DETAILED ACTION
This office action is in response to amendments filed on 03/20/2026.
Claim 13 has been canceled. Claims 1-3, 5, and 23-24 have been amended. Claim 26 has been added. Claims 1-5, 8-12, 14, and 17-26 are pending.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
The information disclosure statements (IDS) submitted on 01/14/2026 and 03/13/2026 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
Response to Arguments
Prior Art Rejections:
Applicant’s arguments regarding the prior art rejections (pg. 10-11) have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Applicant argues that the cited references fail to teach the amended independent claim limitations directed to “detect[ing] a pause in operation of a sorting line within the first sorting facility, wherein the pause in operation is detected based at least in part on a determined change in a conveyance speed associated with the sorting line; and at least in part in response to the detected pause, send[ing] a software update to a compute node or a sorting device located at the first sorting facility, wherein the software update includes the modified machine learning model.”
Examiner notes that the McClain and Rogers references have been brought in to teach these limitations. Claims 1, 2, 5, 8-10, and 18-26 are now rejected under 35 USC § 103 as being unpatentable over Zeng in view of Gurumurthy, McClain, and Rogers. The remaining claims are rejected based on the combination of Zeng, Gurumurthy, McClain, and Rogers, further combined with one or more other references. The prior art rejections have been updated to include the amended limitations and to clarify the reasoning given for the limitations that were not amended.
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.
Claims 1, 2, 5, 8-10, and 18-26 are rejected under 35 U.S.C. 103 as being unpatentable over
Zeng et al. (hereinafter Zeng), U.S. Patent Application Publication US 20210178432 A1 in view of
Gurumurthy et al. (hereinafter Gurumurthy), U.S. Patent Application Publication US 20230125477 A1,
McClain et al. (hereinafter McClain), U.S. Patent Application Publication US 20120037547 A1, and
Rogers et al. (hereinafter Rogers), U.S. Patent Application Publication US 20210117859 A1.
Regarding Claim 1, A system, comprising:
Zeng teaches a processor configured to: (Zeng teaches “a trash sorting and recycling system, comprising: a control device…wherein, the control device includes: a processor and a memory, the memory has a computer program suitable for running by the processor stored therein, and the computer program is run by the processor” (0018).)
Zeng teaches obtain a machine learning model associated with a domain associated with materials to be sorted at a first sorting facility, (Zeng teaches “a deep learning neural network to judge whether or not the trash to be sorted belongs to recyclable trash” (0005). Examiner notes that a deep learning neural network is a machine learning model, and trash disposal is a domain associated with materials to be sorted.)
Zeng teaches a memory coupled to the processor and configured to provide the processor with instructions. (Zeng teaches “a trash sorting and recycling system, comprising: a control device…wherein, the control device includes: a processor and a memory, the memory has a computer program suitable for running by the processor stored therein, and the computer program is run by the processor” (0018).)
Zeng does not appear to explicitly disclose
wherein the machine learning model was trained on data aggregated from a plurality of sorting facilities, wherein the data aggregated from the plurality of sorting facilities comprises a first set of images of objects captured at respective ones of the plurality of sorting facilities;
obtain a set of training data obtained from one or more compute nodes located at the first sorting facility, wherein the set of training data was generated by:
a user input to initiate a training process having been received at the one or more compute nodes;
a second set of images generated by one or more object recognition devices with respect to one or more provided objects at the first sorting facility having been received by the one or more compute nodes;
one or more user input labels corresponding to the one or more provided objects having been received via a user interface associated with the one or more compute nodes; and
associations having been made by the one or more compute nodes between the one or more input labels and the second set of images with respect to the one or more provided objects; and
generate a modified machine learning model by training the machine learning model using the set of training data obtained from the first sorting facility, wherein the modified machine learning model comprises model weights that have been updated for the first sorting facility;
However, Gurumurthy teaches wherein the machine learning model was trained on data aggregated from a plurality of sorting facilities, wherein the data aggregated from the plurality of sorting facilities comprises a first set of images of objects captured at respective ones of the plurality of sorting facilities; and (0503: “In at least one embodiment, an existing machine learning model may be selected from model registry 3624… In at least one embodiment, machine learning models in model registry 3624 may have been trained on imaging data from different facilities than facility 3602 (e.g., facilities remotely located). In at least one embodiment, machine learning models may have been trained on imaging data from one location, two locations, or any number of locations.” The machine learning model obtained from the model registry was trained on a set of image data comprising images from a plurality of facilities.)
