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 .
Priority
The present application has a provisional application No. 63/573,255 filed on April 2, 2024.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1, 2, 3, 4, 5, 9, 11, 12, 13, 14, 15, 19, 20 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al. (Zhang), US Patent Application No. US-2023/0053431-A listed in IDS filed on April 17 2024, Published in February 23, 2023 in view of Dunne et al. (Dunne), US Patent Application No. US-2020/0272899-A1, Published on August 27, 2020.
As to independent Claim 1,
Zhang teaches an apparatus comprising:
sampling circuitry configurable to generate data representing at least one of a voltage or a current of a monitored circuit (Zhang, Pg12, Claim1, Lines7-10, "a sensor in electrical communication with the conductive path and configured to measure an electrical characteristic of the conductive path, the sensor configured to provide a plurality of sensor measurements"
Pg4, Paragraph46, Lines12-14, "In various embodiments, any one of the sensors can be arranged and configured to monitor both the phase and neutral conductive paths."
Pg4, Paragraph47, Lines14-20, "For example, high frequency sensor 22 may be configured to sense/measure high frequency signals, such as high frequency noise. Current sensor 24 may be configured to sense/measure a current value. Differential sensor 26 may be configured to sense/measure a current differential between, for example, the phase and neutral conductive paths", wherein Zhang discloses a sensor (the corresponding sampling circuitry) that is designed to sense and measure a current value of monitoring conductive paths (the corresponding circuit), which is functionally equivalent to the claimed invention);
memory (Zhang, Pg12, Claim1, Lines11-21, "a memory configured to store an arc detection program, the arc detection program configured to implement a machine learning model, the arc detection program including a field-updatable program portion configured to be field-updatable and a non-field-updatable program portion, the field-updatable program portion including a plurality of program parameters, wherein the non-field-updatable program portion is configured to use the plurality of program parameters to decide between presence of an arc event or absence of an arc event"); and
processing circuitry (Zhang, Pg12, Claim1, Lines22-25, "a controller positioned within the housing and coupled to the memory, wherein the arc detection program, when executed by the controller, causes the controller to perform an operation") configurable to:
execute a machine learning model using the data to generate a classification representative of whether an arc has occurred within the monitored circuit; cause output of a result of the classification of whether the arc has occurred within the monitored circuit (Zhang, Pg12, Claim1, Lines26-32, "computing input data for the machine learning model based on the plurality of sensor measurements, providing a decision between presence of an arc event or absence of an arc event, the decision based on the input data, and causing the switch to interrupt the conductive path when the decision indicates presence of an arc event", wherein Zhang discloses the model takes in the data generated from the above sampling sensors, or the sampling circuitry to output a decision of an arc presence to cause the switch to interrupt the path based on the output, which is functionally equivalent to the claimed invention of generating classification representatives of whether an arc is present or not and outputting the result);
record the data from the monitored circuit in the memory (Zhang, Pg10, Paragraph80, Lines11-14, "In various embodiments, the memory 720 may store the sensor measurements for computing the input data and/or may store the input data computed based on the sensor measurements.")
Zhang teaches about performing additional training of the machine learning model (Zhang, Pg10(PDF), Fig10, Step1030, "Further train the trained machine learning model based on the input data to provide an updated machine learning model"). However, Zhang does not teach that this additional training is on-site procedure. From the same field of endeavor, Dunne teaches this (Dunne, Pg15, Paragraph164, "FIG. 9 illustrates a method 900 for updating a neural network on one or more edge devices without requiring transmission of neural network data in accordance with an embodiment. In the example illustrated in FIG. 9, method 900 is performed in a system that includes an edge device 902 and a centralized site / device 904. The edge device 902 may include an inference engine 952 and a learning engine 954. In some embodiments, the learning engine 954 may be a training engine that is configured to use a collected dataset and / or inference results to determine the values of the weights in the layers of a neural engine", wherein Dunne discloses that an edge device has both inference engine and learning engine which the updating or retraining does not rely on an out-site or a cloud system, which is functionally equivalent to the claimed invention.)
