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 .
Examiner Notes
Examiner cites particular columns and line numbers in the references as applied to the claims below for convince of the applicant. Although the specified citations are representative of the teachings in the art and are applied to the specific limitations within the individual claim, other passages and figures may apply as well. It is respectfully requested that, in preparing responses, the applicant fully consider the references cited in their entirety as potentially teaching all or part of the claimed invention, as well as the context of the passage as taught by the prior art or disclosed by the examiner.
Claim Rejections
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.
Claims 11-15 are rejected, as claim 11 recites: “transmitting splitting execution information…to the service server and the network management server”. This is the first mention of “network management server” in the claim yet is introduced with the definite article “the” instead of “a”, lacking proper antecedent basis. Claims 12-15 depend on claim11 and therefore inherit the deficiency.
Claim 16 is rejected as indefinite, as the limitation: “updating the communication session to a communication quality class for smoothly supporting the changed splitting execution type” is unclear. A communication session cannot logically be “updated to” a quality class; it is ambiguous what structural or functional change this requires. Claims 17-20 depend on claim 16 and therefore inherit the deficiency. Furthermore, claims 16-20 are rejected, as claim 16 recites “…by using the network management server” and “…by using a data transfer device” without prior introduction of the “the network management server”, resulting in the antecedent basis being unclear. Claims 17-20 depend on claim11 and therefore inherit the deficiency.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1-6 are rejected under 35 U.S.C. 102 as being anticipated by Karjee (US 2022/0311678 A1) and Lee (US 2023/0386192 A1).
Regarding claim 1, Karjee discloses “A method of supporting dynamic splitting execution of a machine learning model in a communication system including an edge device and a service server connected with each other over a communication network” by disclosing a “[a] method for execution of a deep neural network (DNN) in an internet if things (IoT)-edge network” (claim1 ), in which, “at least one edge device [is selected] from a plurality of edge devices within communication range of the IoT device (claim 1), establishing the claimed communication system of an edge device and a service server connected over an network.
Karjee also discloses, “transmitting device information to the service server by using the edge device” by disclosing that the split ratio is determined by using “a predetermined computation time of each layer of the DNN on the selected at least one edge device obtained by benchmarking the DNN on the selected at least one edge device” (claim 7), and that network throughput is “determined based on a response time of a message sent from the IoT device to the selected at least one edge device” (claim 7). This benchmarked computation-time data and responsive timing data, information describing the edge device’s own processing/communication characteristics, is made available to the split-ratio-determining device, reading on “device information” transmitted to the service server.
Karjee also discloses, “determining a splitting execution type of the machine learning model and transmitting splitting execution information including the determined splitting execution type to the edge device by using the service server, based on the device information” by expressly disclosing: “determining, by the IoT device, a split ratio based on at least one of an inference time of the DNN and a transmission time required for transmitting output of each layer of the DNN from the IoT device to the selected at least one edge device” and ” splitting, by the IoT device, a plurality of layers of the DNN into a first part and a second part based on the determined split ratio, and transmitting the second part to the selected at least one edge device through the identified network, wherein the first part is executed on the IoT device and the second part is executed on the selected at least one edge device.” (claim 1). The transmitted “second part” necessarily conveys the determined splitting execution type (i.e., which layers belong to the edge device), satisfying this limitation.
Karjee also discloses, “performing an arithmetic operation on the machine learning model by using at least one of the edge device and the service server, based on the determined splitting execution type”, by disclosing “a computation task on the first part is executed on the IoT device and computation tasks of the second part is executed on the selected at least one edge device (claim1), reading directly on this limitation.
Lee supports Karjee by disclosing a “loud server configured to infer an acquired image along with an edge device in a split manner” (claim 1), wherein the processor is “configured to analyze at least one of a network resource between the edge device and the cloud server, a computing resource of the edge device, and a computing resource of the cloud server”, and determine a location of the split point…based on an analysis result” (claim4). Lee further discloses: “the edge device 100 may transmit information on the order of the first layer to the cloud server 200 such that the cloud server 200 can determine a layer to which the quantized feature will be input” [0062], an express disclosure of device information (layer/order information) being transmitted to the server to enable the server-side determination, and the resulting quantized feature (embodying the splitting execution type) being sent onward for the arithmetic operation.
