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 .
DETAILED ACTION
This action is responsive to the following communication: Amendment filed Jul. 28, 2026. This Action is made Final.
Claims 1-20 are pending in the case. Claims 1, 15 and 16 are independent claims.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Shrihari et al. (hereinafter SHR) WO 2023/144831 in view of LASKARIDIS et al. (hereinafter LASK) U.S. Patent Publication No. 2022/0245459.
With respect to independent claim 1, SHR teaches a method in a distributed system comprising first computer systems that are configured to connect to at least one second computer system of the distributed system (see e.g., Fig. 17 Para[231]-[238]-“ , the communication system QQlOO may include any number of wired or wireless networks, network nodes, UEs, and/or any other components or systems that may facilitate or participate in the communication of data and/or signals whether via wired or wireless connections. The communication system QQlOO may include and/or interface with any type of communication, telecommunication, data, cellular, radio network, and/or other similar type of system”), the method comprising:
determining, by a specific first computer system of the first computer systems, a need to fine-tune a first artificial intelligence model for performing a specific task (see e.g., Fig. 3C, 4 and Para [91]-[102]-“ in the event that the identified ML model performs poorly on the new data (in other words, not within the threshold) is a fresh ML model creation triggered … In block 419, the node determines whether PERF-TEST is within Tl of PERF. If no, the method proceeds to block 409 and the operations of blocks 411-419 are repeated. If yes, the method proceeds to block 421. In some embodiments, comparing PERF-TEST with existing ML model performance of PERF is a relative performance comparison. In some embodiments, an absolute performance comparison maybe used, e.g., is PERF-TEST within Tl (e.g., is the error within 10 CPU units), to determine ML model compatibility for using the ML model as-is or using the ML model after refinement.”);
performing federated learning for a set of trained second artificial intelligence models for generating a combined artificial intelligence model(see e.g., Fig. 3C, 4 and Para [70] [90]-[102][163][200]-[206] – “either the ML model is refined on D-NEW (e.g., fine tune from COEF) or a new ML model is retrained afresh (e.g., an ordinary least-squares (OLS model)) on D-SEED and D-NEW combined. The resulting ML model is denoted as M-NEW ““The online or offline active learning procedure 209 is used for the refinement, depending on the application context.“); and
using learnable parameters of the combined artificial intelligence model for fine-tuning, at the specific first computer system, the first artificial intelligence model (see e.g., Fig. 4 Para [82][102]-[113]- “The resulting ML model is denoted as M-NEW and its performance is computed as PERF-NEW.”” In block 431, ML model predictions are returned if required, and Model M-NEW is ready to continue serving predictions.”).
SHR does not expressly show the features discussed below. However, LASK teaches the learning type is federated learning (see e.g. para [4]) and selecting a set of trained second artificial intelligence models for participation in federated learning (see e.g. para [4][76][140]-[146] – “Typically federated learning of a machine learning, ML, model requires multiple rounds where i) a number of participating client devices are selected, ii) the latest ML model is sent to the client devices from a central server or orchestrator of the federated learning, iii) the client devices use their local data to update the ML model (local training), iv) the client devices share the updated model with the server, and v) the resulting models (one per client) are aggregated by the server into a single ML model before starting another round.” “At the beginning of each communication round t, the number participating devices St and the number of local iterations are selected. The device pool St of available devices is sampled until the required number of devices need to participate is obtained; the required number could be all available clients A.sub.t or a subset of A.sub.t, depending on server capacity.” “At each local round t, the device i samples p.sub.(l,k)˜U.sub.P, st. p∈(0, p.sup.C.sub.max] and updates the respective weights w.sub.p(i,k) of the local submodel; the weights may be updated using a suitable rule such as FedAvg. Each device i runs E local iterations (rounds) k, and at the end of the local rounds, each device sends back the gradients of the maximally updated submodel“); each second artificial intelligence model having a structure that is at least a substructure of the first artificial intelligence model(see e.g. para [43] – “The machine learning, ML, model may be considered a super-model, from which multiple nested submodels may be extracted and trained. The term “nested submodels” is used herein to mean that the submodels are of graduated sizes and are located one inside the other. Thus, a submodel which is sent to a client device for training may itself contain nested submodels.”)
Both SHR and LASK are directed to method of training machine learning models. Accordingly, it would have been obvious to the skilled artisan before the effective filing date of the claimed invention having SHR and LASK in front of them to modify the system of SHR to include the above feature. The motivation to combine SHR and LASK comes from LASK. LASK discloses the motivation to train selected sub-models instead of larger/whole models so that training is more efficient (see e.g. para [43][44][142][143]). This motivation for combination also applies to the remaining claims which depend on this combination.
