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 .
Claims 1-20 are presented for examination.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 2-7, 17 and 18 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 2 (depending on claim 1): "the audio or visual device" (singular) lacks a clear antecedent because claim 1 recites "one or more audio and/or visual devices" (plural), making it unclear which device is referenced.
Claims 6 and 18 (depending on claim 3/14): "the saved pure examples" (plural) lacks a clear antecedent because claim 1/13 recites "a saved pure example" (singular), creating ambiguity about the plural form.
Claim 7 (depending on claim 3): "saved negative signals" lacks a clear antecedent and is inconsistent because claim 1 recites "one or more negative examples" (not explicitly "signals" or "saved"), though claim 5 mentions "saving the negative example," which may partially support but still raises indefiniteness due to term shift.
Claim 17 (depending on claim 15): "the sound signal" lacks a clear antecedent because claim 13 recites "the signal" (which could be audio or visual), making "sound signal" assume an audio-specific context without explicit support, potentially indefinite if the signal is visual.
Dependent claims 3-5 are also rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to cure the deficiencies of their independent claims.
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, 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) 1-3, 8, 13-15 and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Salekin (US 2021/0005067 A1) in view of Wang (US 2019/0318268 A1) further in view of Sallem (US 2021/0065733 A1).
Regarding Claim 1, Salekin (US 2021/0005067 A1) teaches
A method for automatically detecting events, the method comprising:
receiving a signal from one or more audio and/or visual devices in a deployed environment; ([0019], " In the illustrated embodiment, the surveillance system 10 includes one or more audio input devices 12, one or more output devices 14, and a surveillance computer 20. The audio input devices 12 may comprise, in particular, one or more microphones arranged in the environment (e.g., a home, automobile, etc.) and configured to record audio surveillance signals, which are provided to the surveillance computer 20") Examiner Comments: This passage teaches receiving an audio signal from a microphone (audio device) in a deployed surveillance environment; the mapping is direct as the microphone captures the signal for processing.
running a preprocessing script for buffering the signal to a particular length to feed into a machine learning model; ([0037], " The processor 22 is configured execute program instructions corresponding to the audio feature extractor 32 to extract one or more of high level descriptor (HLD) features HLDi, ..., HLDN from the individual audio clip 102”; [0074] “the audio feature extractor 32 of the audio event detection program 30 to determine a plurality of audio features HLD1, . . . , HLDN based on the received audio clip. As discussed above, in at least one embodiment, the plurality of HLD audio features HLD1, . . . , HLDN include a set of HLD audio features HLD1 corresponding to each window segment Si in a sequence of window segments S1, . . . , SN of the audio clip having a predetermined length and a predetermined amount of overlap. As discussed above, in at least one embodiment, the processor 22 is configured to determine each set of audio features HLD1 by determining a set of LLD audio features LLDij for each sub-segment Sij in a sequence of sub-segments of each window segment Si."; ([0041], “The DCNN audio tagging model 34 utilizes a DCNN (dilated convolution neural network) as a binary classifier to detect and tag the presence of a target audio event in an audio clip.”) Examiner Comments: This passage teaches preprocessing by buffering the audio signal into window segments of particular length to feed into the ML model; the mapping is direct as it prepares the signal for the DCNN audio tagging model.
processing the buffered signal using the machine learning model to identify one or more negative examples; ([0052], "If none of the target audio events are detected in the individual audio clip 102, then the processor 22 does not execute the program instructions corresponding to the Audio2Vec feature model 36 or the BLSTM classifier model 38 and moves on to processing a next audio clip") Examiner Comments: This passage teaches using the ML model (DCNN) to process the buffered signal and identify non-presence of events (negative examples), skipping further processing; the mapping is direct as non-detection identifies negative examples.
mixing the negative examples [with a saved pure example to create one or more positive examples]; ([0033], "to generate each training audio clip, a random number of randomly selected isolated audio samples (which may include no isolated audio samples) are selected for mixture with a randomly selected background audio clip. The isolated audio samples are synthetically mixed with the background audio clip at a randomly selected position(s) to generate a training audio clip") Examiner Comments: This passage teaches mixing background audio clips (negative examples) with saved isolated audio samples.