Gurumurthy teaches obtain a set of training data obtained from one or more compute nodes located at the first sorting facility, wherein the set of training data was generated by:
a user input to initiate a training process having been received at the one or more compute nodes; (0525: “In at least one embodiment, although not illustrated with respect to training system 3604, user interface 3714 (or a different user interface) may be used for selecting models for use in deployment system 3606, for selecting models for training, or retraining, in training system 3604, and/or for otherwise interacting with training system 3604.” A user selects a model for retraining (i.e. initiates a training process).)
a second set of images generated by one or more object recognition devices with respect to one or more provided objects at the first sorting facility having been received by the one or more compute nodes; (0500: “In at least one embodiment, machine learning models may be trained at facility 3602 using data 3608 (such as imaging data) generated at facility 3602 (and stored on one or more picture archiving and communication system (PACS) servers at facility 3602)…” 0507: “In at least one embodiment, a data processing pipeline may receive input data (e.g., imaging data 3608) in a DICOM, RIS, CIS, REST compliant, RPC, raw, and/or other format in response to an inference request (e.g., a request from a user of deployment system 3606, such as a clinician, a doctor, a radiologist, etc.).” Imaging data (i.e. images generated by object recognition devices with respect to provided objects) is generated at facility 3602 (i.e. the first sorting facility) and received by the data processing pipeline.)
one or more user input labels corresponding to the one or more provided objects having been received via a user interface associated with the one or more compute nodes; and (0520: “In at least one embodiment, labels or other annotations may be generated within a drawing program (e.g., an annotation program), a computer aided design (CAD) program, a labeling program, another type of program suitable for generating annotations or labels for ground truth, and/or may be hand drawn, in some examples. In at least one embodiment, ground truth data may be synthetically produced (e.g., generated from computer models or renderings), real produced (e.g., designed and produced from real-world data), machine-automated (e.g., using feature analysis and learning to extract features from data and then generate labels), human annotated (e.g., labeler, or annotation expert, defines location of labels), and/or a combination thereof. In at least one embodiment, for each instance of imaging data 3608 (or other data type used by machine learning models), there may be corresponding ground truth data generated by training system 3604.” Ground truth data (i.e. input labels) for imaging data (i.e. corresponding to provided objects) can be produced by human annotation (i.e. user input) via a drawing program or CAD program (i.e. user interface).)
associations having been made by the one or more compute nodes between the one or more input labels and the second set of images with respect to the one or more provided objects; and (0520: “In at least one embodiment, for each instance of imaging data 3608 (or other data type used by machine learning models), there may be corresponding ground truth data generated by training system 3604.” Instances of imaging data (i.e. the second set of images) correspond to (i.e. are associated with) ground truth data (i.e. input labels).)
generate a modified machine learning model by training the machine learning model using the set of training data obtained from the first sorting facility, wherein the modified machine learning model comprises model weights that have been updated for the first sorting facility; (0500: “In at least one embodiment, machine learning models may be trained at facility 3602 using data 3608 (such as imaging data) generated at facility 3602 (and stored on one or more picture archiving and communication system (PACS) servers at facility 3602)…” 0555: “In at least one embodiment, model training 3614 may include retraining or updating an initial model 4004 (e.g., a pre-trained model) using new training data…” 0130: “…in at least one embodiment, a machine learning model may be trained by calculating weight parameters according to a neural network architecture…” The initial model at facility 3602 (i.e. the first sorting facility) is updated/retrained (i.e. modified) using the set of imaging data (i.e. training data) generated at facility 3602. Training comprises updating model weights.)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zeng and Gurumurthy. Zeng teaches a trash sorting method using a deep neural network to identify recyclable trash. Gurumurthy teaches improvements to automated optical inspection (AOI) using neural networks. One of ordinary skill would have motivation to combine Zeng and Gurumurthy because Gurumurthy’s AOI system “can be used to inspect objects of different materials (e.g., paper, plastic, metal) in a recycling plant for efficient sorting” (0072). According to Gurumurthy, “cost, processing speed, portability, ease in training tasks, flexibility with number of cameras used, flexibility with camera angles, and flexibility with multiple lighting conditions can be improved” (0002).
