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 .
Response to Amendment
The Amendment filed 6/18/2026 has been entered. Claims 1 and 18 have been amended. Claims 1-20 are pending in the application.
Drawings
The drawings are objected to because in FIG. 8 the block designated by reference character 225 bears the label "Knowledge base update model 225," whereas the same block is labeled "Knowledge base update module 225" in FIG. 2, FIG. 9 and FIG. 15, and the description at paragraph [0106] designates 225 the knowledge base update module. Corrected drawing sheets in compliance with 37 CFR 1.121(d) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. The figure or figure number of an amended drawing should not be labeled as "amended." Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either "Replacement Sheet" or "New Sheet" pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
Figure 1 should be designated by a legend such as --Prior Art-- because only that which is old is illustrated. See MPEP 608.02(g). Corrected drawings in compliance with 37 CFR 1.121(d) are required in reply to the Office action to avoid abandonment of the application. The replacement sheet(s) should be labeled "Replacement Sheet" in the page header (as per 37 CFR 1.84(c)) so as not to obstruct any portion of the drawing figures. If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
The drawings are objected to as failing to comply with 37 CFR 1.84(p)(5) because they do not include the following reference sign(s) mentioned in the description: 1601, first occurring at paragraph [0206] and designating the bus of the device 1600; 1602, first occurring at paragraph [0206] and designating the processor; 1603, first occurring at paragraph [0206] and designating the communications interface; and 1604, first occurring at paragraph [0206] and designating the memory. Corrected drawing sheets in compliance with 37 CFR 1.121(d) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either "Replacement Sheet" or "New Sheet" pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
The drawings are objected to as failing to comply with 37 CFR 1.84(p)(5) because they include the following reference character(s) not mentioned in the description: 1610 (FIG. 16, designating the communications interface), 1620 (FIG. 16, designating the processor), and 1630 (FIG. 16, designating the memory). Corrected drawing sheets in compliance with 37 CFR 1.121(d), or amendment to the specification to add the reference character(s) in the description in compliance with 37 CFR 1.121(b) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either "Replacement Sheet" or "New Sheet" pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
The two objections immediately above concern one discrepancy and may be resolved together: FIG. 16 numbers the communications interface, the processor and the memory 1610, 1620 and 1630, while paragraph [0206] numbers the bus, the processor, the communications interface and the memory 1601, 1602, 1603 and 1604. Applicant may reconcile the figure to the description or the description to the figure, provided the bus shown in FIG. 16 is also given a reference character and no new matter is introduced.
Specification
The disclosure is objected to because of the following informalities: paragraph [0092] refers to "the plurality of task models stored in the knowledge base module 230," but the knowledge base module is designated 210 throughout the remainder of the disclosure and no element 230 appears in the drawings; and paragraph [0017] twice refers to "the inference model" where the inference module appears to be intended, in the sentences "the model determining module and the inference model may be randomly deployed" and "both the model determining module and the inference model are deployed in the cloud." Appropriate correction is required.
The lengthy specification has not been checked to the extent necessary to determine the presence of all possible minor errors. Applicant's cooperation is requested in correcting any errors of which applicant may become aware in the specification.