Salekin did not specifically teach
mixing the negative examples with a saved pure example to create one or more positive examples
using the created one or more positive examples and one or more negative examples to retrain the machine learning model at an edge device without the need for human annotation.
However, Wang (US 2019/0318268 A1) teaches
using the created one or more positive examples and one or more negative examples to retrain the machine learning model at an edge device without the need for human annotation. ([0004], "machine learning models can be built from data collected at edge nodes ... to enable the detection, classification, and prediction of future events") Examiner Comments: This passage teaches retraining the ML model at an edge device using local datasets (which can include positive and negative examples from sensor data) without human annotation (automated local training); the mapping is direct as edge nodes use local data for model parameter updates.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Salekin’s teaching into Wang’s in order to perform the retraining on edge devices, as Wang teaches that edge node retraining using local datasets from sensors enables efficient detection, classification, and prediction of future events without sending large amounts of data to central servers, thereby reducing latency and bandwidth usage (Wang [Background/Summary]).
Salekin and Wang did not specifically teach
mixing the negative examples with a saved pure example to create one or more positive examples.
However, Sallem (US 2021/0065733 A1) teaches
mixing the negative examples with a saved pure example to create one or more positive examples. ([0056], " The audio data expansion system 400 can include an audio augmentation system 420 that, in a block 503, can augment the selected frames of the audio data based on the selected augmentation technique(s) to generate an augmented audio data set 402 … the audio augmentation system 420 can insert noise into the frames of the audio data 401 or dampen noise in the frames of the audio data 401 to simulate different environments.") Examiner Comments: This passage teaches mixing noise (negative examples) with audio data (saved pure examples) to create augmented data (positive examples); the mapping is direct as it generates new training examples by mixing.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Salekin and Wang’s teaching into Sallem’s in order to incorporate Sallem's audio augmentation by mixing with noise, as Sallem teaches that adding noise (mixing) to audio data generates an augmented dataset for training machine learning classifiers to identify object types from sounds, improving model robustness to real-world environmental variations (Sallem [Backgorund/Summary]). This combination enhances Salekin’s mixing approach by explicitly handling noise as negative background, making the system more applicable to noisy deployed environments on edge devices as in Wang.
Regarding Claim 2, Salekin, Wang and Sallem teach
The method of claim 1, wherein the audio or visual device includes at least one of a microphone, a video device, and an infra-red or distance sensor. (Salekin, [0008], " The system comprises: a microphone configured to record audio clips of an environment") Examiner Comments: This passage teaches the audio device as a microphone, and implies video devices in surveillance (home-security system); the mapping is direct as the system uses microphones for audio signals.
Regarding Claim 3, Salekin, Wang and Sallem teach
The method of claim 2, further comprising bundling the machine learning model with scripts for at least one of an inference event, a training event, a pure audio event, a video event, or an infra-red or distance sensor event. (Salekin, [0052], " If the classification output(s) Ctag of the DCNN audio tagging model 34 indicate that the individual audio clip 102 includes a target audio event, the processor 22 executes program instructions corresponding to the Audio2Vec feature model 36 and the BLSTM classifier model 38 to determine location(s) and/or boundaries in time of the detected target audio event(s).") Examiner Comments: This passage teaches bundling the ML model (DCNN) with scripts/models for inference (tagging and classification) and training events (synthetic data generation for training); the mapping is direct as the program includes scripts for inference and training on pure audio events.
Regarding Claim 8, Salekin, Wang and Sallem teach
The method of claim 1.
Salekin, did not teach
further comprising optimizing at least one of a software component or a hardware component executing the machine learning model.
However, Wang teaches
further comprising optimizing at least one of a software component or a hardware component executing the machine learning model. ([0061], " a global synchronization step is performed through the synchronization node 120 to update the local parameter at each node to the weighted average of the parameters of all nodes") Examiner Comments: This passage teaches optimizing the software component (model parameters) during execution; the mapping is direct as it optimizes for edge hardware efficiency.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Salekin’s teaching into Wang’s in order to perform the retraining on edge devices, as Wang teaches that edge node retraining using local datasets from sensors enables efficient detection, classification, and prediction of future events without sending large amounts of data to central servers, thereby reducing latency and bandwidth usage (Wang [Background/Summary]).
