DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claims 1-20 are pending for examination. Claims 1, 11, and 20 are independent.
Information Disclosure Statement
The information disclosure statement (IDS) has been submitted on 01/08/2024. The submission is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Specification
The disclosure is objected to because of the following informalities:
Observed clearly on Page 17 of Applicant specification, there is a change in the formatting before and after the section heading "-- Adaptive Model Quantization for Federated Learning --", resulting in the listed line numbers after this heading to be inconsistent with the actual line numbers of the page (e.g., the current line labeled "line 5" is printed on Page 17 as the 12th line from the top of Page 17).
The numbering of lines, as begun on Page 17 and continuing on, is absent in pages 1-17, understood as an inconsistent arrangement of the Applicant specification.
Appropriate correction is required.
Claim Objections
Claims 6 and 16 are objected to because of the following informalities:
“reduce a network bandwidth associated with the trainer clients” should be corrected to “reduce network bandwidth consumption associated with the trainer clients”. The claim will be interpreted as such, in light of Applicant specification [Page 17 (Labeled) Lines 3-7; Page 20 Lines 20-23] that clearly specifies reducing “bandwidth consumptions” and not just reducing “bandwidth”.
Appropriate correction is required.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1: The claim is a process, machine, manufacture, or composition of matter.
In the instant application, Claims 1-10 are directed to a method, Claims 11-19 are directed to a machine, and Claim 20 is directed to a machine. Thus, each of the claims falls within one of the four statutory categories (i.e. process, machine, manufacture, or composition of matter).
With regard to Claim 1:
2A Prong 1: The claim recites an abstract idea, law of nature, or natural phenomenon.
A method comprising: determining, by the supervisory device, an amount of resource savings associated with the trainer clients quantizing their local machine learning models; and (This step of determining an amount of resource savings is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. evaluation).)
2A Prong 2: The judicial exception is not integrated into a practical application.
A method comprising: receiving, at a supervisory device, one or more goal parameters for trainer clients in a federated learning system configured to train local machine learning models using local datasets; (Receiving information is understood as an insignificant extra-solution activity — see MPEP 2106.05(g).)
configuring, by the supervisory device and based on the one or more goal parameters, the trainer clients to quantize their local machine learning models prior to sending them for aggregation into a global model; (This step is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).)
providing, by the supervisory device, an indication of the amount of resource savings for presentation to a user. (Presenting information is understood as an insignificant extra-solution activity — see MPEP 2106.05(g).)
The additional elements as disclosed above alone or in combination do not integrate the judicial exception into practical application as they are insignificant extra-solution activity in combination with merely invoking computers or machinery as a tool to perform the abstract idea.
2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
A method comprising: receiving, at a supervisory device, one or more goal parameters for trainer clients in a federated learning system configured to train local machine learning models using local datasets; (Receiving information is understood as a well-understood, routine, and conventional function, being exemplary of receiving data — see MPEP 2106.05(d)(II)(i).)
configuring, by the supervisory device and based on the one or more goal parameters, the trainer clients to quantize their local machine learning models prior to sending them for aggregation into a global model; (This step is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).)
providing, by the supervisory device, an indication of the amount of resource savings for presentation to a user. (Providing information is understood as a well-understood, routine, and conventional function, being exemplary of presenting offers and gathering statistics — see MPEP 2106.05(d)(II)(iv).)
The additional elements as disclosed above alone or in combination do not recite significantly more than a judicial exception as they are well-understood, routine, and conventional functions in combination with merely invoking computers or machinery as a tool to perform the abstract idea.
With regard to Claim 11:
Claim 11 has similar limitations to Claim 1, therefore it is rejected under the same rationale. Additional limitations of Claim 11 are addressed below.
2A Prong 2 & 2B: An apparatus, comprising: one or more network interfaces; a processor coupled to the one or more network interfaces and configured to execute one or more processes; and a memory configured to store a process that is executable by the processor, the process when executed configured to: (The network interface(s), process(es), and memory are understood as mere instructions to apply the exception using a generic computer component — see MPEP 2106.05(f).)
With regard to Claim 20:
Claim 20 has similar limitations to Claim 1, therefore it is rejected under the same rationale. Additional limitations of Claim 20 are addressed below.
2A Prong 2 & 2B: A tangible, non-transitory, computer-readable medium storing program instructions that cause a supervisory device in a federated learning system to execute a process comprising: (The tangible, non-transitory, computer-readable medium is understood as mere instructions to apply the exception using a generic computer component — see MPEP 2106.05(f).)
With regard to Claims 2 and 12:
2A Prong 1: The claim does not recite an Abstract idea.
2A Prong 2 & 2B: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein the one or more goal parameters include a task for which a specified one of the trainer clients is to train its local machine learning model to perform. (The specification of the one or more goal parameters is understood to be a field of use limitation — see MPEP 2106.05(h).)
