DETAILED ACTION
Summary and Status of Claims
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This Office Action is in response to Applicant’s reply filed 4/9/2026.
Claim 7 is cancelled.
Claim 21 is new.
Claims 1-6 and 8-21 are pending.
Claims 3, 10-12, and 15 are rejected under 35 U.S.C. 112(b).
Claims 1-4, 9, 10, 12-16, 18, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Koshinaka (US Patent Pub 2012/0239400) in view of Ronen et al. (“DeepDPM: Deep Clustering with an Unknown Number of Clusters”, June 2022), further in view of Kindermans et al. (“Learning How to Explain Neural Networks: PatternNet and PatternAttribution”, 10/24/2017).
Claims 5, 11, 17, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Koshinaka (US Patent Pub 2012/0239400) in view of Ronen et al. (“DeepDPM: Deep Clustering with an Unknown Number of Clusters”, June 2022), further in view of Kindermans et al. (“Learning How to Explain Neural Networks: PatternNet and PatternAttribution”, 10/24/2017), further in view of Patterson et al. (US Patent Pub 2007/0174267).
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Koshinaka (US Patent Pub 2012/0239400) in view of Ronen et al. (“DeepDPM: Deep Clustering with an Unknown Number of Clusters”, June 2022), further in view of Kindermans et al. (“Learning How to Explain Neural Networks: PatternNet and PatternAttribution”, 10/24/2017), further in view of Aggarwal et al. (US Patent 12,130,841).
Claims 6 and 21 are objected to for being directed to allowable subject matter.
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
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.
Claims 3, 10-12, and 15 are rejected under 35 U.S.C. 112(b) as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor regards as the invention.
Claim 3 recites “wherein the raw data comprises … a medical image …” which is a single piece of data. However, base claim 1 recites “at least two pieces of data”. In interpretations where the raw data comprises a single medical image, it is unclear how the method of claim 1 is performed. Clarification is required.
Claim 10 recites “aggregating target neurons at the first output layer … ”. However, base claim 1 recites “wherein the one or more first neurons comprise a target neuron.” In other words, regardless of whether there is one “first neuron” or a plurality of “first neurons” they can only comprise a single “target neuron.” Therefore, it is unclear how aggregation of “target neurons” (i.e., a plurality of neurons) can be achieved in claim 10. Clarification is required.
Claims 12 and 15 recite similar limitations as claim 3 and are rejected for the same reasons.
Claim 11 is rejected because it depends on a rejected claim.
Note on Prior Art Rejections
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 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.
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.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied 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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-4, 9, 10, 12-16, 18, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Koshinaka (US Patent Pub 2012/0239400) in view of Ronen et al. (“DeepDPM: Deep Clustering with an Unknown Number of Clusters”, June 2022) (Ronen), further in view of Kindermans et al. (“Learning How to Explain Neural Networks: PatternNet and PatternAttribution”, 10/24/2017) (Kindermans).
In regards to claim 1, Koshinaka discloses a method comprising:
providing a first clustering model having a first output layer that comprises one or more first neurons, wherein the one or more first neurons comprise a target neuron (Koshinaka at para. 0125)1;
obtaining one or more first classes of raw data based on outputs from the one or more first neurons, wherein the one or more first classes comprise a first target class, and wherein the targe neuron outputs the first target class (Koshinaka at para. 0125)2; and
splitting the target neuron to obtain a second clustering model when a first similarity between at least two pieces of data corresponding to the first target class meets a splitting condition (Koshinaka at para. 0199)3,
wherein the second clustering model has a second output layer that comprises second neurons (Koshinaka at para. 0199), and
wherein each of the second neurons outputs a second class. Koshinaka at para. 0199.4
Koshinaka does not expressly disclose that the clustering models are deep clustering models.
Ronen discloses a non-parametric deep clustering model that utilizes a probability calculation (i.e., similarity) to determine whether to split or merge clusters. Ronen at pg. 5, section 4.2.
Koshinaka and Ronen are analogous art because they are directed to the same field of endeavor of clustering models.
At the time before the effective filing date of the instant application, it would have been obvious to one of ordinary skill in the art to modify Koshinaka by adding the features of making the clustering models deep clustering models, as disclosed by Ronen.
