linear activation function (e.g., the rectified liner unit (“relu”) function) to the second projected updated hidden state to generate a second activation vector” layer 230 (i.e., final layer).);
a second conversion step of converting a value obtained by conversion in the first conversion step using an activation function (para [0061] “apply a weight matrix O.sub.4 to the second activation vector to generate second logits, and apply a softmax function (“softmax”) to the second logits to generate the second score distribution.” Softmax function (i.e., activation function).); and
Kalchbrenner does not explicitly disclose
using a bounded nonlinear function in a final layer of a deep neural network
an update step of updating a parameter of the deep neural network so that an objective function based on a value obtained by conversion in the second conversion step is optimized.
However, Chai (US 20220083839 A1) teaches
using a bounded nonlinear function (para [0059] “σ is the activation function sigmoid, and tanh is the activation function”) in a final layer of a deep neural network (para [0020] “Using a double-layer LSTM (Long Short-Term Memory) network for the deep learning algorithm to establish the accuracy compensation model of the caustic alkali concentration measuring device.”)
an update step of updating a parameter of the deep neural network so that an objective function based on a value obtained by conversion in the second conversion step is optimized (para [0056] “wherein, f is usually a nonlinear activation function, such as tanh and relu. s.sub.t is obtained from the hidden output s.sub.t-1 at the previous time and the input x.sub.t at the current time. The softmax function is the activation function of the output layer,” para [0060] “The accuracy compensation model of the caustic alkali concentration measuring device is constructed by using double-layer LSTM networks and one fully connected layer. The accuracy compensation model is trained and tested by using the training set and test set divided previously. The root mean square error is used as the error function for error back propagation learning of neural network.” The error function is optimized (i.e., objective function) and updates the weights (i.e., network parameter) based on the output of the softmax function (i.e., activation function that performs the second conversion).).
Kalchbrenner and Chai are analogous because they are both directed towards recurrent neural networks implementing softmax functions.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the recurrent neural network of Kalchbrenner with the error function of Chai.
Doing so would allow for improving the accuracy and reliability of the model (Chai para [0070]).
Regarding Claim 5,
Kalchbrenner (US 20200401874 A1) teaches an inference device comprising conversion circuitry configured to perform:
a first conversion to convert an output from an intermediate layer (para [0045] “Generally, the single recurrent neural network layer 210 maintains a hidden state h that is updated at each generation time step. The hidden state is a vector of numeric values.” Recurrent layer 210 (i.e. intermediate layer).) using a nonlinear function in a final layer of a deep neural network having the intermediate layer and the final laver (para [0061] “In particular, in the example of FIG. 2, the fine output layers 230 are configured to apply a weight matrix O.sub.3 to y.sub.f to generate a second projected updated hidden state, apply an element-wise non-linear activation function (e.g., the rectified liner unit (“relu”) function) to the second projected updated hidden state to generate a second activation vector” layer 230 (i.e., final layer).); and
a second conversion to convert a value obtained by conversion in the first conversion using an activation function (para [0061] “apply a weight matrix O.sub.4 to the second activation vector to generate second logits, and apply a softmax function (“softmax”) to the second logits to generate the second score distribution.” Softmax function (i.e., activation function).).
Kalchbrenner does not explicitly disclose
using a bounded nonlinear function in a final layer of a deep neural network
However, Chai (US 20220083839 A1) teaches
using a bounded nonlinear function (para [0059] “σ is the activation function sigmoid, and tanh is the activation function”) in a final layer of a deep neural network (para [0020] “Using a double-layer LSTM (Long Short-Term Memory) network for the deep learning algorithm to establish the accuracy compensation model of the caustic alkali concentration measuring device.”)
Kalchbrenner and Chai are analogous because they are both directed towards recurrent neural networks implementing softmax functions.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the recurrent neural network of Kalchbrenner with the error function of Chai.
Doing so would allow for improving the accuracy and reliability of the model (Chai para [0070]).
Regarding Claim 6,
Kalchbrenner (US 20200401874 A1) teaches a learning device comprising:
conversion circuitry configured to perform a first conversion to convert an output from an intermediate layer (para [0045] “Generally, the single recurrent neural network layer 210 maintains a hidden state h that is updated at each generation time step. The hidden state is a vector of numeric values.” Recurrent layer 210 (i.e. intermediate layer).) using a nonlinear function in a final layer of a deep neural network having the intermediate layer and the final layer (para [0061] “In particular, in the example of FIG. 2, the fine output layers 230 are configured to apply a weight matrix O.sub.3 to y.sub.f to generate a second projected updated hidden state, apply an element-wise non-linear activation function (e.g., the rectified liner unit (“relu”) function) to the second projected updated hidden state to generate a second activation vector” layer 230 (i.e., final layer).), and a second conversion to convert a value obtained by conversion in the first conversion step using an activation function (para [0061] “apply a weight matrix O.sub.4 to the second activation vector to generate second logits, and apply a softmax function (“softmax”) to the second logits to generate the second score distribution.” Softmax function (i.e., activation function).); and
Kalchbrenner does not explicitly disclose
using a bounded nonlinear function in a final layer of a deep neural network
update circuitry configured to update a parameter of the deep neural network so that an objective function based on a value obtained by conversion in the second conversion is optimized.