Zeng and Gurumurthy do not appear to explicitly disclose detect a pause in operation of a sorting line within the first sorting facility, wherein the pause in operation is detected based at least in part on a determined change in a conveyance speed associated with the sorting line;
However, McClain teaches detect a pause in operation of a sorting line within the first sorting facility, wherein the pause in operation is detected based at least in part on a determined change in a conveyance speed associated with the sorting line; (0085: “That is, controller 312 is interfaced with sensors 304, 306, and 308 and may receive and record data associated with detected downtime conditions and/or detected actual conditions. For example, if any conveyor motor 310 is not moving and/or there is no sorting material present on in-feed conveyor 102, as determined based on data from sensors 304, 306, and/or 308, controller 312 automatically detects a downtime condition for sorting system 100.” A downtime condition (i.e. pause in operation) of a sorting system is detected when a conveyor motor is not moving (i.e. based on a determined change in conveyance speed).)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zeng, Gurumurthy, and McClain. Zeng teaches a trash sorting method using a deep neural network to identify recyclable trash. Gurumurthy teaches improvements to automated optical inspection (AOI) using neural networks. McClain teaches detecting and recording downtime experienced by a sorting system for recyclable materials. One of ordinary skill would have motivation to combine Zeng, Gurumurthy, and McClain because “[t]he systems and methods described herein provide a world-class operations performance module for sorting systems to improve efficiency and reduce downtime of sorting systems. More specifically, the systems described herein establish baselines to drive performance goals, trend system performance and throughput, and provide tools to analyze causal effects, quantify potential improvements, and/or direct focus on highest areas of return. By performing the above-described methods, an operator can determine which sorting systems experience the longest downtimes and/or which shifts, operators, and/or maintenance personnel experience the most downtime” (McClain, 0187).
Zeng, Gurumurthy, and McClain do not appear to explicitly disclose at least in part in response to the detected pause, send a software update to a compute node or a sorting device located at the first sorting facility, wherein the software update includes the modified machine learning model;
However, Rogers teaches at least in part in response to the detected pause, send a software update to a compute node or a sorting device located at the first sorting facility, wherein the software update includes the modified machine learning model; (0022: “[T]o update to a newer, better model, a new container image, which then becomes a new version of the application, must be built and distributed to the computing nodes at the edge locations. The deployment of a new version of the application means that the container currently running at the node must be stopped and a new container started, which results in downtime of inferencing and analytics.” During inferencing downtime (i.e. a pause in operation), a new model version (i.e. modified machine learning model) is deployed to a computing node at an edge location (i.e. a compute node at the first sorting facility).)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zeng, Gurumurthy, McClain, and Rogers. Zeng teaches a trash sorting method using a deep neural network to identify recyclable trash. Gurumurthy teaches improvements to automated optical inspection (AOI) using neural networks. McClain teaches detecting and recording downtime experienced by a sorting system for recyclable materials. Rogers teaches that in the field of machine learning-based video stream inference, deploying an updated model to an edge compute node requires inferencing downtime. One of ordinary skill would have motivation to combine Zeng, Gurumurthy, McClain, and Rogers because McClain teaches detecting downtime, Rogers teaches that downtime is required for model deployment, and one of ordinary skill in the art would have recognized that updated model deployment, as disclosed by Rogers, could take advantage of pre-existing inferencing downtime, as detected in McClain, in order to avoid creating additional downtime, which leads to “frequent stoppages, as well as inferencing or analytics gaps” (Rogers, 0022).
