Detailed Action
This action is written in response to the remarks and amendments dated 7/28/26. This action is made final. 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
The Examiner is persuaded by the Applicant’s arguments pertaining to the outstanding rejections under §112 (a) and (b). Accordingly—and in view of the current claim amendments—these rejections are withdrawn.
The Applicants argue that the previous art of record does not anticipate or render obvious the claims as currently amended. The Examiner provides updated prior art rejections below necessitated by the current amendments.
Title of Invention Is Not Descriptive
The title of the invention is not descriptive. A new title is required that is clearly indicative of the invention to which the claims are directed.
Subject Matter Eligibility
In determining whether the claims are subject matter eligible, the examiner has considered and applied guidance from MPEP § 2106. The examiner finds that the independent claims are directed to the practical application of training an improved autoencoder model, with potential applications to communication and classification tasks in internet of things systems (see spec. pp. 1-2). Furthermore, the combination of steps performed in the recited method cannot be practically performed as a mental process.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103(a) which forms the basis for all obviousness rejections set forth in this Office action:
(a) A patent may not be obtained though the invention is not identically disclosed or described as set forth in section 102 of this title, if the differences between the subject matter sought to be patented and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art to which said subject matter pertains. Patentability shall not be negatived by the manner in which the invention was made.
The following references are relied upon in the rejections below:
James (James G, Witten D, Hastie T, Tibshirani R. An introduction to statistical learning. New York: springer; 2013 Jun, corrected 8th printing 2017. PP. 1-264.)
• Ross (US 2015/0257679 A1)
• Zhang (Zhang, Chenggang, et al. "An imbalanced data classification algorithm of improved autoencoder neural network." 2016 Eighth International Conference on Advanced Computational Intelligence (ICACI). IEEE, 2016.)
Claims 1-2, 4-7 and 12-13 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang and James.
Regarding claims 1, 12 and 13, Zhang discloses a training device (and a related method and non-transitory recording medium) comprising:
a first memory storing instructions; and
a first processor connected to the first memory and configured to execute the instructions to:
A memory and a processor are inherent throughout the disclosure.
acquire a data set of raw data and correct answer data; …
P. 97, sec. (IV)(A) Experimental data for classification task.
construct an encoding model that outputs a code related to the feature amount in response to an input of the raw data, a reconstruction model that outputs a reconstructed feature amount related to an original feature amount of the code in response to an input of the code, and an estimation model that outputs an estimation result related to the raw data in response to an input of the code;
P. 96, fig. 2 (reproduced below).
PNG
media_image1.png
248
490
media_image1.png
Greyscale
encoding model :: encoder fθ
reconstruction model :: decoder gθ
estimation model :: loss function LH(x, z) representing encoding error
train the encoding model and the estimation model in such a way that the estimation result matches the correct answer data based on a relationship between the code and the feature amount, wherein the relationship between the code and the feature amount is maintained as a constraint condition during the training.
P. 96, “The training nature of autoencoder neural network is to optimize reconstruction error using the given samples. Optimization target is Cross Entropy or Mean Square Error. The cost function of autoencoder neural network defined in the study is (4).”
PNG
media_image2.png
84
500
media_image2.png
Greyscale
James discloses the following further limitations which Zhang does not disclose:
calculate a feature amount using the raw data, the feature amount being calculated independently from any encoding model by performing one or more of dimension reduction, statistical processing, frequency conversion, integration, and differentiation on the raw data;
P. 231, “An Overview of Principal Components Analysis PCA is a technique for reducing the dimension of an n×p data matrix X. The first principal component direction of the data is that along which the observations vary the most. For instance, consider Figure 6.14, which shows population size (pop) in tens of thousands of people, and ad spending for a particular company (ad) in thousands of dollars, for 100 cities5. The green solid line represents the first principal component direction of the data. We can see by eye that this is the direction along which there is the greatest variability in the data. That is, if we projected the 100 observations onto this line (as shown in the left-hand panel of Figure 6.15), then the resulting projected observations would have the largest possible variance; projecting the observations onto any other line would yield projected observations with lower variance. Projecting a point onto a line simply involves finding the location on the line which is closest to the point.”