Regarding claim 2, it is rejected for the same reasons mentioned in claim 1 due to its dependency. Karjee also teaches wherein the splitting execution information comprises information representing one of a type where the edge device performs all arithmetic operations on the machine learning model, a type where the service server performs all arithmetic operations on the machine learning model, and a type where the edge device and the service server split and perform all arithmetic operations on the machine learning model, by disclosing [in table 4] 3 types of benchmarking results. Firstly, an all-edge device type: “Less than 6 Mbps -> complete on-device execution -> 181 ms”. Secondly, an all-service server type: “Greater than 11 Mbps -> Complete on-edge execution -> Less than 145 ms”. Thirdly, a split type: “Greater than 6 Mbps and less than 11 Mbps -> Layers 1-11 -> on-device and Layers 12-31 -> on-edge”. This is confirmed again in Table 6: “TH <= 4 Mbps -> Complete on-device execution”; “TH > 31 Mbps -> Complete on-edge execution”; and intermediate throughputs yielding “Layer 1-12 (on-device) and Layer 13-31 (on-edge)”. All three claimed types are disclosed.
Regarding claim 3, it is rejected for the same reasons mentioned in claim 1 due to its dependency. Karjee also teaches, wherein the splitting execution information comprises layer information representing layers allocated to each of the edge device among a plurality of layers included in the machine learning model and layer information representing layers allocated to the service server among the plurality of layers by stating “the MobileNet DNN layers 1-11 are executed on-device and the layers 12-31 on-edge” [0120], hence explicitly allocating specific layer ranges to each device. Lee also teaches this via its split-point architecture: the first artificial intelligence model executes “a first layer corresponding to a predetermined split point,” [0085] while the cloud server executes” a second layer immediately after the first layer” (claim 1), i.e. explicit layer allocation to each device.
Regarding claim 4, it is rejected for the same reasons mentioned in claim 1 due to its dependency. Additionally, Karjee teaches wherein the performing of the arithmetic operation on the machine learning model comprises performing an arithmetic operation on all layers included in the machine learning model, based on the determined splitting execution type by having the DSC framework perform, in the complete-on-device or complete-on-edge cases, inference across “L1, L2, L3, …LN”; all layers of the model (Algorithm 1 (0125)), satisfying this general limitation regardless of which single device executes them. Lee also discloses, “The function of the edge device 100 can be designed in various ways. For example, the edge device 100 may be designed to process data by itself without sending the data to the cloud server 200 [0046], showing an all-layer, single device embodiment.
Regarding claim 5, it is rejected for the same reasons mentioned in claim 1 due to its dependency. Additionally, Karjee teaches performing of the arithmetic operation on the machine learning model comprises performing an arithmetic operation on all layers included in the machine learning model by using the service server, based on the determined splitting execution type by teaching “Greater than 11 Mbps -> Complete on-edge” (Table 4), i.e. the “:edge device” 103 in this case is not the recipient of the arithmetic operation and the operation is instead concentrated so that the mapped “service server” role performs the entire computation, and conversely under the mapping where the IoT device is the “service server”, the “Complete on-device execution” entries in tables 4-6 show the IoT device (service server) executing all layers.
Regarding claim 6, it is rejected for the same reasons mentioned in claim 1 due to its dependency. Additionally, Karjee teaches performing of the arithmetic operation on the machine learning model comprises performing an arithmetic operation on some of all layers included in the machine learning model by using the edge device and performing an arithmetic operation on the other layers by using the service server by disclosing “Greater than 3 Mbps -> Layers 1-28 -> on-device ->Less than 584 ms” and “and less than 71 Mbps -> Layers 29-31 -> on-edge -> greater than 560 ms”. (Table 5), directly disclosing execution of some layers by one device and the remaining layers by the other device.
Claims 7-10 are rejected under 35 U.S.C. 102 as being anticipated by Karjee (US 2022/0311678 A1)
Regarding Claim 7, changing a splitting execution type of the machine learning model previously determined based on a resource status change of the edge device by using the edge device is taught by Karjee expressly disclosing, “the determined split ratio is modified based on variations in the one or more inference parameters” (claim 9), where the inference parameters include “throughput of the identified network is re-computed periodically” (claim 8); highlighting a direct disclose of dynamically changing a previously determined splitting type based on a resource status change. transmitting splitting execution change information including the changed splitting execution type to the service server by using the edge device is shown when Karjee discloses that the RCNS mechanism continuously “checks the bandwidth statistics of another available network” ([0103]) and upon detecting a higher-bandwidth network, “ The network with higher bandwidth measurements value is selected to the next network. During the selected network, the IoT device applies rule-based policy to map with the bandwidth measurements value (BWi) w.r.t the split computing point.” ([0103]) Here, the updated network/split information is used to re-transmit an updated portion of the DNN to the edge device, i.e., communicating the change. changing an operation scheme of the machine learning model by using at least one of the edge device and the service server, based on the changed splitting execution type is shown by Karjee ‘s table 4 showing the split point itself shifting (from “Complete on-device” to “Layers 1-11 -> On-device and Layers 12-31 -> on-edge” to “Complete on-edge”) as a direct function of throughput changes, evidencing the changed operation scheme.