Regarding Claim 2, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Zeng also teaches wherein the processor is remote to the first sorting facility, and (Zeng teaches “the control device is a remote control device for the trash sorting device, and the trash sorting device and the control device communicate with each other through a wired network or a wireless network” (0026). The control device includes the processor, and it is remote to the sorting device, i.e. the sorting facility.)
Zeng teaches wherein the processor is further configured to:
obtain, over a first network, a sensed signal associated with a target object located at the first sorting facility; (Zeng teaches “acquiring a detection image of trash to be sorted” (0005). The detection image is a sensed signal, and the target object is recyclable trash.)
apply the modified machine learning model to the sensed signal to identify the target object; and (Zeng teaches “processing the detection image with a deep learning neural network to judge whether or not the trash to be sorted belongs to recyclable trash” (0005).)
send, over a second network, a control signal from the processor to the sorting device located at the first sorting facility to cause the sorting device at the first sorting facility to perform a capture operation on the target object. (Zeng teaches “sending a first control signal, to control to deliver the trash to be sorted into a recycling region” (0005).)
Regarding Claim 5, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Gurumurthy also teaches wherein the processor comprises a first processor, and wherein a second processor of a second facility compute node located in a second sorting facility is further configured to: (Gurumurthy teaches a “distributed computer system comprising multiple devices” (0627), where devices of the computer system include “any number and type of processor(s)” (0255) distributed among “facility 3602” and “different facilities than facility 3602” (0503).)
Gurumurthy teaches receive, over a first network, the modified machine learning model; (Gurumurthy teaches a “model registry” which “may be accessible through, for example, a cloud storage” (0501). “In at least one embodiment, an existing machine learning model may be selected from model registry” which “may have been trained on imaging data from different facilities than facility 3602” (0503). I.e., a second facility can receive, through the model registry network, a model that was trained/modified at a first facility.)
Gurumurthy teaches deploying the model at the second sorting facility. (0503: “In at least one embodiment, a machine learning model may then be selected from model registry 3624—and referred to as output model 3616—and may be used in deployment system 3606 to perform one or more processing tasks for one or more applications of a deployment system.”)
Zeng teaches obtain, over a second network, a sensed signal associated with a target object [located at the second sorting facility]; (Zeng teaches “acquiring a detection image of trash to be sorted” (0005). The detection image is a sensed signal, and the target object is recyclable trash.)
Zeng teaches apply the modified machine learning model to the sensed signal to identify the target object; and (Zeng teaches “processing the detection image with a deep learning neural network to judge whether or not the trash to be sorted belongs to recyclable trash” (0005).)
Zeng teaches send, over a third network, a control signal from the second processor to a [second facility] sorting device [located at the second sorting facility] to cause the sorting device [at the second sorting facility] to perform a sorting operation on the target object. (Zeng teaches “sending a first control signal, to control to deliver the trash to be sorted into a recycling region” (0005).)
Regarding Claim 8, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Gurumurthy also teaches wherein the data aggregated from the plurality of sorting facilities further comprises metadata. (Gurumurthy teaches that the training framework can perform supervised training, where “supervision comprises input information such as tags or labels, where a tag or label identifies that an input training data 104 item contains a specific object or objects or is of a specific classification” (0061). Examiner notes that these tags/labels are metadata.)
Regarding Claim 9, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Gurumurthy also teaches wherein the data obtained from the first sorting facility comprises a panorama with respect to a plurality of objects at the first sorting facility, wherein the panorama comprises a combination of a plurality of image frames of the plurality of objects. (Gurumurthy teaches “two or more different images of one or more manufactured objects (e.g., PCBs) may be captured… one or more combined images are generated based, at least in part, on multiple images” (0070). These combined image(s) make up the input data captured by the system which is used to train/re-train the neural network.)