At the time of filing, it would have been obvious to a skilled machine learning engineer to apply the PCA technique disclosed by James to the autoencoder system of Zhang because, by reducing dimensionality, system engineers can expect to improve model performance (eg model training times) and reduce model size without unduly impacting model predictive performance. Both disclosures pertain to machine learning.
Regarding claim 2, Zhang discloses the further limitation wherein the first processor is configured to execute the instructions to construct the estimation model using the code output from the encoding model in response to the input of the raw data and the feature amount calculated using the raw data.
P. 96, “In Fig. 1, SAE is network structure that has a connection between layers, but has no connection inside each layer. L1 is visual layer, L2 and L3 is hidden layer, x is input sample, xˆ is output feature, and 1+ is bias neuron.”
(cont.) “The training nature of autoencoder neural network is to optimize reconstruction error using the given samples. Optimization target is Cross Entropy or Mean Square Error. The cost function of autoencoder neural network defined in the study is (4).”
Regarding claim 4, Zhang discloses the further limitation wherein
the first processor is configured to execute the instructions to construct the estimation model using the reconstructed feature amount output from the reconstruction model in response to the input of the code and the code.
P. 96, fig. 1 (reproduced supra), output layer L3.
Regarding claim 5, Zhang discloses the further limitation wherein
the first processor is configured to execute the instructions to train the estimation model in such a way that a mutual information amount of the code output from the encoding model in response to the input of the raw data and the feature amount calculated using the raw data is reduced.
P. 96, eqn. (4). The examiner notes that “in such a way that the mutual information amount… is reduced” is an intended result of the claimed algorithm.
Regarding claim 6, Zhang discloses the further limitation wherein
the first processor is configured to execute the instructions to construct a first pre-training model that outputs a first conversion value related to the feature amount in response to the input of the raw data,
P. 96, “In Fig. 2, sample x is corrupted to x~ according to probability Dq, which then mapped to y through encoder, which further generates z through decoder that makes it reconstruct x as possible. Lost function LH (x, z) represents reconstruction error [15].”(Emphasis added.)
train the first pre-training model in such a way that the first conversion value matches the feature amount,
P. 96, eqn. 5.
set a model parameter of the first pre-training model after training as an initial value of the encoding model, and
P. 96, “It initializes the network weights using the greedy layer-wise training method,”.
train the encoding model and the estimation model in such a way that the estimation result matches the correct answer data.
See generally p. 96.
Regarding claim 7, Zhang discloses the further limitation wherein
the first processor is configured to execute the instructions to construct a second pre-training model that outputs a second conversion value related to the raw data in response to an input of the feature amount calculated using the raw data, and wherein
P. 97, steps d) “Pretrain DAE” and 3) “Finetune DAE”.
train the second pre-training model in such a way that the second conversion value matches the estimation result,
See generally p. 96, eqns. (3) and (4) reproduced supra.
set a model parameter of the second pre-training model after training as an initial value of the estimation model, and
P. 97, steps 1) “Set parameter[s].”
train the encoding model and the estimation model in such a way that the estimation result matches the correct answer data.
See generally p. 96, eqns. (3) and (4) reproduced supra.
Claims 8-11 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang, James and Ross.
Regarding claim 8, Zhang discloses the further limitation comprising an estimation system comprising an estimation device in which an encoding model and an estimation model constructed by the training device according to claim 1 are implemented, the system comprising: …
a second memory storing instructions, and
a second processor connected to the second memory and configured to execute the instructions to
A memory and a processor are inherent throughout the Zhang disclosure.
input raw data measured by the measurement instrument to the encoding model, and
P. 97, sec. (IV)(A) Experimental data for classification task.
transmit a code output from the encoding model in response to an input of the raw data, and
P. 96, fig. 1 (reproduced supra), output from L2.
the estimation device that includes the estimation model, a third memory storing instructions, and a third processor connected to the third memory and configured to execute the instructions to …
input the received code to the estimation model, and
P. 96, fig. 1 (reproduced supra), input to L3.
output an estimation result output from the estimation model in response to an input of the code.