Regarding claim 8, it is rejected for the same reasons mentioned in claim 7 due to its dependency. Additionally, wherein the changing of the operation scheme of the machine learning model comprises changing a scheme, which performs an arithmetic operation on all layers included in the machine learning model by using the edge device, to a scheme which performs an arithmetic operation on all layers included in the machine learning model by using the service server with table 4 showing “Less than 6 Mbps -> Complete on-device execution” transition to “Greater than 11 Mbps -> Complete on-edge execution” as through[put increases; a direct disclosure of switching from all-one-device scheme to an all-other devoice scheme.
Regarding claim 9, it is rejected for the same reasons mentioned in claim 7 due to its dependency. Additionally, changing a scheme, which performs an arithmetic operation on all layers included in the machine learning model by using the service server, to a scheme which performs an arithmetic operation on all layers included in the machine learning model by using the edge device is shown in Table 4, which supports the reverse transition, as throughput decreases from high to low, he scheme reverts from complete on-edge to complete on-device execution, which is the mirror image of the same disclosed mechanism and therefore equally anticipated.
Regarding claim 10, it is rejected for the same reasons mentioned in claim 7 due to its dependency. Additionally, changing the number of layers allocated to the edge device and the number of layers allocated to the service server among all layers included in the machine learning mode, is shown in table 6 where the layer boundary shifting from “Layers 1-28 (on-device) and Layer 29-31 (on-edge)” to “Layer 1-12 (on-devoice) and Layer 13-31(on edge)” as throughput increases; showing an explicit change in the number of layers allocated to each device.
Claims 16-20 are rejected under 35 U.S.C. 102 as being anticipated by Lee (US 2018/0279180 A1)
Regarding claim 16, Lee teaches configuring a communication session for performing data communication by using the edge device and the service server. By disclosing UE session establishment through the 5EG CN connecting to both central and edge clouds: “UE receives a service from the edge cloud and the central cloud through a 5G CN” ([0162])
changing a splitting execution type of the machine learning model by using one of the edge device and the service server, based on a resource status thereof is disclosed when “ An SMF device determines that there is a need to connect a new local gateway, UPF(PSA2), for a corresponding PDU session, based on a load of a central cloud and edge cloud 1, or on a lifecycle of UPF(PSA1).”, hence showing a change in processing/session allocation triggered by resource status of the edge or central side device.
sharing the changed splitting execution type over the communication network by using the edge device and the service server is disclosed when “The SMF device transmits an N4 session establishment request to PSA2 and provides PSA2 with a tunnel identifier (ID) of a branching point or an UL CL to be installed, and packet detection, enforcement, and reporting rules” ([143]), showing the changed configuration is shared with the relevant network devices over the network.
performing data communication based on the changed splitting execution type to perform an arithmetic operation on the machine learning model by using the edge device and the service server is disclosed by that following relocation “a service that is provided once through UPF(PSA1) is provided through UPF(PSA2) as UPF(PSA1) is replaced with UPF(PSA2” ([0170]). Showing data communication for service execution occurs per the changed configuration.
monitoring data traffic based on the data communication to determine whether to change the splitting execution type, by using the network management server is disclosed when An SMF device determines that there is a need to connect a new local gateway, UPF(PSA2), for a corresponding PDU session, based on a load of a central cloud and edge cloud 1” ([0172]), showing the SMF’s monitoring of load determining the need for change.
and in a case where the network management server determines whether to change the splitting execution type, updating the communication session to a communication quality class for smoothly supporting the changed splitting execution type by using a data transfer device, based on a result of the determination of the network management server is disclosed by “The SMF device requests an UL CL to set a rule to select traffic to be transmitted to the newly set UPE(PSA2) for edge cloud 2 or 1. The rule may be set based on an index of a previous rule or a tunnel ID of a next hope, or a single value of a port number or a combination of each value.” ([0174]). Here the UL CL updates/reconfigured the session’s traffic-routing rule in response to and based on the SMF’s determination, reading on updating the session to smoothly support the changed splitting arrangement.
Regarding claim 17, it is rejected for the same reasons mentioned in claim 16 due to its dependency. Additionally, wherein the determining whether to change the splitting execution type comprises monitoring a level change of the data traffic and a destination change of the data traffic to determine whether to change the splitting execution type by discloses monitoring “based on a load of a central cloud and edge cloud 1, or on a lifecycle of UPF(PSA1)” [(0172)] in combination with tracking which gateway is currently handling traffic, as evidenced by the SMF’s subsequent rule-setting to “ select traffic to be transmitted to the newly set UPE(PSA2)” [(0174)]; hence showing a destination change determination paired with the load determination.