Regarding Claim 10, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Gurumurthy also teaches wherein the machine learning model associated with the domain comprises a first machine learning model associated with a first domain, and wherein the processor is further configured to generate a second machine learning model associated with a second domain by using the first machine learning model in pretraining. (Gurumurthy teaches “a user may not have a model for use, so a user may select a pre-trained model 3706 to use with an application… pre-trained model 3706 may not be optimized for generating accurate results on customer dataset 4006 of a facility of a user (e.g., based on patient diversity, demographics, types of medical imaging devices used, etc.)… prior to deploying pre-trained model 3706 into deployment pipeline 3710 for use with an application(s), pre-trained model 3706 may be updated, retrained, and/or fine-tuned for use at a respective facility” (0557). Examiner notes that the differences between data sets in demographics, etc. represent different domains, and updating the pre-trained model corresponds to generating a second model using the first model in pretraining. Gurumurthy also specifies that “refined model 4012 may be further refined on new datasets any number of times” (0562).)
Regarding Claim 18, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Gurumurthy also teaches wherein the modified machine learning model comprises a first modified machine learning model and wherein the processor is further configured to generate a second modified machine learning model associated with the first sorting facility including by: (Gurumurthy teaches “retraining or updating a machine learning model” to generate a refined model (0504). Gurumurthy also specifies that “refined model 4012 may be further refined on new datasets any number of times” (0562).)
Gurumurthy teaches determining a third set of images associated with objects that are not identified by the first modified machine learning model at the first sorting facility; (Gurumurthy teaches that the model “might not be fine-tuned or optimized… because of differences in populations, genetic variations, robustness of training data used to train a machine learning model, diversity in anomalies of training data, and/or other issues with training data” (0504). Issues with robustness of training data indicates a lack of representation of certain classifications, necessitating the identification of new “imaging data 3608 to be used as ground truth data” (0504).)
Gurumurthy teaches receiving annotations corresponding to the third set of images; and (Gurumurthy teaches “AI-assisted annotation 3610 may be used to aid in generating annotations corresponding to imaging data 3608 to be used as ground truth data… annotations provided by a clinician, doctor, scientist, etc.” may also be used (0504).)
Gurumurthy teaches generating the second modified machine learning model by training the first modified machine learning model using the annotations. (Gurumurthy teaches “AI-assisted annotations 3610, labeled clinic data 3612, or a combination thereof—may be used as ground truth data for retraining or updating” the model (0504).)
Regarding Claim 19, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Gurumurthy also teaches wherein the machine learning model comprises a parent machine learning model and wherein the modified machine learning model comprises a child machine learning model. (Gurumurthy teaches “once a model is trained—or partially trained,” it may be “added to model registry” and “may then be retrained, or updated, at any number of other facilities” (0503). This creates a structure of descendants, where the partially trained model is the original parent model, and each retrained/modified model at each facility is a child model of the parent model.)
Regarding Claim 20, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Gurumurthy also teaches wherein to generate the modified machine learning model comprises to add a new output layer corresponding to the machine learning model. (Gurumurthy teaches “to retrain, or update, initial model 4004, output or loss layer(s) of initial model 4004 may be reset, or deleted, and/or replaced with an updated or new output or loss layer(s)” (0555).)
Regarding Claim 21, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Gurumurthy also teaches wherein to generate the modified machine learning model comprises to train the machine learning model using the data obtained from the first sorting facility in addition to data obtained from a second sorting facility, wherein the first sorting facility and the second sorting facility share a common attribute, and (Gurumurthy teaches “machine learning models may be trained at facility 3602 using data 3608 (such as imaging data) generated at facility 3602…, may be trained using imaging or sequencing data 3608 from another facility or facilities (e.g., a different hospital, lab, clinic, etc.), or a combination thereof” (0500). Examiner notes that the phrase “a different hospital, lab, clinic, etc.” implies that both facilities are of the same type and process similar types of data (i.e., they share a common attribute).)
Gurumurthy teaches wherein the processor is further configured to: provide the modified machine learning model to the first sorting facility or the second sorting facility. (Gurumurthy teaches that a trained or retrained model “may be added to model registry” (0503). When “facility 3602 needs a machine learning model for use in performing one or more processing tasks…but facility 3602 may not currently have such a machine learning model…an existing machine learning model may be selected from model registry” which “may have been trained on imaging data from one location, two locations, or any number of locations” including facility 3602 (0503).)