P. 96, fig. 1 (reproduced supra), output from L3.
Ross discloses the following further limitation which Zhang/James does not disclose:
a measurement device that includes at least one measurement instrument and the encoding model, …
Abstract: “a pair of sensor insoles that include a plurality of force sensors [and] an accelerometer”.
See also fig. 1 (reproduced below).
PNG
media_image3.png
682
380
media_image3.png
Greyscale
receive the code transmitted from the measurement device,
Abstract “a transmitter for transmitting data to a portable electronic device”.
At the time of filing, it would have been obvious to a person of ordinary skill to apply the classification system of Zhang/James to gait data, as measured and transmitted by Ross, because this would provide for the diagnosis of problems with a subject’s gait, if present.
Regarding claim 9, Zhang discloses the further limitation wherein
the second processor of the measurement device is configured to execute the instructions to calculate a feature amount using the raw data, and
P. 96, fig. 1, (reproduced supra) layer L2.
The examiner notes that a processor is inherent throughout Zhang (as well as Ross).
transmit the code output from the encoding model in response to the input of the raw data and the feature amount calculated using the raw data to the estimation device, and wherein
P. 96, fig. 1, output from layer L2.
the third processor of the estimation device is configured to execute the instructions to
receive the code and the feature amount transmitted from the measurement device,
P. 96, fig. 1, input to layer L3.
input the received code and the received feature amount to the estimation model, and
P. 96, eqns. (3) and (4).
output the estimation result output from the estimation model in response to inputs of the code and the feature amount.
P. 96, fig. 1, output from layer L3.
Regarding claim 10, Zhang discloses the further limitation wherein
the estimation device includes a reconstruction model that outputs a reconstructed feature amount related to the original feature amount of the code in response to the input of the code, and wherein the third processor of the estimation device is configured to execute the instructions to input, to the estimation model, a reconstructed feature amount output from the reconstruction model in response to the input of the code, and
See generally p. 96, describing autoencoder models.
output the estimation result output from the estimation model in response to an input of the reconstructed feature amount.
P. 96, fig. 1, output from L3.
Regarding claim 11, Ross discloses the further limitation wherein
the measurement device is worn by a user, wherein the second processor of the measurement device is configured to execute the instructions to measure sensor data related to a motion of the user, and
Abstract: “a pair of sensor insoles that include a plurality of force sensors [and] an accelerometer”.
See also fig. 1 (reproduced below).
PNG
media_image3.png
682
380
media_image3.png
Greyscale
transmit the code output from the encoding model to the estimation device in response to an input of the measured sensor data, and
Abstract “a transmitter for transmitting data to a portable electronic device”.
Zhang discloses the following further limitation:
wherein the third processor of the estimation device is configured to execute the instructions to receive the code transmitted from the measurement device and transmit information related to the estimation result to a mobile terminal having a display means with which the estimation result can be visually recognizable by the user.
See generally p. 96, describing autoencoder classification models.
Additional Relevant Prior Art
The following references were identified by the Examiner as being relevant to the disclosed invention, but are not relied upon in any rejection:
Sarkodie-Gyan discloses a technique for the automated diagnosis of gait problems using sensor data (eg accelerometers and gyroscopes). (US 2014/0100494 A1)
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
Information regarding the status of an application may be found at the USPTO Patent Center at https://patentcenter.uspto.gov. Any inquiry concerning this communication should be directed to Vincent Gonzales at (571) 270-3837. The examiner can normally be reached Monday-Friday 7 am to 4 pm MT. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Miranda Huang, can be reached at (571) 270-7092.
/Vincent Gonzales/Primary Examiner, Art Unit 2124
Prosecution Insights