Zhang and Dunne are analogous to the claimed invention as they are from the same field of endeavor of deploying and updating machine learning models on resource-constrained edge devices for real-time monitoring and anomaly detection. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date, to combine the arc fault circuit interrupter apparatus of Zhang with the on-device learning engine architecture of Dunne. The motivation is as recited by Dunne (Dunne, Pg4, Paragraph47, Lines7-10, "Such challenges are particularly acute in systems or applications where bandwidth between the centralized site / device and the edge device is restricted, limited, intermittent , and / or non-reliable" and Pg6, Paragraph64, Lines7-9, "Reducing the amount of data that is transmitted or received may extend the operational life of the edge device due to power savings") such that the incorporation of Dunne's local retraining mechanism directly onto Zhang's processing circuitry enables Zhang's internal processing circuitry to perform additional training of its arc detection model on-device using the circuit data recorded in its local memory, thereby eliminating reliance on continuous cloud/network connectivity, drastically reducing communication bandwidth and operational power consumption, and achieving real-time autonomous model adaptation to localized circuit environments and electrical noise profiles.
As to dependent Claim 2,
The combination of Zhang and Dunne teaches, as mentioned above, all the limitations of Claim 1. It teaches about the overall architecture of detecting arc faults using on-site sensors, which collect the real-time current data that is inputted into the local AI model. It also teaches about the on-site retraining mechanism of an edge device, which has both inference and training engines, that does not rely on cloud or wireless system to transfer the collected data for the update.
Zhang further teaches the apparatus of claim 1, wherein
the data recorded in the memory is labeled as not including the arc (Zhang, Pg3, Paragraph23, Lines2-9, "preparing the updated program portion having reduced false positive arc event decisions includes: in case the event data is the plurality of sensor measurements, computing the input data for the machine learning model based on the plurality of sensor measurements; associating the input data with a label indicating absence of an arc event; training an updated machine learning model using the input data and the label"
Pg8, Paragraph69, Lines8-12, "As persons skilled in the art will understand, when supervised learning is implemented, input data corresponding to an arc fault are labeled as such, and input data corresponding to no arc fault are labeled as such", wherein Zhang explicitly discloses about using non-arc labeled input data to reduce the false positive arc event decisions, rendering it functionally equivalent to the claimed invention.)
As to dependent Claim 3,
The combination of Zhang and Dunne teaches, as mentioned above, all the limitations of Claim 1. It teaches about the overall architecture of detecting arc faults using on-site sensors, which collect the real-time current data that is inputted into the local AI model. It also teaches about the on-site retraining mechanism of an edge device, which has both inference and training engines, that does not rely on cloud or wireless system to transfer the collected data for the update.
Zhang further teaches the apparatus of claim 1, wherein to perform the additional training of the machine learning model,
the processing circuitry is configurable to not alter at least a portion of the machine learning model as part of the additional training (Zhang, Pg10, Paragraph81, Lines2-7, "the arc detection program 730 includes a field-updatable program portion 732 and a non-field-updatable program portion 734. The non-field-updatable program portion 734 may include instructions which implement a machine learning model and which do not change when the machine learning model is improved by a computing system"
Pg11, Paragraph92, Lines4-10, "neural networks have nodes and certain neural networks may be implemented using the same node operations for each node. Accordingly, the non-field updatable program portion 734 may implement node operations for a neural network, and the field-updatable program portion 732 may include parameters to be used by the node operations", wherein Zhang explicitly discloses that when it comes to updating or retraining the model, there are portions that are subject to the modification and portions that are not, rendering it functionally equivalent to the claimed invention.)
As to dependent Claim 4,
The combination of Zhang and Dunne teaches, as mentioned above, all the limitations of Claim 1. It teaches about the overall architecture of detecting arc faults using on-site sensors, which collect the real-time current data that is inputted into the local AI model. It also teaches about the on-site retraining mechanism of an edge device, which has both inference and training engines, that does not rely on cloud or wireless system to transfer the collected data for the update.
Zhang further teaches the apparatus of claim 1, wherein
the processing circuitry is configurable to provide the data to model trainer circuitry for training of an updated machine learning model (Zhang, Pg9, Paragraph73, "In various embodiments, the AFCI devices 620-640n can communicate sensor data/measurements to the computing system 600 for the computing system 600 to use as further training data... In this manner, the computing system 600 can receive sensor data/measurements and/or input data from AFCI devices operating in real-world situations and can update and train machine learning models using such data", wherein Zhang discloses about using the sensor data/measurement to update and train the machine learning model (updating implies it can be applied to the updated machine learning model) from the computing system (the corresponding model trainer) albeit it relies on the remote communications over one or more networks. As mentioned above, Dunne teaches about local training of an edge device using the on-site collected data, thus the combination of Zhang and Dunne is functionally equivalent to the claimed invention.)