With respect to dependent claim 2, the modified SHR teaches selecting the set of second artificial intelligence models includes selecting from a larger set of second artificial intelligence models based on the specific task (see e.g., Para [6]-[11][91][95][117] - “Challenges in ML model identification also may be resolved based on the method including ML model selection from a repository of ML models, with or without deployment metadata or a target performance metric.” Also see LASK Para [43][44][142][143]).
With respect to dependent claim 3, the modified SHR teaches the set of second artificial intelligence models being a subset of a larger set of second artificial intelligence models, wherein the set of second artificial intelligence models are described by model attributes, the method further comprising: evaluating the model attributes for the larger set of second artificial intelligence models (see e.g., Para[76]-[91]-“Relevant ML models with matching deployment metadata are pulled up from storage (e.g., from a model registry). Incoming data (or a sample of incoming data) (0-TEST) is evaluated against the relevant ML models. The best performing ML model is identified.”); storing, in a database, records representing the larger set of second artificial intelligence models, the records comprising the evaluated model attributes (see e.g., Para [51]-[57][166]-[179]-“Model registry 111 can include all model related artifacts (e.g., model data 111b, performance metrics 111c, deployment metadata 111d). A deployed ML model can continue serving dimensioning outcomes for incoming (e.g., test) data.”); indexing the database using one or more of the model attributes, resulting in an index (see e.g., Para [50][154] – “controller that can work with a ML model registry 111 (e.g., a memory, a database, a repository, etc.). LCM controller of node 107 can (1) select 107a a minimal informative subset of data from incoming data 101; and (2) identify 107b a ML model(s) from ML registry 111 ““the ML models are sorted based on PERF-TEST-S, and the top-K ML models are selected”); and using the index and the specific task for identifying the set of trained second artificial intelligence models (see e.g., Para[91][95]-“Relevant ML models with matching deployment metadata are pulled up from storage (e.g., from a model registry). Incoming data (or a sample of incoming data) (0-TEST) is evaluated against the relevant ML models.”).
With respect to dependent claim 4, the modified SHR teaches the evaluating and the storing are performed on a periodic basis (see e.g., Para [43][45]-“a proactive periodic refinement to the current ML model to proactively to follow a drift in a subsequent dataset”).
With respect to dependent claim 5, the modified SHR teaches logging changes to the database in a log file, using the log file for tracking changes to the database; and using the index based on the tracked changes (see e.g., Fig. 4 Para [104] – steps 425 427).
With respect to dependent claim 6, the modified SHR teaches using the index comprising: evaluating at least part of the model attributes for the first artificial intelligence model (see e.g., Para[97]-[103]-“a performance metric PERF-TEST is computed”); defining a query based on the evaluated model attributes; querying the database using the index and the defined query (see e.g., Para [117]-“ Relevant ML models with matching deployment metadata are pulled up from storage (e.g., from a model registry)”); receiving a response of the query comprising candidate second artificial intelligence models (see e.g., Para[96]-[112][154]); and selecting the set of trained second artificial intelligence models from the candidate second artificial intelligence models using a selection criterion (see e.g., Para[99]-[103][154][171] – “the ML models are sorted based on PERF-TEST-S, and the top-
K ML models are selected”).
With respect to dependent claim 7, the modified SHR teaches the candidate second artificial intelligence models have a matching level with the defined query that is higher than a minimum threshold (see e.g., Para [114][141][155] - “the node determines whether a best model (denoted in this example embodiment as Ml) performance is less than threshold Tl.”).
With respect to dependent claim 8, the modified SHR teaches the minimum threshold being defined by the specific first computer system (see e.g., Para [63][76] - “Tl and T2 have default values defined” SHR does not expressly indicate whether the threshold is the minimum or maximum threshold. However, it would have been obvious to set minimum threshold).
With respect to dependent claim 9, the modified SHR teaches the selection criterion requiring an inference accuracy of the candidate second artificial intelligence model that is better than an accuracy threshold (see e.g., Para [100]-[114][145]-“PERF-TEST with existing ML model performance of PERF is a relative performance comparison. In some embodiments, an absolute performance comparison maybe used,”).
With respect to dependent claim 10, the modified SHR teaches the federated learning being performed by aggregating the learnable parameters of the set of second artificial intelligence models (see e.g., LASK Para[52][57][68]-[70][132]-[146]- “aggregating, using the received gradients, the changes in weights of the submodel received from each client device; and updating the ML model. “ “where w.sub.s.sub.j\w.sub.s.sub.j−1 are the weights that belong to F.sub.s.sub.j but not to F.sub.s.sub.j−1, w.sup.t+1 the global weights at communication round t+1, w.sup.(i,t,E) the weights on client i at communication round t after E local iterations, S.sub.t.sup.j=i{i∈S.sub.t:p.sub.max.sup.i≥s.sub.j} a set of clients that have the capacity to update w.sub.s.sub.j, and WA stands for weighted average, where weights are proportional to the amount of data on each client.”).