The motivation for doing so would have been deep learning models can cluster large and high dimensional datasets better than non-deep clustering methods. Ronen at pg. 1, section 1.
Koshinaka in view of Ronen does not expressly disclose wherein each weight of the second neurons is a superposition value of a weight of the target neuron and Gaussian noise.
Kindermans discloses methods to explain linear models used in multilayer networks with millions of parameters (i.e., deep networks). Kindermans at abstract. Kindermans explains linear models, wherein weight vectors are affected by distractors and Gaussian noise. Kindermans explains that it is well known that adding Gaussian noise shrinks a weight vector. Kindermans at pg. 3. Kindermans goes on to explain that in linear models, the back projections for linear components correspond to superpositions of what are assumed to be the signal directions (i.e., weight vector) of each neuron. Kindermans at pg. 4. Kindermans proposes signals are approximated as a superposition of neuron-wise nonlinear signal estimators in each layer, which is equal to the computation of the gradient during a backward pass of the weights of the network (i.e., first neurons of the first output layer). Kindermans at pg. 6-7. Accordingly, weights of the second layer are based on a superposition value of weights of the first layer (i.e., target neuron) and all weights are affected by Gaussian noise, which shrinks the weight.
Koshinaka, Ronen, and Kindermans are analogous art because they are directed to the same field of endeavor of clustering and classifier models.
At the time before the effective filing date of the instant application, it would have been obvious to one of ordinary skill in the art to modify Koshinaka in view of Ronen by adding the features of wherein each weight of the second neurons is a superposition value of a weight of the target neuron and Gaussian noise, as disclosed by Kindermans.
The motivation for doing so would have been because doing so improves layer-wise relevance propagation. Kindermans at pg. 7.
In regards to claim 2, Koshinaka in view of Ronen and Kindermans discloses the method of claim 1, wherein a number of the one or more first classes is preset. Koshinaka at para. 0125.5
In regards to claim 3, Koshinaka in view of Ronen and Kindermans discloses the method of claim 1, wherein the raw data comprises picture data, text data, voice data, one or more medical images, or video data. Koshinaka at para. 0045.6
In regards to claim 4¸ Koshinaka in view of Ronen and Kindermans discloses the method of claim 1, further comprising:
determining probability distributions of the at least two pieces of data (Koshinaka at para. 0199)7; and
obtaining, based on a second similarity between the probability distributions, the first similarity. Koshinaka at para. 0199.8
In regards to claim 9, Koshinaka in view of Ronen and Kindermans discloses the method of claim 1, further comprising obtaining the second class using a trained second deep clustering model. Ronen at pg. 5, section 4.2.9
In regards to claim 10, Koshinaka in view of Ronen and Kindermans discloses the method of claim 1, further comprising:
aggregating one or more target neurons at the first output layer to obtain a third deep clustering model when the first similarity meets the splitting condition and when a second similarity between the data corresponding to the first target class and second data corresponding to a second target class meets an aggregation condition (Koshinaka at paras. 0199-0200)10,
wherein the one or more first neurons comprise the one or more target neurons that output the first target class and the second target class (Koshinaka at paras. 0199-200)11,
wherein the third deep clustering model comprises a third output layer that comprises third neurons resulting from aggregating the one or more target neurons, and wherein each of the third neurons outputs a third class. Koshinaka at para. 0200.12
In regards to claim 12, Koshinaka discloses a deep clustering system, comprising:
one or more processors (Koshinaka at para. 0077); and
a non-transitory computer-readable storage media storing instructions that, when executed by the one or more processors, configure the deep clustering system (Koshinaka at para. 0077) to comprise:
an intelligent device configured to collect raw data, wherein the raw data comprises picture data, text data, voice data, a medical image, or video data (Koshinaka at paras. 0041-42)13; and
a deep clustering apparatus coupled to the intelligent device (Koshinaka at para. 0041) and configured to:
cluster the raw data (Koshinaka at para. 0050)14;
provide a first clustering model having a first output layer that comprises one or more first neurons, wherein the one or more first neurons comprise a target neuron (Koshinaka at para. 0125)15;
obtain one or more first classes of the raw data based on outputs from the one or more first neurons, wherein the one or more first classes comprises a first target class, and wherein the target neuron outputs the first target class (Koshinaka at para. 0125)16; and
split the target neuron to obtain a second clustering model when a first similarity between at least two pieces of data corresponding to the first target class meets a splitting condition (Koshinaka at para. 0199)17,
wherein the second clustering model has a second output layer that comprises second neurons (Koshinaka at para. 0199), and
wherein each of the second neuron outputs a second class. Koshinaka at para. 0199.18
Koshinaka does not expressly disclose that the clustering models are deep clustering models.