As to dependent Claim 5,
The combination of Zhang and Dunne teaches, as mentioned above, all the limitations of Claim 4. It teaches about the locally collected data is used to update or train the local model.
Zhang further teaches the apparatus of claim 4, wherein
the processing circuitry is configurable to store the updated machine learning model in the memory (Zhang, Pg10, Paragraph82, Lines4-8, "Updates to the field-updatable program portion 732 may be stored in the memory locations referenced by the memory pointers. The non-field-updatable program portion 734 may access the updates by accessing the same memory locations referenced by the memory pointers"
Pg1, Paragraph12, Lines2-8, "the memory is configured to store a firmware update program which, when executed by the controller, causes the controller to perform a further operation that includes causing the communication device to receive an updated program portion having reduced false positive arc event decisions, and replacing the field-updatable program portion of the arc detection program with the updated program portion", wherein Zhang discloses about the memory storing the updated model although the updated model is received using the communication device in which Dunne's local training comes in place such that the locally updated model can be stored in the same memory, rendering it functionally equivalent to the claimed invention).
As to dependent Claim 9,
The combination of Zhang and Dunne teaches, as mentioned above, all the limitations of Claim 1. It teaches about the overall architecture of detecting arc faults using on-site sensors, which collect the real-time current data that is inputted into the local AI model. It also teaches about the on-site retraining mechanism of an edge device, which has both inference and training engines, that does not rely on cloud or wireless system to transfer the collected data for the update.
Zhang further teaches the apparatus of claim 1, wherein
the processing circuitry is configurable to generate the classification without performance of a Fourier transform (Zhang, Pg6, Paragraph61, Lines8-13, "For example, signal samples/measurements from one or more sensors may be directly used as inputs to a deep learning neural network, which may be trained to decide whether the signal samples/measures indicate presence of an arc event or absence of an arc event", wherein Zhang explicitly discloses that signal samples/measures do not required to go through Fourier Transformation, which is used to transform signal information that can be perceived by the machine learning models, such that the raw data can be directly inputted into the model to be classified, rendering it functionally equivalent to the claimed invention.)
As to dependent Claim 11,
The combination of Zhang and Dunne teaches, as mentioned above, all the limitations of Claim 1. It teaches about the overall architecture of detecting arc faults using on-site sensors, which collect the real-time current data that is inputted into the local AI model. It also teaches about the on-site retraining mechanism of an edge device, which has both inference and training engines, that does not rely on cloud or wireless system to transfer the collected data for the update.
Zhang further teaches the apparatus of claim 1, wherein
the processing circuitry is to perform the additional training of the machine learning model in response to at least one of a user input, a number of samples in the data exceeding a threshold, or an instruction from an external source (Zhang, Pg9, Paragraph78, Lines5-13, "the AFCI devices may be configured to communicate sensor data or input data to the computing system 600 based on certain criteria. For example, the AFCI devices may be configured to communicate sensor data or input data to the computing system 600 when a user instructs the AFCI device to transmit the data, such as by pressing a button on the AFCI device, or engaging an interface of a connected smartphone app, or directing the transmission through a cloud interface, or by other ways"
Pg9, Paragraph78, Lines14-18, "an AFCI device may be configured to communicate sensor data or input data to the computing system 600 when the AFCI device detects a predetermined number of arc events followed by resets, which occur in a predetermined period of time", wherein Zhang explicitly discloses about user inputs, instructions from an external source and a threshold for number of samples to trigger to communicate sensor data to the computing system for additional training, rendering it functionally equivalent to the claimed invention).
As to independent Claim 12,
it is a non-transitory computer-readable medium claim that contains similar limitations of Claim 1 and thus rejected under the same rationale.
As to dependent Claim 13,
it is a non-transitory computer-readable medium claim that contains similar limitations of Claim 3 and thus rejected under the same rationale.
As to dependent Claim 14,
it is a non-transitory computer-readable medium claim that contains similar limitations of Claim 4 and thus rejected under the same rationale.
As to dependent Claim 15,
it is a non-transitory computer-readable medium claim that contains similar limitations of Claim 5 and thus rejected under the same rationale.
As to independent Claim 19,
it is a method claim that contains similar limitations of Claim 1 and thus rejected under the same rationale.
As to dependent Claim 20,
it is a method claim that contains similar limitations of Claim 3 and thus rejected under the same rationale.