Regarding Claim 22, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Zeng also teaches wherein to generate the modified machine learning model includes to train the machine learning model using the set of training data obtained from the first sorting facility in addition to noisy data. (Zeng teaches that prior to training the model, “the training image may be subjected to several times of convoluting and sampling… the sampling process can process the training image by a method such as average combining, maximum combining, and random combining… if the weight value is relatively large, according to a magnitude of the offset, the sampling process may be equivalent to a “OR” operation with noisy or a “AND” operation with noisy” (0067).)
Claim 23 is a method claim, containing substantially the same elements as system claim 1. Zeng, Gurumurthy, McClain, and Rogers teach the elements of claim 1, as shown above.
Claim 24 is a product claim, containing substantially the same elements as system claim 1. Zeng, Gurumurthy, McClain, and Rogers teach the elements of claim 1, as shown above.
Zeng also teaches A computer program product, the computer program product being embodied in a non- transitory computer-readable storage medium and comprising computer instructions (Zeng teaches “the computer program product may include various forms of computer readable storage media, for example, a volatile memory and/or a nonvolatile memory… One or more computer programs may be stored on the computer readable storage medium” (0134).)
Regarding Claim 25, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Gurumurthy also teaches wherein the second set of images captured at the first sorting facility is different from the first set of images. (0503: “In at least one embodiment, an existing machine learning model may be selected from model registry 3624… In at least one embodiment, machine learning models in model registry 3624 may have been trained on imaging data from different facilities than facility 3602 (e.g., facilities remotely located).” 0555: “In at least one embodiment, model training 3614 may include retraining or updating an initial model 4004 (e.g., a pre-trained model) using new training data (e.g., new input data, such as customer dataset 4006, and/or new ground truth data associated with input data).” The model obtained from the registry was trained on imaging data from different facilities than the current facility (i.e. the first set of images). The model is then retrained (i.e. modified) at the current facility using new input data (i.e. the second set of images).)
Regarding Claim 26, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
McClain also teaches wherein the conveyance speed associated with the sorting line comprises a global speed of a conveyor device within the sorting line, wherein the global speed of the conveyor device is determined based at least in part on respective trajectories of objects being transported by the conveyor device. (0075: “[C]ontrol system 278 controls a speed of conveyors… based on the sensor inputs, control system 278 automatically stops a component and/or sorting system 100, for example, without limitation, when a mechanical, electrical, structural, jam, overload, and/or no material situation occurs within sorting system 100.” The control system controls conveyor speed, and can stop the sorting system (i.e. determine the global speed of the conveyor device) based on, e.g., a jam or overload (i.e. based on trajectories of objects being transported).)
Claims 3-4 are rejected under 35 U.S.C. 103 as being unpatentable over Zeng in view of Gurumurthy, McClain, and Rogers, and further in view of
Cheruvu et al. (hereinafter Cheruvu), U.S. Patent Application Publication US 20210390447 A1.
Regarding Claim 3, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Zeng also teaches wherein the processor is remote to the first sorting facility, and (Zeng teaches “the control device is a remote control device for the trash sorting device, and the trash sorting device and the control device communicate with each other through a wired network or a wireless network” (0026). The control device includes the processor, and it is remote to the sorting device, i.e. the sorting facility.)
Zeng, Gurumurthy, McClain, and Rogers do not appear to explicitly disclose wherein the processor is further configured to cryptographically sign the modified machine learning model prior to providing the modified machine learning model to the compute node located at the first sorting facility.
However, Cheruvu teaches wherein the processor is further configured to cryptographically sign the modified machine learning model prior to providing the modified machine learning model to the compute node located at the first sorting facility. (Cheruvu teaches a system for securely transmitting machine learning models, in which “the processor may identify a global unique identifier (GUID) for the ML model… the processing may generate a digital signature for the content, the digital signature based on at least the GUID of the ML model…the processor may transmit the content and the digital signature to a content consumer platform” (0068-0069).)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zeng, Gurumurthy, McClain, Rogers, and Cheruvu. Zeng teaches a trash sorting method using a deep neural network to identify recyclable trash. Gurumurthy teaches improvements to automated optical inspection (AOI) using neural networks. McClain teaches detecting and recording downtime experienced by a sorting system for recyclable materials. Rogers teaches that in the field of machine learning-based video stream inference, deploying an updated model to an edge compute node requires inferencing downtime. Cheruvu teaches a system for authenticating transmitted ML models using digital signatures. One of ordinary skill would have motivation to combine Zeng, Gurumurthy, McClain, Rogers, and Cheruvu in order to “provide protection of AI and ML models, and the output data generated from such models, in addition to supporting an end-to-end ML security pipeline by helping secure the authenticity and validity of output data” (0021).