Regarding Claim 13, is a system claim corresponding to the method claim above (Claim 1) and, therefore, is rejected for the same reasons set forth in the rejection of claim 1.
Regarding Claim 14, is a system claim corresponding to the method claim above (Claim 2) and, therefore, is rejected for the same reasons set forth in the rejection of claim 2.
Regarding Claim 15, is a system claim corresponding to the method claim above (Claim 3) and, therefore, is rejected for the same reasons set forth in the rejection of claim 3.
Regarding Claim 20, is a system claim corresponding to the method claim above (Claim 8) and, therefore, is rejected for the same reasons set forth in the rejection of claim 8.
Claim(s) 4-7 and 16-19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Salekin (US 2021/0005067 A1) in view of Wang (US 2019/0318268 A1), Sallem (US 2021/0065733 A1) further in view of Nemala (US 9,640,194 B1).
Regarding Claim 4, Salekin, Wang and Sallem teach
The method of claim 3.
Salekin, Wang and Sallem did not specifically teach
wherein the preprocessing script acts as a sensor to buffer the signal received from the microphone.
However, Nemala (US 9,640,194 B1) teaches
wherein the preprocessing script acts as a sensor to buffer the signal received from the microphone. (Col. 7, lines 24-33, "The frequency analysis module 310 may be used to combine the signals from the primary microphone 106 and the secondary microphone 108 and optionally transform them into a frequency-domain for further noise suppression pre-processing.") Examiner Comments: This passage teaches the frequency analysis module (preprocessing script) acting to receive and buffer microphone signals for transformation; the mapping is direct as it processes the received signal like a sensor input for frequency analysis.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Salekin, Wang and Sallem’s teaching into Nemala’s in order to incorporate Nemala's frequency analysis preprocessing for buffering microphone signals, as Nemala teaches that transforming signals into frequency domain enables effective noise suppression in speech processing, improving accuracy in real-world noisy environments by extracting one or more first cues from first noisy speech, the clean automatic speech processing features are created using a mapping and the extracted first cues, the clean speech and noise is received and the second noisy speech is produced using the clean speech and the noise (Nemala [Backgorund/Summary]).
Regarding Claim 5, Salekin, Wang and Sallem teach
The method of claim 3.
Salekin, Wang and Sallem did not specifically teach
wherein running the signal into the machine learning model to identify negative examples comprises: calling, by at least one of the scripts, the machine learning model at initialization; extracting a spectrogram from the signal; providing a labeled positive and negative output; and saving the negative example on the edge device.
However, Nemala teaches
wherein running the signal into the machine learning model to identify negative examples comprises: calling, by at least one of the scripts, the machine learning model at initialization; extracting a spectrogram from the signal; providing a labeled positive and negative output; and saving the negative example on the edge device. (Col. 8, lines 15-41, "a learning module 480 of the training system 410 may apply one or more machine-learning algorithms such as regression trees, a non-linear transform algorithms, linear transform algorithms, statistical or heuristic algorithms, neural networks, or a GMM to determine mapping between the cues and gain coefficients using reference clean speech and noise signals."; Col. 7, lines 23-33, "optionally transform them into a frequency-domain for further noise suppression pre-processing."; Col. 9, lines 29-47, "The method 700 may commence in operation 710 with the frequency analysis module 450 receiving predetermined reference clean speech from the clean speech database 420 and predetermined reference noise from the noise database 430.") Examiner Comments: This passage teaches calling the learning module (model) for training (initialization), transforming to frequency domain (spectrogram extraction), determining mapping using clean (positive) and noise (negative) for output gains (labeled), and saving negative examples in the noise database on the device; the mapping is direct as it uses clean/positive and noise/negative for labeled training outputs and stores noise.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Salekin, Wang and Sallem’s teaching into Nemala’s in order to incorporate Nemala's frequency analysis preprocessing for buffering microphone signals, as Nemala teaches that transforming signals into frequency domain enables effective noise suppression in speech processing, improving accuracy in real-world noisy environments by extracting one or more first cues from first noisy speech, the clean automatic speech processing features are created using a mapping and the extracted first cues, the clean speech and noise is received and the second noisy speech is produced using the clean speech and the noise (Nemala [Backgorund/Summary]).