Claims 6, 7, 10, 16, 17 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang and Dunne as mentioned in Claim 1 in further view of Tian et al. (Tian), Non-Patent Literature listed in IDS filed on April 17, 2024, “Arc fault detection using artificial intelligence: Challenges and benefits”, Published in May 23, 2023, 29.
As to dependent Claim 6,
The combination of Zhang and Dunne teaches, as mentioned above, all the limitations of Claim 1. It teaches about the overall architecture of detecting arc faults using on-site sensors, which collect the real-time current data that is inputted into the local AI model. It also teaches about the on-site retraining mechanism of an edge device, which has both inference and training engines, that does not rely on cloud or wireless system to transfer the collected data for the update.
However, both Zhang and Dunne do not teach the following limitations but from the same field of endeavor Tian teaches the apparatus of claim 1, wherein
the machine learning model includes a plurality of two-dimensional convolution layers followed by a fully connected layer (Tian, Pg12420, Section3.2.2, Paragraph3, Lines6-10, "Chu et al. [83] presented a new approach for the detection of series AC arc faults in residential low-voltage distribution networks. The approach utilized a high-frequency coupling sensor to gather high-frequency feature signals, which were subsequently transformed into two-dimensional gray images and analyzed by means of a three-layer convolutional neural network, as depicted in Figure 15"
Pg12422, Figure15,
PNG
media_image1.png
210
658
media_image1.png
Greyscale
, wherein Tian explicitly discloses about detecting arc faults using CNN that comprises several two-dimensional convolution layers followed by a fully connected layer, rendering it functionally equivalent to the claimed invention.)
Zhang, Dunne and Tian are analogous to the claimed invention as they are from the same field of endeavor of implementing artificial intelligence and machine learning models on embedded processors for real-time electrical circuit monitoring and arc fault detection. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date, to combine the arc fault circuit interrupter apparatus of Zhang, the on-device learning engine architecture of Dunne with Tian's application of machine learning and AI classification algorithms specifically targeting direct current arc faults in DC electrical environments. The motivation is as recited by Tian (Tian, Pg12404, Abstract, Lines2-3, "The rising demand for electricity and concomitant expansion of energy systems has resulted in a heightened risk of arc faults and the likelihood of related fire" and Pg12423, Section4, Paragraph1, Lines14-17, "AI-based arc fault detection systems have the advantage of fast fault detection, thereby reducing the risk of electrical fires. Furthermore, the algorithms can be customized to meet specific requirements and can be trained to detect specific types of faults") such that the incorporation of Tian's application enables Zhang and Dunne's processing circuity to perform on-device retraining of its arc detection model to dynamically adapt to localized DC electrical noise profiles without network latency or external bandwidth dependency, thereby providing fast, highly customized, and autonomous DC arc fault protection to prevent catastrophic electrical fires.
As to dependent Claim 7,
The combination of Zhang and Dunne teaches, as mentioned above, all the limitations of Claim 1. It teaches about the overall architecture of detecting arc faults using on-site sensors, which collect the real-time current data that is inputted into the local AI model. It also teaches about the on-site retraining mechanism of an edge device, which has both inference and training engines, that does not rely on cloud or wireless system to transfer the collected data for the update.
However, both Zhang and Dunne do not teach the following limitations but from the same field of endeavor Tian teaches the apparatus of claim 1, wherein
the electrical arc is a direct current (DC) electrical arc (Tian, Pg12418, Figure10, "Arc fault detection in DC(direct current) distribution using semi-supervised ensemble machine learning",
Pg12416, Section3.2.1., Paragraph2, Lines2-3, "A number of studies have aimed to develop machine learning algorithms to identify arc faults in DC(direct current) distribution systems", wherein Tian explicitly discloses that identifying direct current arc fault using machine learning is pretty common in the field, rendering it functionally equivalent to the claimed invention.)
As to dependent Claim 10,
The combination of Zhang and Dunne teaches, as mentioned above, all the limitations of Claim 1. It teaches about the overall architecture of detecting arc faults using on-site sensors, which collect the real-time current data that is inputted into the local AI model. It also teaches about the on-site retraining mechanism of an edge device, which has both inference and training engines, that does not rely on cloud or wireless system to transfer the collected data for the update.