Ronen discloses a non-parametric deep clustering model that utilizes a probability calculation (i.e., similarity) to determine whether to split or merge clusters. Ronen at pg. 5, section 4.2.
Koshinaka and Ronen are analogous art because they are directed to the same field of endeavor of clustering models.
At the time before the effective filing date of the instant application, it would have been obvious to one of ordinary skill in the art to modify Koshinaka by adding the features of making the clustering models deep clustering models, as disclosed by Ronen.
The motivation for doing so would have been deep learning models can cluster large and high dimensional datasets better than non-deep clustering methods. Ronen at pg. 1, section 1.
Koshinaka in view of Ronen does not expressly disclose wherein each weight of the second neurons is a superposition value of a weight of the target neuron and Gaussian noise.
Kindermans discloses methods to explain linear models used in multilayer networks with millions of parameters (i.e., deep networks). Kindermans at abstract. Kindermans explains linear models, wherein weight vectors are affected by distractors and Gaussian noise. Kindermans explains that it is well known that adding Gaussian noise shrinks a weight vector. Kindermans at pg. 3. Kindermans goes on to explain that in linear models, the back projections for linear components correspond to superpositions of what are assumed to be the signal directions (i.e., weight vector) of each neuron. Kindermans at pg. 4. Kindermans proposes signals are approximated as a superposition of neuron-wise nonlinear signal estimators in each layer, which is equal to the computation of the gradient during a backward pass of the weights of the network (i.e., first neurons of the first output layer). Kindermans at pg. 6-7. Accordingly, weights of the second layer are based on a superposition value of weights of the first layer (i.e., target neuron) and all weights are affected by Gaussian noise, which shrinks the weight.
Koshinaka, Ronen, and Kindermans are analogous art because they are directed to the same field of endeavor of clustering and classifier models.
At the time before the effective filing date of the instant application, it would have been obvious to one of ordinary skill in the art to modify Koshinaka in view of Ronen by adding the features of wherein each weight of the second neurons is a superposition value of a weight of the target neuron and Gaussian noise, as disclosed by Kindermans.
The motivation for doing so would have been because doing so improves layer-wise relevance propagation. Kindermans at pg. 7.
In regards to claim 13, Koshinaka discloses a computing device, comprising:
a memory configured to store instructions (Koshinaka at paras. 0021-2); and
one or more processors coupled to the memory, wherein when executed by the one or more processors (Koshinaka at paras. 0021-2), the instructions cause the computing device to:
provide a first clustering model having a first output layer that comprises one or more first neurons, wherein the one or more first neurons comprise a target neuron (Koshinaka at para. 0125)19;
obtain one or more first classes of the raw data based on outputs from the one or more first neurons, wherein the one or more first classes comprises a first target class, and wherein the target neuron outputs the first target class (Koshinaka at para. 0125)20; and
split the target neuron to obtain a second clustering model when a first similarity between at least two pieces of data corresponding to the first target class meets a splitting condition (Koshinaka at para. 0199)21,
wherein the second clustering model has a second output layer that comprises second neurons (Koshinaka at para. 0199), and
wherein each of the second neuron outputs a second class. Koshinaka at para. 0199.22
Koshinaka does not expressly disclose that the clustering models are deep clustering models.
Ronen discloses a non-parametric deep clustering model that utilizes a probability calculation (i.e., similarity) to determine whether to split or merge clusters. Ronen at pg. 5, section 4.2.
Koshinaka and Ronen are analogous art because they are directed to the same field of endeavor of clustering models.
At the time before the effective filing date of the instant application, it would have been obvious to one of ordinary skill in the art to modify Koshinaka by adding the features of making the clustering models deep clustering models, as disclosed by Ronen.
The motivation for doing so would have been deep learning models can cluster large and high dimensional datasets better than non-deep clustering methods. Ronen at pg. 1, section 1.