Regarding Claim 6, Salekin, Wang and Sallem teach
The method of claim 3.
Salekin, Wang and Sallem did not specifically teach
wherein the mixing the negative examples with the saved pure example to create positive examples comprises: receiving a trigger event that precedes retraining the machine learning model; mixing the saved pure examples of supported types stored in the edge device with the negative examples from an environment to create positive examples; and storing the positive examples in the edge device.
However, Nemala teaches
wherein the mixing the negative examples with the saved pure example to create positive examples comprises: receiving a trigger event that precedes retraining the machine learning model; mixing the saved pure examples of supported types stored in the edge device with the negative examples from an environment to create positive examples; and storing the positive examples in the edge device. (Col. 8, lines 1-14, "These reference clean speech and noise signals may be combined by a combination module 460 of the training system 410 into “synthetic” noisy speech signals."; Col. 9, lines 29-47, "the frequency analysis module 450 receiving predetermined reference clean speech from the clean speech database 420 and predetermined reference noise from the noise database 430.") Examiner Comments: This passage teaches receiving signals for training (trigger event), mixing clean (pure examples from database) with noise (negative from environment/database) to create synthetic noisy (positive), and storing for training use on the device; the mapping is direct as synthetic noisy is created and used for model training.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Salekin, Wang and Sallem’s teaching into Nemala’s in order to incorporate Nemala's frequency analysis preprocessing for buffering microphone signals, as Nemala teaches that transforming signals into frequency domain enables effective noise suppression in speech processing, improving accuracy in real-world noisy environments by extracting one or more first cues from first noisy speech, the clean automatic speech processing features are created using a mapping and the extracted first cues, the clean speech and noise is received and the second noisy speech is produced using the clean speech and the noise (Nemala [Backgorund/Summary]).
Regarding Claim 7, Salekin, Wang and Sallem teach
The method of claim 3.
Salekin, Wang and Sallem did not specifically teach
further comprising: calling, based on a trigger event, for re-training the machine learning model that is stored in the edge device; re-training the machine learning model on the created positive example and saved negative signals; bundling up the machine learning model to create a new edge machine learning version; and replacing the current version of edge machine learning with the new edge machine learning version.
However, Nemala teaches
further comprising: calling, based on a trigger event, for re-training the machine learning model that is stored in the edge device; re-training the machine learning model on the created positive example and saved negative signals; bundling up the machine learning model to create a new edge machine learning version; and replacing the current version of edge machine learning with the new edge machine learning version. (Col. 4, lines 51-63, "The mapping may be learned from a training database, and one such mapping may exist per frequency domain tap or per group of frequency domain taps."; Col. 08, lines 15-41, " a learning module 480 of the training system 410 may apply one or more machine-learning algorithms such as regression trees, a non-linear transform algorithms, linear transform algorithms, statistical or heuristic algorithms, neural networks, or a GMM to determine mapping between the cues and gain coefficients using reference clean speech and noise signals."; Col. 3, lines 34-39, "Embodiments described herein may be practiced on any device that is configured to receive and/or provide audio such as, but not limited to, personal computers (PCs), tablet computers, phablet computers; mobile devices, cellular phones...") Examiner Comments: This passage teaches calling for learning (re-training trigger), re-training on synthetic positive (noisy) and saved negative (noise), creating new mapping (bundled version), and using it on edge devices (replacing old mapping); the mapping is direct as new learned mappings update the model for deployment on mobile devices.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Salekin, Wang and Sallem’s teaching into Nemala’s in order to incorporate Nemala's frequency analysis preprocessing for buffering microphone signals, as Nemala teaches that transforming signals into frequency domain enables effective noise suppression in speech processing, improving accuracy in real-world noisy environments by extracting one or more first cues from first noisy speech, the clean automatic speech processing features are created using a mapping and the extracted first cues, the clean speech and noise is received and the second noisy speech is produced using the clean speech and the noise (Nemala [Backgorund/Summary]).
Regarding Claim 16, is a system claim corresponding to the method claim above (Claim 4) and, therefore, is rejected for the same reasons set forth in the rejection of claim 4.