However, Chai (US 20220083839 A1) teaches
using a bounded nonlinear function (para [0059] “σ is the activation function sigmoid, and tanh is the activation function”) in a final layer of a deep neural network (para [0020] “Using a double-layer LSTM (Long Short-Term Memory) network for the deep learning algorithm to establish the accuracy compensation model of the caustic alkali concentration measuring device.”)
update circuitry configured to update a parameter of the deep neural network so that an objective function based on a value obtained by conversion in the second conversion is optimized (para [0056] “wherein, f is usually a nonlinear activation function, such as tanh and relu. s.sub.t is obtained from the hidden output s.sub.t-1 at the previous time and the input x.sub.t at the current time. The softmax function is the activation function of the output layer,” para [0060] “The accuracy compensation model of the caustic alkali concentration measuring device is constructed by using double-layer LSTM networks and one fully connected layer. The accuracy compensation model is trained and tested by using the training set and test set divided previously. The root mean square error is used as the error function for error back propagation learning of neural network.” The error function is optimized (i.e., objective function) and updates the weights (i.e., network parameter) based on the output of the softmax function (i.e., activation function that performs the second conversion).).
Kalchbrenner and Chai are analogous because they are both directed towards recurrent neural networks implementing softmax functions.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the recurrent neural network of Kalchbrenner with the error function of Chai.
Doing so would allow for improving the accuracy and reliability of the model (Chai para [0070]).
Regarding Claim 7,
Kalchbrenner and Chai teach the inference device according to claim 5. Kalchbrenner further teaches a non-transitory computer readable medium including a program for causing a computer to function as the inference device according to claim 5 (para [0074] “Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible non transitory storage medium for execution by, or to control the operation of, data processing apparatus.”).
Regarding Claim 8,
Kalchbrenner and Chai teach the learning device according to claim 6. Kalchbrenner further teaches a non-transitory computer readable medium including a program for causing a computer to function as the learning device according to claim 6 (para [0074] “Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible non transitory storage medium for execution by, or to control the operation of, data processing apparatus.”).
Regarding Claim 9,
Kalchbrenner and Chai teach the method of Claim 4. Kalchbrenner further teaches a non-transitory computer readable medium including a program for causing a computer to perform the method of Claim 4 (para [0074] “Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible non transitory storage medium for execution by, or to control the operation of, data processing apparatus.”).
Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over the combination of Kalchbrenner/Chai, as applied above, and further in view of Baker et al. (US-20200279165-A1).
Regarding Claim 2,
Kalchbrenner and Chai teach the inference method according to claim 1.
Kalchbrenner and Chai do not explicitly disclose
wherein conversion in the first conversion step is performed using a nonlinear function in which a maximum value of an absolute value is not infinite and a value of an argument when taking a maximum value is not infinite.
However, Baker (US 20200279165 A1) teaches
wherein conversion in the first conversion step is performed using a nonlinear function in which a maximum value of an absolute value is not infinite (para [0025] “With sigmoid outputs, the maximum activation may be any number in the range [0,1]. With a softmax normalization, the maximum score must be at least 1/3.”) and a value of an argument when taking a maximum value is not infinite (para [0025] “Training with sigmoid activations or softmax activations for output nodes are well-known to those skilled in the art of training neural networks. In either case, the activation of each output node is in the range [0,1].” activations (i.e., arguments) is less than infinity but greater than zero.).
Kalchbrenner, Chai, and Baker are analogous because they are both directed towards neural networks.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the recurrent neural network of Kalchbrenner and Chai with the activation function of Baker.
Doing so would allow for training each node separately allowing for training with multiple objective (Baker para [0029]).
Claim 3 is rejected under 35 U.S.C. 103 as being unpatentable over the combination of Kalchbrenner/Chai, as applied above, and further in view of Kuo et al. (US-20210064341-A1).
Regarding Claim 3,
Kalchbrenner and Chai teach the inference method according to claim 1. Kalchbrenner further teaches
wherein conversion in the first conversion step is performed by multiplying an output of the nonlinear function by a parameter y (where 0 < y <
∞
) determined by trial and error.
Kalchbrenner and Chai do not explicitly disclose
wherein conversion in the first conversion step is performed by multiplying an output of the nonlinear function by a parameter y (where 0 < y <
∞
) determined by trial and error.
However, Kuo (US 20210064341 A1) teaches
wherein conversion in the first conversion step is performed by multiplying an output of the nonlinear function by a parameter y (where 0 < y <
∞
) determined by trial and error (para [0022] “Different from the foregoing embodiment in which the Tan h function (or the Sigmoid function) is used as a target for approximation, in the following embodiments, a value obtained by multiplying the Tan h function (or the Sigmoid function) by a power of 2 (that is, 2.sup.N, and N is an integer) is used as a target for approximation. Herein, “multiplying the Tan h function (or the Sigmoid function) by a power of 2” may be referred to as an error estimation mechanism. Through the error estimation mechanism, the error value between the approximate value (that is, the value of the linear function) and the value of the Tan h function (or the Sigmoid function) can meet the design requirement.” An integer cannot be infinity and anything to the power of 2 is positive.).
Kalchbrenner, Chai, and Kuo are analogous because they are both directed towards neural networks.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the neural network of Kalchbrenner and Chai with the method of approximating error of the activation function of Kuo.
Doing so would allow for the error value between the approximate value and the value of the Tan h function (or the Sigmoid function) can meet the design requirement (Kuo para [0022]).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Zhao et al. (US-20200401835-A1) – discloses a softmax layer with nonlinear mapping functions.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HENRY K NGUYEN whose telephone number is (571)272-0217. The examiner can normally be reached Mon - Fri 7:00am-4:30pm.
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, Li B Zhen can be reached at 5712723768. 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.
/HENRY NGUYEN/Examiner, Art Unit 2121