Regarding Claim 4, Zeng, Gurumurthy, McClain, Rogers, and Cheruvu teach The system of claim 3, as shown above.
Cheruvu also teaches wherein the compute node is configured to decrypt the cryptographically signed modified machine learning model. (Cheruvu teaches “the digital signature is utilized to verify authenticity of the ML model generating the content at the content consumer platform” (0093).)
Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Zeng in view of Gurumurthy, McClain, and Rogers, and further in view of
Jannink et al. (hereinafter Jannink), U.S. Patent Application Publication US 20200184273 A1.
Regarding Claim 11, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Zeng, Gurumurthy, McClain, and Rogers do not appear to explicitly disclose wherein the processor is further configured to compare at least two machine learning models run against one or more training data sets.
However, Jannink teaches wherein the processor is further configured to compare at least two machine learning models run against one or more training data sets. (Jannink teaches “testing may compare the output of the new model against a previously validated ‘golden data set’ output of the existing models” (0042). Examiner notes that a “golden data set” is a training data set.)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zeng, Gurumurthy, McClain, Rogers, and Jannink. Zeng teaches a trash sorting method using sensors and a deep neural network to identify recyclable trash. Gurumurthy teaches improvements to automated optical inspection (AOI) using neural networks. McClain teaches detecting and recording downtime experienced by a sorting system for recyclable materials. Rogers teaches that in the field of machine learning-based video stream inference, deploying an updated model to an edge compute node requires inferencing downtime. Jannink teaches developing and optimizing a model of an environment using sensor data. One of ordinary skill would have motivation to combine Zeng, Gurumurthy, McClain, Rogers, and Jannink because Jannink’s “refinable model” is “amenable to iterative updates that improve its accuracy” (0030).
Claims 12 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Zeng in view of Gurumurthy, McClain, and Rogers, and in further view of
Parr et al. (hereinafter Parr), U.S. Patent Application Publication US 20190217342 A1.
Regarding Claim 12, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Gurumurthy also teaches communication between the plurality of sorting facilities (Gurumurthy teaches “communication between facilities and components of system 3700 (e.g., for transmitting inference requests, for receiving results of inference requests, etc.) may be communicated over a data bus or data busses, wireless data protocols (Wi-Fi), wired data protocols (e.g., Ethernet), etc.” (0517).)
Zeng, Gurumurthy, McClain, and Rogers do not appear to explicitly disclose wherein the processor is further configured to collect operational data associated with [the plurality of] sorting facilities and generate one or more reports based on the operational data.
However, Parr teaches wherein the processor is further configured to collect operational data associated with [the plurality of] sorting facilities and generate one or more reports based on the operational data. (Parr teaches “facility environmental sensors… can report number, weight and type of bales or waste material (recyclable or otherwise) produced, moisture sensing instrumentation (to determine if materials are wet and so need to be discarded or rerouted for additional processing), inclinometers on screens, sorters, feeders and conveyors, induction sensing arrays (which may help detect metal contaminants or types of metal for appropriate recycling), laser measurement devices that report volumetric characteristics of the material stream, smart current sensing meters, for detection of overloads or frequency drives that report running amperage of system equipment, positive and negative pressure transducers to compute system vacuum and pressure required to remove objects in positive and negative sorting applications, flow switches and meters to report total air consumed by optical and robotic sorters, and fire detection and identification sensors and cameras, to name a few” (0069). Additionally, “If an adverse condition is detected…control system 302 may notify an operator of the MRF of the adverse condition to dispatch manual correction” (0072).)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zeng, Gurumurthy, McClain, Rogers, and Parr. Zeng teaches a trash sorting method using a deep neural network to identify recyclable trash. Gurumurthy teaches improvements to automated optical inspection (AOI) using neural networks. McClain teaches detecting and recording downtime experienced by a sorting system for recyclable materials. Rogers teaches that in the field of machine learning-based video stream inference, deploying an updated model to an edge compute node requires inferencing downtime. Parr teaches an autonomous system for control of material recovery and recycling sorting. One of ordinary skill would have motivation to combine Zeng, Gurumurthy, McClain, Rogers, and Parr because, according to Parr, “human staffed positions can be minimized and the plant can be dynamically configured to accommodate waste streams of a fluctuating nature and composition, thereby allowing the plant to be operated more efficiently” (0025) while “limiting the amount of contaminants found in the final recovered commodity, maximizing the amount of commodity that can be recovered, and minimizing the amount of material that is sent to a landfill” (0019).