Regarding Claim 17, is a system claim corresponding to the method claim above (Claim 5) and, therefore, is rejected for the same reasons set forth in the rejection of claim 5, with "sound signal" mapped to audio signal in Salekin.
Regarding Claim 18, is a system claim corresponding to the method claim above (Claim 6) and, therefore, is rejected for the same reasons set forth in the rejection of claim 6.
Regarding Claim 19, is a system claim corresponding to the method claim above (Claim 7) and, therefore, is rejected for the same reasons set forth in the rejection of claim 7.
Claim(s) 9-12 is/are rejected under 35 U.S.C. 103 as being unpatentable over Salekin (US 2021/0005067 A1) in view of Wang (US 2019/0318268 A1), Sallem (US 2021/0065733 A1) further in view of Bhattacharyya (US 2020/0285997 A1).
Regarding Claim 9, Salekin, Wang and Sallem teach
The method of claim 1.
Salekin, Wang and Sallem did not specifically teach
wherein retraining the machine learning model further comprises automatically detecting a trigger event for re-training a machine learning model, wherein automatically detecting the trigger event comprises: receiving a distribution of training data; creating a distribution of current data based on the distribution of training data; compare the difference between the distribution of training data and the distribution of current data; and in response to the difference being above a first threshold, detect the trigger event for re- training the machine learning model.
However, Bhattacharyya (US 2020/0285997 A1) teaches
wherein retraining the machine learning model further comprises automatically detecting a trigger event for re-training a machine learning model, wherein automatically detecting the trigger event comprises: receiving a distribution of training data; ([0486], " capturing a plurality of streams of training data representing sensor readings over a range of states of the system during a training phase; characterizing joint statistical properties of the plurality of streams of data representing sensor readings over the range of states of the system during the training phase, … determining a statistical norm for the characterized joint statistical properties that reliably distinguishes between a normal state of the system and an anomalous state of the system") Examiner Comments: This passage teaches receiving the distribution of training data as joint statistical properties during a training phase to establish a baseline norm; the mapping is direct as the statistical norm represents the training data distribution for comparison.
creating a distribution of current data based on the distribution of training data; ([0486], " capturing a plurality of streams of training data representing sensor readings over a range of states of the system during a training phase; characterizing joint statistical properties of the plurality of streams of data representing sensor readings over the range of states of the system during the training phase, comprising determining a plurality of quantitative standardized errors between a predicted value of a respective training datum, and a measured value of the respective training datum, and a variance of the respective plurality of quantitative standardized errors over time; determining a statistical norm for the characterized joint statistical properties that reliably distinguishes between a normal state of the system and an anomalous state of the system;") Examiner Comments: This passage teaches creating a distribution of current data by computing joint statistical properties in operational phases, aligned with the training norm; the mapping is direct as operational data properties are computed similarly to training for comparison.
compare the difference between the distribution of training data and the distribution of current data; ([0494], " comparing the plurality of quantitative standardized errors and the variance of the respective plurality of quantitative standardized errors with the determined statistical norm, to determine whether the plurality of streams of operational data representing the sensor readings during the operational phase represent an anomalous state of system operation ") Examiner Comments: This passage teaches comparing the difference between training (statistical norm) and current data distributions using standardized errors and variance; the mapping is direct as it quantifies shifts in distributions to detect anomalies or drift.
and in response to the difference being above a first threshold, detect the trigger event for re- training the machine learning model. ([0504], "Drift can be detected for a sensor when models no longer fit the most recent data well and the frequency of type I errors the system detects exceeds an acceptable, pre-specified threshold.") Examiner Comments: This passage teaches detecting the trigger event (drift) when differences (model fit degradation or error frequency) exceed a threshold, leading to retraining; the mapping is direct as exceeding the threshold triggers model regeneration on recent data.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Salekin, Wang and Sallem’s teaching into Bhattacharyya’s in order to incorporate Bhattacharyya's drift detection using distribution comparisons and thresholds to trigger retraining, as Bhattacharyya teaches that this enables self-adaptive anomaly detection in real-time sensor systems, ensuring models remain accurate despite changing data statistics over time by capturing multiple streams of training data representing sensor readings over a range of states of the system during a training phase, the range of states includes a normal state of the system (Bhattacharyaa [Background/Summary]).