With regard to Claims 3 and 13:
2A Prong 1: The claim does not recite an Abstract idea.
2A Prong 2 & 2B: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein the one or more goal parameters include a compression ratio for a specified one of the trainer clients. (The specification of the one or more goal parameters is understood to be a field of use limitation — see MPEP 2106.05(h).)
With regard to Claims 4 and 14:
2A Prong 1: The claim does not recite an Abstract idea.
2A Prong 2 & 2B: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein at least one of the trainer clients quantizes its local machine learning model after training that model. (This step is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).)
With regard to Claims 5 and 15:
2A Prong 1: The claim does not recite an Abstract idea.
2A Prong 2 & 2B: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein two or more of the trainer clients apply different degrees of quantization to their local machine learning models based on the one or more goal parameters. (This step is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).)
With regard to Claims 6 and 16:
2A Prong 1: The claim does not recite an Abstract idea.
2A Prong 2 & 2B: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein the one or more goal parameters seek to reduce a network bandwidth associated with the trainer clients sending the local machine learning models for aggregation. (The specification of the goal parameter(s) is understood to be a field of use limitation — see MPEP 2106.05(h).)
With regard to Claims 7 and 17:
2A Prong 1: The claim does not recite an Abstract idea.
2A Prong 2 & 2B: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein the amount of resource savings indicates a power consumption savings. (The specification of the resource savings is understood to be a field of use limitation — see MPEP 2106.05(h).)
With regard to Claims 8 and 18:
2A Prong 1: The claim does not recite an Abstract idea.
2A Prong 2 & 2B: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein the one or more goal parameters indicate a particular hardware platform that one of the trainer clients is to use to train its local machine learning model. (The specification of the goal parameter(s) is understood to be a field of use limitation — see MPEP 2106.05(h).)
With regard to Claims 9 and 19:
2A Prong 1: The claim does not recite an Abstract idea.
2A Prong 2 & 2B: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein a global aggregator server in the federated learning system de-quantizes the local machine learning models and aggregates them into the global model. (This step is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).)
With regard to Claim 10:
2A Prong 1: The claim does not recite an Abstract idea.
2A Prong 2 & 2B: The method as in claim 1, wherein the global model is configured to classify image data. (This step of configuring the global model is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. observation/evaluation/judgement/opinion) — see MPEP 2106.05(f).)
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.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Choi et al. (US 2022/0245527 A1), hereinafter “Choi”, in view of Chen et al. (US 2024/0174254 A1), hereinafter “Chen”, further in view of Venkatesan et al. (US 2021/0174214 A1), hereinafter “Venkatesan”.
With regard to Claim 1:
Choi teaches: A method comprising:
receiving, at a supervisory device, one or more goal parameters for trainer clients in a federated learning system configured to train local machine learning models using local datasets; ([0004] Choi recites “the server may determine a respective quantization level for each [user equipment] UE (i.e. trainer client)… the server or the base station (i.e. supervisory device) or both may determine a quantization level for a UE based on a channel condition (e.g., a link budget, a channel bandwidth, a channel quality, or some other channel condition).” [0004, 0077] Choi also recites “the US may transmit a capability message to the server or the base station (i.e. supervisory device) or both that indicates a set of quantization levels supported by the UE, and the server or the base station or both may select and indicate, at each iteration, quantization levels from the set of supported quantization levels for the UE to use,” wherein the base station or supervisory device is receiving the set of quantization levels supported by the UE which is considered a goal parameter as it would be a goal of the base station to determine a quantization level within the parameters of the set received from the UE. [0032-0033] Choi clarifies that the quantization may be based on a bandwidth of a channel, and that the machine learning model is a federated learning model. As the channel condition is stated to act as a limit or condition that indicates the level of quantization to be performed, it may be understood as a constraint or a goal parameter for the UE. [0046] Choi recites “workers (e.g. UEs per [0045]) may continue to train local models using local datasets that are unique to each worker.”)
configuring, by the supervisory device and based on the one or more goal parameters, the trainer clients to quantize their local machine learning ([0004] Choi recites that from the server or base station, “each [user equipment] UE may receive the information [from the server or the base station] and the indication of the respective quantization level and may compress (i.e. quantize) gradient data output by a respective machine learning model according to the indicated quantization level.” Choi goes on to recite “each UE may transmit the compressed gradient data to the server or the base station or both,” and [0045] “The server may aggregate the local gradient data received from the set of workers and may update the global model at the server (e.g., using gradient averaging).” The UEs (i.e. trainer clients) performing compression after receiving the quantization level indication from the base station (i.e. supervisory device) is understood as being the trainer clients being configured to quantize by the supervisory device. The compressed or quantized model gradient data is then sent by the base station into the server where it is all aggregated into the global model.)