Koshinaka in view of Ronen does not expressly disclose wherein each weight of the second neurons is a superposition value of a weight of the target neuron and Gaussian noise.
Kindermans discloses methods to explain linear models used in multilayer networks with millions of parameters (i.e., deep networks). Kindermans at abstract. Kindermans explains linear models, wherein weight vectors are affected by distractors and Gaussian noise. Kindermans explains that it is well known that adding Gaussian noise shrinks a weight vector. Kindermans at pg. 3. Kindermans goes on to explain that in linear models, the back projections for linear components correspond to superpositions of what are assumed to be the signal directions (i.e., weight vector) of each neuron. Kindermans at pg. 4. Kindermans proposes signals are approximated as a superposition of neuron-wise nonlinear signal estimators in each layer, which is equal to the computation of the gradient during a backward pass of the weights of the network (i.e., first neurons of the first output layer). Kindermans at pg. 6-7. Accordingly, weights of the second layer are based on a superposition value of weights of the first layer (i.e., target neuron) and all weights are affected by Gaussian noise, which shrinks the weight.
Koshinaka, Ronen, and Kindermans are analogous art because they are directed to the same field of endeavor of clustering and classifier models.
At the time before the effective filing date of the instant application, it would have been obvious to one of ordinary skill in the art to modify Koshinaka in view of Ronen by adding the features of wherein each weight of the second neurons is a superposition value of a weight of the target neuron and Gaussian noise, as disclosed by Kindermans.
The motivation for doing so would have been because doing so improves layer-wise relevance propagation. Kindermans at pg. 7.
Claims 14-16 are essentially the same as claims 2-4, respectively, in the form of a computing device. Therefore, they are rejected for the same reasons.
Claims 18 and 19 are essentially the same as claims 2 and 4, respectively, in the form of a system. Therefore, they are rejected for the same reasons.
Claims 5, 11, 17, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Koshinaka (US Patent Pub 2012/0239400) in view of Ronen et al. (“DeepDPM: Deep Clustering with an Unknown Number of Clusters”, June 2022) (Ronen), further in view of Kindermans et al. (“Learning How to Explain Neural Networks: PatternNet and PatternAttribution”, 10/24/2017) (Kindermans), further in view of Patterson et al. (US Patent Pub 2007/0174267) (Patterson).
In regards to claim 5, Koshinaka in view of Ronen and Kindermans discloses the method of claim 1, but does not expressly disclose wherein the splitting conditions comprises that a first Jensen-Shannon (JS) divergence is greater than a splitting threshold, and wherein the first JS divergence indicates the first similarity.
It is noted that Koshinaka discloses a similarity calculation is used to determine whether to split a cluster based on the calculating being greater than a threshold. Koshinaka at para. 0199.
Patterson discloses a system and method for computer aided document retrieval. The method includes determining cluster attractors for a plurality of documents, calculating a probability distribution of a term, and calculating the entropy of the probability distribution. Patterson at abstract. Patterson further discloses calculating a Jensen Shannon divergence between probability distributions to estimate the similarity. Patterson at para. 0072.
Koshinaka, Ronen, Kindermans, and Patterson are analogous art because they are directed to the same field of endeavor of classifiers and clustering.
At the time before the effective filing date of the instant application, it would have been obvious to one of ordinary skill in the art to modify Koshinaka in view of Ronen and Kindermans by adding the features of wherein the splitting conditions comprises that a first Jensen-Shannon (JS) divergence is greater than a splitting threshold, and wherein the first JS divergence indicates the first similarity, as disclosed by Patterson.
The motivation for doing so would have been because using JS divergence results in a more natural for estimating similarity. Patterson at para. 0031.
In regards to claim 11, Koshinaka in view of Ronen and Kindermans discloses the method of claim 1, but does not expressly disclose wherein the aggregation condition comprises that a second Jensen-Shannon (JS) divergence is less than an aggregation threshold, and wherein the second JS divergence indicates the second similarity.
It is noted that Koshinaka discloses a similarity calculation is used to determine whether to merge (i.e., aggregate) clusters based on the calculating being less than a threshold. Koshinaka at para. 0200.