However, both Zhang and Dunne do not teach the following limitations but from the same field of endeavor Tian teaches the apparatus of claim 1, wherein
the monitored circuit is a power conversion circuit (Tian, Pg12411, Paragraph1, Lines7-10, "The efficacy of the proposed solution was confirmed through experiments using a photovoltaic power supply and DC-DC converter as DC power sources, with results indicating that the method can effectively distinguish arc faults with clear physical significance and low computational requirements"
Pg12413, Paragraph1, Lines1-2, "The deployment of modern power electronic equipment, such as photovoltaic inverters and energy storage converters, which operate at high-frequency switching states"
Pg12419, Paragraph1, Lines6-7, "Le et al. utilized random forest-based detectors to design a master controller for series dc arc fault detection and localization in power electronics systems", wherein Tian explicitly discloses that arc faults can happen in the mentioned power converters, rendering it functionally equivalent to the claimed invention.)
As to dependent Claim 16,
it is a non-transitory computer-readable medium claim that contains similar limitations of Claim 6 and thus rejected under the same rationale.
As to dependent Claim 17,
it is a non-transitory computer-readable medium claim that contains similar limitations of Claim 7 and thus rejected under the same rationale.
Claims 8 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang and Dunne as mentioned in Claim 1 in further view of Claussen et al. (Claussen), Chinese Patent No. CN-113272829-A, Published in August 17, 2021.
As to dependent Claim 8,
The combination of Zhang and Dunne teaches, as mentioned above, all the limitations of Claim 1. It teaches about the overall architecture of detecting arc faults using on-site sensors, which collect the real-time current data that is inputted into the local AI model. It also teaches about the on-site retraining mechanism of an edge device, which has both inference and training engines, that does not rely on cloud or wireless system to transfer the collected data for the update.
However, both Zhang and Dunne do not teach the following limitations but from the same field of endeavor Claussen teaches the apparatus of claim 1, wherein to perform the additional training of the machine learning model,
the processing circuitry is configurable to exclude feature detection layers of the machine learning model from the additional training (Claussen, Pg7, Lines2-7, "The input layer and a series of convolution and pooling layers are used as domain independent feature extractors. The fully-connected layer of the trained CNN is replaced with a domain-dependent fully-connected layer on the edge device 105. In practice, this may be understood as preserving the complete CNN, where only the weight of the fully connected layer is allowed to change when performing the gradient descent/optimization procedure"
Pg4, Last paragraph, Lines1-5, "Using the techniques described herein, a neural network is divided into a fixed portion and a flexible portion. The fixed part is pre-trained (e.g., with millions of examples in the cloud) and used as a feature extraction layer. The fixed portion can then be deployed on the inferentially optimized portion of the hardware accelerator. The flexible part of the neural network is used to train new classes and dynamically adapt the current classifier"
Pg9, Last Paragraph, Lines3-7, "The advantage of this approach is that previously learned classifiers are frozen and not affected by the training of new classifiers. This enables classification to continue during training without loss of accuracy. Furthermore, the network can be divided into a fixed high performance optimization part and a flexible adaptation part", wherein Claussen explicitly discloses the technique adapted to retrain an edge device where the feature extraction layers that takes the heavy portion of the model is fixed during the retraining such that only the fully connected layer is being updated, rendering it functionally equivalent to the claimed invention.)
Zhang, Dunne and Claussen are analogous to the claimed invention as they are from the same field of endeavor of deploying and training machine learning models on edge hardware accelerators. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date, to combine the arc fault circuit interrupter apparatus of Zhang, the on-device learning engine architecture of Dunne with Claussen’s partitioning a trained neural network into a fixed feature extraction layer and a flexible domain-dependent layer, which excludes the feature extraction layers from retraining during edge adaptation. The motivation is as recited by Claussen (Claussen, Pg5, Lines2-6, "By performing neural network training/adaptation on low power hardware at the edge, we can achieve more flexible applications, including continuous learning-based methods. This will make the update and refinement process of the neural network significantly more efficient than conventional solutions" and Pg9, Last Paragraph, Lines3-7, "The advantage of this approach is that previously learned classifiers are frozen and not affected by the training of new classifiers. This enables classification to continue during training without loss of accuracy. Furthermore, the network can be divided into a fixed high performance optimization part and a flexible adaptation part") such that adapting the edge systems to freeze pre-trained feature detection layers and retrain only the flexible domain-dependent layers, thereby achieving fast and efficient continuous on-device learning with now computational overhead while allowing continuous inference operation.