Choi does not explicitly teach: the trainer clients quantizing their local machine learning models prior to sending them for aggregation into a global model, teaching instead the trainer clients quantizing the local machine learning model gradient data to be aggregated to update a global model.
Chen teaches: the trainer clients quantizing their local machine learning models prior to sending them for aggregation into a global model ([Abstract, 0016, 0044] Chen teaches the one or more edge nodes (i.e. trainer clients) “compressing (i.e. quantizing) the trained machine learning model based on the determined level of compression, and transmits the compressed trained machine learning model to the server (i.e. supervisory device). The server decompresses the compressed trained machine learning models and aggregates the decompressed trained machine learning models to obtain the aggregated (i.e. global) machine learning model.”)
It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Choi’s teachings by compressing and transmitting the full models of the trainer clients instead of just the model gradients as taught by Chen. One would have been motivated to make this modification in order to obtain an aggregate model that fully encompasses the aspects of the local models from each trainer client, providing a more nuanced model better equipped to perform any potential tasks within its scope.
Chen does not explicitly teach: determining, by the supervisory device, an amount of resource savings associated with the trainer clients quantizing their local machine learning models; and
providing, by the supervisory device, an indication of the amount of resource savings for presentation to a user.
However, Venkatesan teaches in the same field of endeavor: determining, by the supervisory device, an amount of resource savings associated with the trainer clients quantizing their local machine learning models; and ([0051] Venkatesan discusses validating the quantized version of the DNN (deep neural network), wherein “performance of the quantized version of one or both of the DNNs may include… memory usage, and processing time” and how the system “may present the performance data… in one or more visualization tools. [Fig. 21, 0202] Venkatesan provides an example of a visualization tool UI, wherein [Column 2114 of Fig. 21] the percent change of different resource metrics may be viewed between the DNN before and after quantization. Further indication of a resource savings is stated: “quantization achieved nearly 75% reduction in memory utilization of the learnable parameters.”)
providing, by the supervisory device, an indication of the amount of resource savings for presentation to a user. ([Figs. 12, 17B, 21] present example graphical user interfaces that show an indication of performance metrics for original and quantized neural networks, in which the user may compare these metrics to determine the relative differences or resource savings.)
It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Choi in view of Chen’s teachings by gathering and presenting or visualizing data about the amount of resources saved through model quantization as taught by Venkatesan. One would have been motivated to make this modification in order to allow users to more easily access feedback information and subsequently adjust their federated learning systems accordingly in order to further improve those resource savings and accuracy-cost tradeoffs.
With respect to Claim 11:
Choi in view of Chen and Venkatesan teaches: An apparatus, comprising: one or more network interfaces; a processor coupled to the one or more network interfaces and configured to execute one or more processes; and a memory configured to store a process that is executable by the processor, the process when executed configured to: ([Fig. 11, 0174] Choi describes “device 1105” that includes “a network communications manager 1110”, “a memory 1130”, and “a processor 1140” “coupled via one or more buses (e.g., a bus 1150).”)
Claim 11 is a device claim that corresponds to Claim 1 and the remaining limitations are rejected for at least the same reasons therein.
With respect to Claim 20:
Choi in view of Chen and Venkatesan teaches: A tangible, non-transitory, computer-readable medium storing program instructions that cause a supervisory device in a federated learning system to execute a process comprising: ([0008] Choi teaches “A non-transitory computer-readable medium storing code” that “may include instructions executable by a processor” and is related to a [0017] “federated learning model associated with a set of UEs.”)
Claim 20 is a tangible, non-transitory, computer-readable medium claim that corresponds to Claim 1 and the remaining limitations are rejected for at least the same reasons therein.
With regard to Claims 2 and 12:
Choi in view of Chen and Venkatesan teaches: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein the one or more goal parameters include a task for which a specified one of the trainer clients is to train its local machine learning model to perform. ([0020] Chen describes the server. or supervisory device sending an initialized model to each of the edge nodes or trainer clients, wherein “the initialized model may be any model that may be utilized for operating a vehicle, for example, an image processing model, an object detection model, or any other model for advanced driver assistance systems. Each of the edge nodes trains the received initialized model using local data,” wherein the intended function of each model is understood as a task or goal parameter.)
With regard to Claims 3 and 13:
Choi in view of Chen and Venkatesan teaches: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein the one or more goal parameters include a compression ratio for a specified one of the trainer clients. ([0005, 0014, 0030, 0046] Choi describes receiving an indication of “a respective quantization level for each UE,” wherein “each quantization level of the set of quantization levels [is] associated with a dimensional parameter of a set of dimensional parameters associated with the gradient data output by the machine learning model.” The quantization level is understood as a compression ratio, as it informs the trainer client as to the amount of compression that should be performed at some ratio (e.g., 32 bits, 64 bits, 128 bits, or some other quantity of bits).)