Regarding claim 18, it is rejected for the same reasons mentioned in claim 16 due to its dependency. Additionally, wherein the determining whether to change the splitting execution type comprises: calculating a movement average value of the data traffic during a certain period; and determining whether to change the splitting execution type, based on a result of comparison of the movement average value and a reference value is disclosed by Lee’s load-based determination, “based on a load of a central cloud and edge cloud 1, or on a lifecycle of UPF(PSA1)” ([0172]), which inherently requires evaluating traffic/load over an operative period against an operator-configured threshold, since the specification confirms rule updates are governed by “an operator configuration and SMF logic” [(0082]) for organizing the data path; this periodic, policy driven load evaluation reads on calculating an average traffic value over a period and comparing it to a reference value.
Regarding claim 19, it is rejected for the same reasons mentioned in claim 16 due to its dependency. Additionally, wherein the reference value is a value which is previously set for determining whether to change the splitting execution type is disclosed when Lee discloses that PSA relocation and routing rule decisions are dictated by “an operator configuration and SMF logic” [(0082]), i.e. a previously set value/policy governing the change determination.
Regarding claim 20, it is rejected for the same reasons mentioned in claim 16 due to its dependency. Additionally, wherein the updating of the communication session comprises updating routing information about a user plane function (UPF) so that the data transfer device transfers the data traffic to a changed destination, based on the changed splitting execution type is disclosed when “The SMF device transmits an N4 session modification request to the branching point or the UL CL to update an UL traffic filter according to the new IPv6 prefix allocated to PSA2 or an UL CL rule regarding a traffic flow that the SMF device tries to move from PSA1 to PSA2.” [(0146)] and further that “the branching point or the UL CL updates PSA2 to an N9 interface for the traffic filters that need the PSA relocation” ([0146)]. This shows an express disclosure of updating UPF routing information so that the UPF device transfers traffic to a changed destination based on the changed configuration.
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 11-15 are rejected under 35 U.S.C. 103 as being unpatentable by Karjee (US 2022/0311678 A1) and Lee (US 2020/0196201 A1).
Regarding claim 11, Karjee teaches A method of supporting dynamic splitting execution of a machine learning model in a communication system including an edge device and a service server connected with each other over a communication network, the method comprising by disclosing “Referring to FIG. 1, at least one edge device, for example, an edge device 103 is selected from a plurality of edge devices 103 by an IoT device 101 for execution of DNN in an IoT-edge network. The edge device 103 is within the communication range with the IoT device 101.” ([0037]). Additionally, Karjee teaches transmitting splitting execution information including a splitting execution type of the machine learning model to the service server and the network management server by using the edge device by disclosing the ML-model splitting core of this limitation: “ At operation 307, the method 300 includes splitting, by the IoT device 101, a plurality of layers of the DNN into a first part and a second part based on the split ratio.” ([0059]), disclosing transmission of a splitting execution type from the edge device to the service server. Karjee does not disclose transmission of the same information to a distinct network management server. This gap is filled by Lee, which discloses a network management server receiving and acting upon session configuration information associated with data being split/routed between a edge cloud and central cloud: “Session management function (SMF) device: As a function of managing a session of the UE in the 5G CN, the SMF connects the UE to the central cloud and the edge cloud by controlling UPFs through the SMF.” ([0163]). Additionally, Karjee teaches, performing an arithmetic operation on the machine learning model by using at least one of the edge device and the service server, based on the splitting execution type by disclosing “wherein the first part is executed on the IoT device and the second part is executed on the selected at least one edge device.” ([0009)]. Additionally, Lee teaches monitoring a current communication session generated between the edge device and the service server to determine a current communication quality class of the current communication session by using the network management server by disclosing, “An SMF device determines that there is a need to connect a new local gateway, UPF(PSA2), for a corresponding PDU session, based on a load of a central cloud and edge cloud 1, or on a lifecycle of UPF(PSA1). That is, the SMF device determines whether there is a need for PSA relocation.” ([0167]). Additionally, Lee teaches configuring a final communication session by using the network management server, based on a result of comparison of the current communication quality class and a reference communication quality class by disclosing, “The SMF device selects a 5G gateway, UPF(PSA2), to connect a new local gateway, and sets a tunnel ID of the UPF device” ([0168]) and further continues with explaining, “ When all traffic is moved to UPF(PSA0) or UPF(PSA2) through (iii), for example, scenario 2, the SMF device releases UPF(PSA1). However, when the UE still receives a service from edge cloud 1 through UPF(PSA1) even after (iii), the SMF device does not perform (iv) and not release UPF(PSA1).” ([0170]), thus expressly disclosing both branches of the claimed comparison: the SMF configures either the current session or a new session as the final communication session, based on the comparison outcome. Additionally, Karjee teaches and performing data communication for performing an arithmetic operation on the machine learning model by using the edge device and the service server, based on the configured final communication session by disclosing “Once the suitable network is selected, the DNN inference task is partitioned among IoT and edge device, respectively. Based on the partitioned DNN inference, an optimal splitting ratio is determined as described in DSC mechanism. The selected network bandwidth measurements value is considered as BWi∈{cell, Wi-Fi, Bluetooth} where cell, Wi-Fi and Bluetooth are the bandwidth measurements of cellular, Wi-Fi and Bluetooth network, respectively.” [(0103)] “splitting is started based on the optimal splitting ratio and a splitted part of the DNN is transferred from the IoT device to the edge device” [(0104]).