Regarding Claim 10, Salekin, Wang, Sallem and Bhattacharyya teach
The method of claim 9.
Salekin, Wang and Sallem did not teach
wherein comparing the differences between the distribution of training data and the distribution of current data comprises measuring a Kullback- Leibler divergence between the distribution of training data and the distribution of current data.
However, Bhattacharyya teaches
wherein comparing the differences between the distribution of training data and the distribution of current data comprises measuring a Kullback- Leibler divergence between the distribution of training data and the distribution of current data. ([0496], "The method may further comprise capturing a plurality of streams of operational data representing sensor readings during an operational phase; and determining at least one of a Mahalanobis distance, a Bhattacharyya distance, Chernoff distance, a Matusita distance, a KL divergence, a Symmetric KL divergence, a Patrick-Fisher distance, a Lissack-Fu distance and a Kolmogorov distance of the captured plurality of streams of operational data with respect to the determined statistical norm") Examiner Comments: This passage teaches measuring Kullback-Leibler (KL) divergence as one of the probabilistic distance measures for comparing distributions; the mapping is direct as KL divergence is explicitly listed for evaluating differences between training norm and current data in drift detection.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Salekin, Wang and Sallem’s teaching into Bhattacharyya’s in order to incorporate Bhattacharyya's drift detection using distribution comparisons and thresholds to trigger retraining, as Bhattacharyya teaches that this enables self-adaptive anomaly detection in real-time sensor systems, ensuring models remain accurate despite changing data statistics over time by capturing multiple streams of training data representing sensor readings over a range of states of the system during a training phase, the range of states includes a normal state of the system (Bhattacharyaa [Background/Summary]).
Regarding Claim 11, Salekin, Wang, Sallem and Bhattacharyya teach
The method of claim 9.
Salekin, Wang, and Sallem did not teach
wherein comparing the differences between the distribution of training data and the distribution of current data comprises measuring the difference in accuracy between the distribution of training data and the distribution of current data.
However, Bhattacharyya teaches
wherein comparing the differences between the distribution of training data and the distribution of current data comprises measuring the difference in accuracy between the distribution of training data and the distribution of current data. ([0504], "Drift can be detected for a sensor when models no longer fit the most recent data well and the frequency of type I errors the system detects exceeds an acceptable, pre-specified threshold.") Examiner Comments: This passage teaches measuring the difference in accuracy via model fit degradation and increased Type I error frequency (false positives, indicating accuracy drop); the mapping is direct as error rates reflect accuracy differences between training and current distributions, triggering drift detection.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Salekin, Wang and Sallem’s teaching into Bhattacharyya’s in order to incorporate Bhattacharyya's drift detection using distribution comparisons and thresholds to trigger retraining, as Bhattacharyya teaches that this enables self-adaptive anomaly detection in real-time sensor systems, ensuring models remain accurate despite changing data statistics over time by capturing multiple streams of training data representing sensor readings over a range of states of the system during a training phase, the range of states includes a normal state of the system (Bhattacharyaa [Background/Summary]).
Regarding Claim 12, Salekin, Wang, Sallem and Bhattacharyya teach
The method of claim 9, further comprising re-training the machine learning model using close loop learning in response to detecting the trigger event. (Salekin, [0049], " during a training process, a set of weights and/or parameters are learned and/or optimized for all of the filters in the DCNN audio tagging model 34 for each individual target audio event based on the corresponding synthetic training dataset for the individual target audio event. In at least one embodiment, the optimized values for the set of weights and/or parameters are determined by minimizing a loss function (e.g., a mean squared loss function) that evaluates a classification output Ctag of the deep DCNN audio tagging model 34 compared to the correct classification identified by the labeled training data in the synthetic training dataset") Examiner Comments: This passage teaches re-training using closed loop (optimized loss function in iterations) in response to data (trigger implied by new data); the mapping is direct as it is iterative learning.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to AMIR SOLTANZADEH whose telephone number is (571)272-3451. The examiner can normally be reached M-F, 9am - 5pm ET.
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, Wei Mui can be reached at (571) 272-3708. 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.
/AMIR SOLTANZADEH/Examiner, Art Unit 2191