With regard to Claims 4 and 14:
Choi in view of Chen and Venkatesan teaches: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein at least one of the trainer clients quantizes its local machine learning model after training that model. ([Fig. 3, 0047-0050] Chen describes the process flow of the system, wherein an earlier step 310 “each of a plurality of edge nodes trains a local machine learning model using first local data”, and then later in step 340 “each of the plurality of edge nodes compresses corresponding trained machine learning model based on the determined level of compression.”)
With regard to Claims 5 and 15:
Choi in view of Chen and Venkatesan teaches: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein two or more of the trainer clients apply different degrees of quantization to their local machine learning models based on the one or more goal parameters. ([0021] Chen recites “each of the edge nodes 101, 103, 105, 107, and 109 may determine whether or not to compress its locally trained machine learning model before transmitting the model to the server 106 based on bandwidth information between corresponding edge node and the server 106… each of the edge nodes 101, 103, 105, 107, and 109 may determine a level of compression for the locally trained machine learning model based on the current bandwidth information.” [0023] Chen clarifies that “While FIG. 1B depicts three vehicles [or edge nodes], more than or less than three vehicles may be included in the present system.”)
With regard to Claims 6 and 16:
Choi in view of Chen and Venkatesan teaches: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein the one or more goal parameters seek to reduce a network bandwidth associated with the trainer clients sending the local machine learning models for aggregation. (Examiner understands “reduce a network bandwidth” as “reduce network bandwidth consumption”. [0021, 0033] Chen discusses the compression level adapting to different levels of network bandwidth, exemplifying “if the network bandwidth is relatively low or the channel between a vehicle and a server allows relatively low speed data communication, the level of compression may need to be relatively high. i.e., a high compression ratio.” This is understood to mean that the compression ratio goal parameter, associated with the edge node compressing its trained machine learning model prior to sending it for aggregation, is increased in order to reduce the amount of bandwidth used in sending the model from the edge node to the server and allowing for a full transmission of the model to the server within a designated time window.)
With regard to Claims 7 and 17:
Choi in view of Chen and Venkatesan teaches: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein the amount of resource savings indicates a power consumption savings. ([0041] Venkatesan recites “quantizing a trained Deep Neural Network (DNN)… to meet desired performance objectives, such as one or more of throughput, inference accuracy, power/energy usage, memory usage, latency, and execution speed,” as well as [0191] “a trade-off analysis among… power/energy usage may be performed,” in which the trade-off analysis may be between such metrics as power/energy usage and DNN quantization. While [Fig. 21] does not explicitly illustrate an example of power/energy usage as a metric for viewing, it would be obvious to add that metric into the UI as it is defined in the same “performance objectives” category as the shown “memory usage” in [2108b] and interpret the negative percent change as a resource savings indication.)
With regard to Claims 8 and 18:
Choi in view of Chen and Venkatesan teaches: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein the one or more goal parameters indicate a particular hardware platform that one of the trainer clients is to use to train its local machine learning model. ([0081] Venkatesan describes the “UI 900” that includes a “data entry field 904” to indicate or specify “one or more target hardware platforms” upon which the neural network may be quantized. The data entry of the target hardware platform is understood as a goal parameter received indicating a particular hardware platform.)
With regard to Claims 9 and 19:
Choi in view of Chen and Venkatesan teaches: The method as in claim 1 and the apparatus as in claim 11 respectively, wherein a global aggregator server in the federated learning system de-quantizes the local machine learning models and aggregates them into the global model. ([0042] Chen states that the server includes one or more memory modules that may include a model decompressor and a global model update module. [0043] Chen recites “The model decompressor 245 may decompress one or more compressed trained machine learning models received from edge nodes.” [0044] Chen recites “The global model update module 247 aggregates decompressed local machine learning models… to generate a global model.”)
With regard to Claim 10:
Choi in view of Chen and Venkatesan teaches: The method as in claim 1, wherein the global model is configured to classify image data. ([0031] Chen recites “The ML model training module 207 may train the initial model received from the server 106 using local data obtained by the first edge node system 200, for example, images obtained by imaging sensors such as cameras of a vehicle… The ML model training module 207 may also train an aggregated global model received from the server 106 using local data obtained by the first edge node system 200.”)
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BRIAN D. BUI whose telephone number is (571)270-0463. The examiner can normally be reached Monday - Friday 8:00am - 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, ABDULLAH AL KAWSAR can be reached at (571) 270-3169. 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.
/BRIAN D. BUI/Examiner, Art Unit 2127
/ABDULLAH AL KAWSAR/Supervisory Patent Examiner, Art Unit 2127