Patterson discloses a system and method for computer aided document retrieval. The method includes determining cluster attractors for a plurality of documents, calculating a probability distribution of a term, and calculating the entropy of the probability distribution. Patterson at abstract. Patterson further discloses calculating a Jensen Shannon divergence between probability distributions to estimate the similarity. Patterson at para. 0072.
Koshinaka, Ronen, Kindermans, and Patterson are analogous art because they are directed to the same field of endeavor of classifiers and clustering.
At the time before the effective filing date of the instant application, it would have been obvious to one of ordinary skill in the art to modify Koshinaka in view of Ronen and Kindermans by adding the features of wherein the aggregation condition comprises that a second Jensen-Shannon (JS) divergence is less than an aggregation threshold, and wherein the second JS divergence indicates the second similarity, as disclosed by Patterson.
The motivation for doing so would have been because using JS divergence results in a more natural for estimating similarity. Patterson at para. 0031.
Claim 17 is essentially the same as claim 5 in the form of a computing device. Therefore, it is rejected for the same reasons.
Claim 20 is essentially the same as claim 5 in the form of a system. Therefore, it is rejected for the same reasons.
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Koshinaka (US Patent Pub 2012/0239400) in view of Ronen et al. (“DeepDPM: Deep Clustering with an Unknown Number of Clusters”, June 2022) (Ronen), further in view of Kindermans et al. (“Learning How to Explain Neural Networks: PatternNet and PatternAttribution”, 10/24/2017) (Kindermans), further in view of Aggarwal et al. (US Patent 12,130,841) (Aggarwal).
In regards to claim 8, Koshinaka in view of Ronen and Kindermans discloses the method of claim 1, but does not expressly disclose further comprising training the second deep clustering model by:
inputting the raw data and a number of second classes into the second deep clustering model; and
optimizing and updating weights of the second neurons at the second output layer in a training process to obtain a trained second deep clustering model.
Aggarwal discloses a framework for dynamic clustering for events. The neural network used for clustering is trained so that it is optimized. This is done by iteratively training the neural network and modifying the model parameters associated with the clustering component. Aggarwal at col. 5, lines 59-67; col. 6, lines 1-34. The model is trained on training data including the sequence of events (i.e., raw data). Model parameters also include the number of clusters. Aggarwal at col. 19, lines 49-60.
Koshinaka, Ronen, Kindermans, and Aggarwal are analogous art because they are directed to the same field of endeavor of classifiers/clustering models.
At the time before the effective filing date of the instant application, it would have been obvious to one of ordinary skill in the art to modify Koshinaka in view of Ronen and Kindermans by adding the features of training the second deep clustering model by inputting the raw data and a number of second classes into the second deep clustering model and optimizing and updating weights of the second neurons at the second output layer in a training process to obtain a trained second deep clustering model, as disclosed by Aggarwal.
The motivation for doing so would have been to be able to more accurately perform clustering of new data based on the training data. Aggarwal at col. 6, lines 29-34.
Allowable Subject Matter
Claims 6 and 21 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Response to Amendment
Objection to claim 7 for Minor Informalities
Claim 7 is cancelled rendering the objection moot.
Rejection of Claims 3 and 10-12 under 35 U.S.C 112(b)
Applicant’s amendment to claims 3 and 10-12 is acknowledged. However, the amendment does not address the issue because the broadest reasonable interpretation of “one or more medical images,” as amended, is still one piece of data while claim 1, as set forth in the rejection above, recites “at least two pieces of data”. Applicant’s remarks also do not explain how the amendment resolves the issue.
Similarly, Applicant’s amendment to claim 10 now recites “… aggregating one or more target neurons at the first output layer …”. As explained in the rejection, the first output layer comprises “one or more first neurons, wherein the one or more first neurons comprise a target neuron.” In other words, there is only a single target neuron. As stated in the rejection, it’s unclear how a single “target neuron” is aggregated and further, the amendment of “one or more target neurons” creates a potential contradiction with claim 1 if there is more than one target neuron. Applicant’s remarks also do not explain how the amendment resolves the issue. Consequently, The rejection to claims 3 and 10-12 under 35 U.S.C. 112(b) is maintained. Additionally, claim 15 is rejected for the same reasons as claim 3.
Rejection of Claims 12 and 18-20 under 35 U.S.C 101
Applicant’s amendment to claims 12 and 18-20 is acknowledged. Consequently, the rejection to claims 12 and 18-20 under 35 U.S.C. 101 is withdrawn.