With respect to dependent claim 11, the modified SHR teaches the aggregating is done using a weighted sum, wherein weights are inference accuracies of the set of second artificial intelligence models, respectively (see e.g., LASK Para[52][57][132]-[146]).
With respect to dependent claim 12, the modified SHR teaches the first computer system has an amount of processing resources that is smaller than the processing resources of the at least one second computer system (see e.g., Para[42]-“deploy a ML model may not be feasible (e.g., on-premises low-resource cloud infrastructure, such as limited computational resources) or at it may be time-consuming and wasteful. Thus, it may be desirable to compress the new dataset sufficiently to enable memory-efficient and/or fast training, re-training, deployment, and/or re-deployment of a ML model(s)” SHR does not require certain amount of processing resources).
With respect to dependent claim 13, the modified SHR teaches the distributed system being a wireless communication system, wherein the first computer systems are multi-access edge computing (MEC) nodes and the at least one second computer system is a cloud system (see e.g., Fig. 17 Para [32]-[38]).
With respect to dependent claim 14, the modified SHR teaches each artificial intelligence model is a foundation model (see e.g., Para [9][10] – The examiner notes that it is not clear how foundation model is defined.).
Claim 15 is rejected for similar reasons discussed above with respect to claim 1.
Claim 16 is rejected for similar reasons discussed above with respect to claim 1.
With respect to dependent claim 17, the modified SHR teaches the computer system of claim 16, being a first computer system of the first computer systems (see e.g., Fig. 17).
With respect to dependent claim 18, the modified SHR teaches the set of trained second artificial intelligence models is a subset of a larger set of artificial intelligence models(see e.g., Para[76]-[91] [144][204]-“Relevant ML models with matching deployment metadata are pulled up from storage (e.g., from a model registry). Incoming data (or a sample of incoming data) (0-TEST) is evaluated against the relevant ML models. The best performing ML model is identified.” “evaluating a performance of each of the plurality of ML models on a subset of the new data, (ii) identifying a subset of the plurality of ML models to retrain on a new dataset, the subset of the plurality of ML models identified based on a defined value that sets the number”).
With respect to dependent claim 19, the modified SHR teaches selecting the set of trained second artificial intelligence models from the larger set of artificial intelligence models using a selection criterion (see e.g., Para [154] [204]- “In block 615, the ML models are sorted based on PERF-TEST-S, and the top- K ML models are selected. The method proceeds to block 617 to perform operations to use an existing ML model.” “identifying a subset of the plurality of ML models to retrain on a new dataset, the subset of the plurality of ML models identified based on a defined value that sets the number best performing ML models to include in the subset of the plurality of ML models “The selection criteria is the tested performance metric PERF-TEST-S).
With respect to dependent claim 20, the modified SHR teaches the selection criterion indicates that an inference accuracy of artificial intelligence models in the set of trained second artificial intelligence models satisfies an accuracy threshold (see e.g., Para [145][152][154][191] – “ML model M is applied on D-TEST-S-PP, and a performance metric PERF-TEST-S is computed … the ML models are sorted based on PERF-TEST-S, and the top- K ML models are selected.”).
It is noted that any citation to specific pages, columns, lines, or figures in the prior art references and any interpretation of the references should not be considered to be limiting in any way. “The use of patents as references is not limited to what the patentees describe as their own inventions or to the problems with which they are concerned. They are part of the literature of the art, relevant for all they contain.” In re Heck, 699 F.2d 1331, 1332-33, 216 USPQ 1038, 1039 (Fed. Cir. 1983) (quoting In re Lemelson, 397 F.2d 1006, 1009, 158 USPQ 275, 277 (CCPA 1968)). Further, a reference may be relied upon for all that it would have reasonably suggested to one having ordinary skill the art, including nonpreferred embodiments. Merck & Co. v. Biocraft Laboratories, 874 F.2d 804, 10 USPQ2d 1843 (Fed. Cir.), cert. denied, 493 U.S. 975 (1989). See also Upsher-Smith Labs. v. Pamlab, LLC, 412 F.3d 1319, 1323, 75 USPQ2d 1213, 1215 (Fed. Cir. 2005); Celeritas Technologies Ltd. v. Rockwell International Corp., 150 F.3d 1354, 1361, 47 USPQ2d 1516, 1522-23 (Fed. Cir. 1998).
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 PEIYONG WENG whose telephone number is (571)270-1660. The examiner can normally be reached on Mon.-Fri. 8 am to 5 pm.
If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, Matthew Ell, can be reached on (571) 270-3264. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://portal.uspto.gov/external/portal. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
/PEI YONG WENG/Primary Examiner, Art Unit 2141
Prosecution Insights