Regarding Claim 14, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Zeng, Gurumurthy, McClain, and Rogers do not appear to explicitly disclose wherein the processor is further configured to:
obtain commodity values associated with a plurality of material types; and
use the commodity values to assign priorities to target objects to perform sorting operations on at the first sorting facility based on the commodity values.
However, Parr teaches wherein the processor is further configured to:
obtain commodity values associated with a plurality of material types; and (Parr teaches using a “central control system 302 to research commodity processes and pricing, such as via an external information source like the Internet” (0078).)
use the commodity values to assign priorities to target objects to perform sorting operations on at the first sorting facility based on the commodity values. (Parr teaches using the commodity pricing information “to adjust the system to recover the highest possible value stream” (0078) by identifying “targeted commodities” (0023).)
Claim 17 is rejected under 35 U.S.C. 103 as being unpatentable over Zeng in view of Gurumurthy, McClain, and Rogers, and in further view of
Stauffenberg, U.S. Patent Application Publication US 20220005002 A1.
Regarding Claim 17, Zeng, Gurumurthy, McClain, and Rogers teach The system of claim 1, as shown above.
Zeng, Gurumurthy, McClain, and Rogers do not appear to explicitly disclose wherein the one or more user input labels comprise one or more stock keeping units (SKUs).
However, Stauffenberg teaches wherein the one or more user input labels comprise one or more stock keeping units (SKUs). (Stauffenberg teaches a system for sorting recycling materials in which “software assisted item recognition may be based on traditional programmed algorithms and/or machine learning algorithms” (0065). Machine learning models are trained “using tagged images of material items” (0232), where the tags identify an item, and “the ‘identity’ of an item refers to the identification of a unique item based on an identifier such as a tag or other features that are specific (‘unique item’), or pseudo-specific to said individual item… for example its Stock Keeping Unit SKU” (0036).)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zeng, Gurumurthy, McClain, Rogers, and Stauffenberg. Zeng teaches a trash sorting method using a deep neural network to identify recyclable trash. Gurumurthy teaches improvements to automated optical inspection (AOI) using neural networks. McClain teaches detecting and recording downtime experienced by a sorting system for recyclable materials. Rogers teaches that in the field of machine learning-based video stream inference, deploying an updated model to an edge compute node requires inferencing downtime. Stauffenberg teaches a software-assisted system for sorting recycling materials while capturing information about them. One of ordinary skill would have motivation to combine Zeng, Gurumurthy, McClain, Rogers, and Stauffenberg because Stauffenberg’s use of Stock Keeping Units (SKUs) “allows the identification and documentation of materials collected, enabling easy manual sorting, and tracking their progress” (0025) which in turn “facilitates cost-effective and efficacious recycling, as well as creating dependable information” (0019).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BENJAMIN M ROHD whose telephone number is (571)272-6445. The examiner can normally be reached Mon-Thurs 8:00-6:00 EST.
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, Viker Lamardo can be reached at (571) 270-5871. 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.
/B.M.R./Examiner, Art Unit 2147 /VIKER A LAMARDO/Supervisory Patent Examiner, Art Unit 2147
Prosecution Insights