Response to Arguments
Rejection of claims 1-20 under 35 U.S.C. 103
Claim 7 is cancelled rendering its rejection moot.
Applicant’s arguments in regards to the rejections to claims 1-6 and 8-20 under 35 U.S.C. 103, have been fully considered and they are persuasive in that Ronen does not expressly disclose the weight of second neurons is a superposition value of a weight of the target neuron and Gaussian noise.
However, upon further search and consideration, new grounds of rejection are set forth above. The new grounds of rejection rely on Kindermans, which discloses aspects of deep learning neural networks including how Gaussian noise affects weights of neuron signals and how to best approximate the weight based on superposition.
Additional Prior Art
Additional relevant prior art are listed on the attached PTO-892 form. Some examples are:
Boahen et al. (US Patent Pub 202/0019839) discloses a system and method for spiking neural network computing based on threshold accumulation.
Weerasinghe et al. (US Patent Pub 2023/0256248) discloses a system and method for deep brain simulation using a model.
Wang et al. (US Patent 11,614,508) discloses a system and method for sparse representation of measurements using deep learning techniques and linear superposition of parameters.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Examiner Michael Le whose telephone number is 571-272-7970 and fax number is 571-273-7970. The examiner can normally be reached Mon-Fri 9:30 AM – 6 PM.
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, Tony Mahmoudi can be reached on 571-272-4078. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/MICHAEL LE/Examiner, Art Unit 2163
/TONY MAHMOUDI/Supervisory Patent Examiner, Art Unit 2163
1 A learning means and classification means (i.e., first clustering model) output a defined number of clusters (i.e., first neuron, wherein the first neuron comprises a target neuron).
2 The classification means outputs clusters (i.e., first classes) of the analyzed speech data (i.e., raw data).
3 A cluster is split (i.e., split the target neuron) to update the structure of the model (i.e., obtain a second clustering model) when the probability and degree of difference (i.e., a first similarity between at least two pieces of data) is larger than a threshold (i.e., meets a splitting condition).
4 The split clusters produce the new clusters (i.e., second neurons output second classes).
5 Number of clusters T is previously defined (i.e., number of the one or more first classes is preset).
6 The data being clustered is speech data (i.e., voice data).
7 A probability that session pairs (i.e., at least two pieces of data) belong to the same cluster and a degree of difference the appearance probabilities of the speakers for respective session pairs.
8 Based on the probabilities of the speakers for respective session pairs (i.e., probability distributions), a degree of difference is determined (i.e., the first similarity is obtained from the probability distribution).
9 Changing K via splits and merges is performed during training (i.e., using a trained model to obtain the second class).
10 When the splitting condition is met and the merging condition is met (i.e., aggregation condition), clusters are merged (i.e., aggregating target neurons at the first output layer …).
11 The clustering model outputs clusters (i.e., first target class and second target class).
12 Merged clusters (i.e., third neuron) outputs a merged cluster (i.e., third class).
13 Speech data (i.e., raw data comprising voice data) is received and stored to be analyzed.
14 Speech data is clustered.
15 A learning means and classification means (i.e., first clustering model) output a defined number of clusters (i.e., first neuron, wherein the first neuron comprises a target neuron).
16 The classification means outputs clusters (i.e., first classes) of the analyzed speech data (i.e., raw data).
17 A cluster is split (i.e., split the target neuron) to update the structure of the model (i.e., obtain a second clustering model) when the probability and degree of difference (i.e., a first similarity between at least two pieces of data) is larger than a threshold (i.e., meets a splitting condition).
18 The split clusters produce the new clusters (i.e., second neurons output second classes).
19 A learning means and classification means (i.e., first clustering model) output a defined number of clusters (i.e., first neuron, wherein the first neuron comprises a target neuron).
20 The classification means outputs clusters (i.e., first classes) of the analyzed speech data (i.e., raw data).
21 A cluster is split (i.e., split the target neuron) to update the structure of the model (i.e., obtain a second clustering model) when the probability and degree of difference (i.e., a first similarity between at least two pieces of data) is larger than a threshold (i.e., meets a splitting condition).
22 The split clusters produce the new clusters (i.e., second neurons output second classes).
Prosecution Insights