Claims 18 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang and Dunne as mentioned in Claim 12 in further view of Yamada et al. (Yamada), Non-Patent Literature, “A SEQUENTIAL CONCEPT DRIFT DETECTION METHOD FOR ON-DEVICE LEARNING ON LOW-END EDGE DEVICES”, Published in January 30, 2023, 12 Pages.
As to dependent Claim 18,
The combination of Zhang and Dunne teaches, as mentioned above, all the limitations of Claim 12. It teaches about the overall architecture of detecting arc faults using on-site sensors, which collect the real-time current data that is inputted into the local AI model. It also teaches about the on-site retraining mechanism of an edge device, which has both inference and training engines, that does not rely on cloud or wireless system to transfer the collected data for the update.
Zhang teaches about retraining the system when predetermined number of arc events over a predetermined period of time (Zhang, Pg9, Paragraph78, Lines14-18, "an AFCI device may be configured to communicate sensor data or input data to the computing system 600 when the AFCI device detects a predetermined number of arc events followed by resets, which occur in a predetermined period of time").
However, both Zhang and Dunne do not teach the following limitations but from the same field of endeavor Yamada teaches the at least one non-transitory machine-readable medium of claim 12, wherein the machine-readable instructions are to cause one or more of the at least one processor circuit to
perform the additional training of the machine learning model in response to an amount of time having elapsed without the determination that the arc has occurred within the monitored circuit (Yamada, Pg1, Abstract, Lines1-5, "practical issue of edge AI systems is that data distributions of trained dataset and deployed environment may differ due to noise and environmental changes over time. Such a phenomenon is known as a concept drift, and this gap degrades the performance of edge AI systems and may introduce system failures. To address this gap, retraining of neural network models triggered by concept drift detection is a practical approach"
Pg3, Section2.2.2, Line1, "In the active approach, a machine learning model is retrained only when a concept drift is detected"
Pg4, Section3.1, Lines3-4, "Each discriminative model instance forms an autoencoder [13] for unsupervised anomaly detection"
Pg4, Section3.2, Lines3-6, "A drift rate is calculated based on a sum of the distance between the trained centroid and corresponding recent centroid for each label (see line 14 in Algorithm 1). Then a concept drift is detected when the drift rate exceeds a pre-determined threshold value"
Pg11, Summary, Paragraph2, Lines3-5, "the combination of the neural network retraining and the proposed concept drift detection method was demonstrated on Raspberry Pi Pico that has 264kB memory", wherein Yamada explicitly teaches an on-device machine learning system for resource-constrained microcontroller where data distribution changes over time due to noise and environmental changes (concept drift, the corresponding state of receiving only regular current data or no arc data for a period of time with noises). Yamada teaches monitoring operational data stream and triggering model retraining when a drift rate exceeds a threshold, this is functionally equivalent to the claimed invention of retraining when no arcs have been detected for a period of time as it continuously receives no-arc labeled data or regular current data with continuous noise that can cause concept drift.)
Zhang, Dunne, and Yamada are analogous to the claimed invention as they are from the same field of endeavor of machine learning model execution, anomaly detection, and on-device model updating/retraining on resource-constrained embedded edge devices. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date, to combine the arc fault circuit interrupter apparatus of Zhang, the on-device learning engine architecture of Dunne with the concept-drift-triggered sequential model retraining mechanism on low-end edge microcontrollers of Yamada. The motivation is as recited by Yamada (Yamada, Pg1, Abstract, Lines1-5, "practical issue of edge AI systems is that data distributions of trained dataset and deployed environment may differ due to noise and environmental changes over time. Such a phenomenon is known as a concept drift, and this gap degrades the performance of edge AI systems and may introduce system failures. To address this gap, retraining of neural network models triggered by concept drift detection is a practical approach") such that applying Yamada’s drift-triggered retraining mechanism to Zhang’s circuit interrupter results in triggering model retraining/updating during an elapsed period of normal operation without detecting an arc fault, thereby allowing the microcontroller to adapt its detection model to background noise shifts and sensor aging over time to prevent false positive decisions.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DONG YOON JUNG whose telephone number is (571)270-0198. The examiner can normally be reached 8am-5pm.
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, Cesar Paula can be reached at (571) 272-4128. 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.
/DONG YOON JUNG/ Examiner, Art Unit 2145
/CESAR B PAULA/ Supervisory Patent Examiner, Art Unit 2145
Prosecution Insights