Both references address the same underlying technical problem: reliable, network aware distribution of computing/data tasks between a resource constrained edge device and a more capable server/cloud over a wireless network subject to variable link quality. Karjee teaches that the network provides unreliable connectivity and that it’s necessary to develop an intelligent mechanism to switch to a suitable network, while expressly disclosing implementations of 5G technologies and 6G network deployments. However, Karjee implements network awareness as an internal IoT device function without the standardized session management function that an actual 5G/6G core network uses to monitor and reconfigure sessions at the network level. A POSITA would combine Lee’s SMF/UPF session management framework to monitor and reconfigure Karjee’s split-computing task, yielding predictable results.
Regarding Claim 12, it is rejected for the same reasons mentioned in claim 11 due to its dependency. Additionally, Karjee teaches wherein the splitting execution information comprises a splitting execution type changed based on a resource status change of the edge device by disclosing “ the determined split ratio may be modified based on variations in the one or more inference parameters.” ([0059]) and “the battery level of the Edge device is also tracked. If battery level goes below a certain threshold, the task is unloaded to some other edge device which will give minimum inference time.” ([0082]), hence disclosing the splitting type changing in response to a resource status change of the edge-side device.
Regarding claim 13, it is rejected for the same reasons mentioned in claim 11 due to its dependency. Additionally, Lee teaches, further comprising determining a reference communication quality class for smoothly supporting the splitting execution type by using the network management server by disclosing that the SMF’s relocation determination inherently evaluates the monitored load against an underlying reference/threshold level: “ An SMF device determines that there is a need to connect a new local gateway, UPF(PSA2), for a corresponding PDU session, based on a load of a central cloud and edge cloud 1” ([0167]), this shows a determination that necessarily presupposes a reference load level against which adequacy is judged.
Regarding claim 14, it is rejected for the same reasons mentioned in claim 13 due to its dependency. Additionally, Karjee teaches wherein the determining of the reference communication quality class comprises calculating the reference communication quality class, based on a mapping table where a mapping relationship between the splitting execution type and a communication quality class is previously defined, by disclosing an explicit mapping table correlating communication quality to splitting execution type: “ Less than 6 Mbps -> Complete on-device execution, Greater than 6 Mbps Layers 1-11 -> on-device and less than 11 Mbps Layers 12-31 -> on-edge; Greater than 11 Mbps -> Complete on-edge.
Regarding claim 15, it is rejected for the same reasons mentioned in claim 11 due to its dependency. Additionally, Lee teaches, wherein the configuring of the final communication session comprises: when the current communication quality class is equal to the reference communication quality class, configuring the current communication session as the final communication session; and when the current communication quality class differs from the reference communication quality class, configuring a new communication session as the final communication session by disclosing: “ When all traffic is moved to UPF(PSA0) or UPF(PSA2) through (iii), for example, scenario 2, the SMF device releases UPF(PSA1). However, when the UE still receives a service from edge cloud 1 through UPF(PSA1) even after (iii), the SMF device does not perform (iv) and not release UPF(PSA1)” ([0170]). Disclosing retention of the current session where conditions remain adequate and establishment of a new session when they do not.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SAMUEL NWUHA whose telephone number is (571)272-9367. The examiner can normally be reached Monday-Friday; 7:30 am - 5:00pm.
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, Kevin Young can be reached at (571) 270-3180. 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. 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.
/SAMUEL OBINNA NNAJI NWUHA/
Examiner, Art Unit 2194
/KEVIN L YOUNG/
Supervisory Patent Examiner, Art Unit 2194
Prosecution Insights