The specification is objected to as failing to provide proper antecedent basis for the claimed subject matter. See 37 CFR 1.75(d)(1) and MPEP 608.01(o). Correction of the following is required: claim 10 recites a "model determiner" and an "inference performer," claim 11 recites an "attribute determiner" and a "task determiner," and claim 16 recites a "knowledge base updater," but none of those terms appears in the description, which instead describes a model determining module 223, an inference module 224, an attribute definition module 221, a task determining module 222, and a knowledge base update module 225. Correction of the specification to provide antecedent basis for these claim terms is required, and no new matter may be introduced.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. - An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term "means" or "step" or a term used as a substitute for "means" that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term "means" or "step" or the generic placeholder is modified by functional language, typically, but not always linked by the transition word "for" (e.g., "means for") or another linking word or phrase, such as "configured to" or "so that"; and
(C) the term "means" or "step" or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word "means" (or "step") in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word "means" (or "step") in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word "means" (or "step") are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word "means" (or "step") are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word "means," but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitations are:
(1) a model determiner deployed in a cloud or an edge side network and configured to generate, when the inference task corresponding to the input sample is the unknown task, the inference model based on the at least one task attribute and the corresponding first task model, recited in claim 10. The term "determiner" is a generic placeholder, and the modifier "model" identifies its task rather than implementing structure. The recitation "deployed in a cloud or an edge side network" specifies where the placeholder is deployed rather than structure for performing the entire recited function. The claimed function is to generate, when the inference task corresponding to the input sample is the unknown task, the inference model based on the at least one task attribute and the corresponding first task model. The corresponding structure is the processor 1602 of the device 1600 executing the program code of the model determining module 223 stored in the memory 1604, as described in [0206] and [0211]-[0213], programmed with the procedure of [0017] and [0092]-[0102], namely selecting candidate task models whose task attributes differ from the target task attribute by less than a threshold, combining the selected candidate task models, retraining one or more of them on the stored training samples, or training a new task model on those samples, and using the result as the inference model. This limitation is interpreted to cover the foregoing structure and its disclosed procedure, and equivalents thereof.
(2) an inference performer deployed in the edge side network and configured to perform inference on the input sample using the inference model to obtain the target inference result, recited in claim 10. The term "performer" is a generic placeholder, and the modifier "inference" identifies its task rather than implementing structure. The recitation "deployed in the edge side network" specifies where the placeholder is deployed rather than structure for performing the entire recited function. The claimed function is to perform inference on the input sample using the inference model to obtain the target inference result. The corresponding structure is the processor 1602 of the device 1600 executing the program code of the inference module 224 stored in the memory 1604, as described in [0206] and [0211]-[0213], programmed with the procedure of [0017], [0091], [0095], and [0198]-[0199], namely applying the received inference model to the input sample and, when the inference model comprises a plurality of models, applying every model of the plurality and reducing the resulting inference results to a single target inference result by averaging or by majority selection. This limitation is interpreted to cover the foregoing structure and its disclosed procedure, and equivalents thereof.
(3) an attribute determiner deployed in an edge side network and configured to determine a target task attribute of the input sample based on the input sample and the at least one task attribute, recited in claim 11. The term "determiner" is a generic placeholder, and the modifier "attribute" identifies its task rather than implementing structure. The recitation "deployed in an edge side network" specifies where the placeholder is deployed rather than structure for performing the entire recited function. The claimed function is to determine a target task attribute of the input sample based on the input sample and the at least one task attribute. The corresponding structure is the processor 1602 of the device 1600 executing the program code of the attribute definition module 221 stored in the memory 1604, as described in [0206] and [0211]-[0213], programmed with the procedure of [0018] and [0075]-[0077], namely matching the input sample against each stored task attribute to determine the task attribute category present in the input sample and its attribute value when the input sample is data item information, and extracting the target task attribute with a pre-trained classifier when the input sample is not a data item. This limitation is interpreted to cover the foregoing structure and its disclosed procedure, and equivalents thereof.
(4) a task determiner deployed in the edge side network and configured to determine, based on the target task attribute, the at least one task attribute, and the corresponding first task model, that the inference task is the unknown task, recited in claim 11. The term "determiner" is a generic placeholder, and the modifier "task" identifies its task rather than implementing structure. The recitation "deployed in the edge side network" specifies where the placeholder is deployed rather than structure for performing the entire recited function. The claimed function is to determine, based on the target task attribute, the at least one task attribute, and the corresponding first task model, that the inference task is the unknown task. The corresponding structure is the processor 1602 of the device 1600 executing the program code of the task determining module 222 stored in the memory 1604, as described in [0206] and [0211]-[0213], programmed with the procedure of [0018] and [0078]-[0090], namely measuring the difference between the target task attribute and each stored task attribute as a minimum editing distance and comparing that difference against a threshold, and alternatively or additionally evaluating the confidence of inference on the input sample by a stored task model, a model migration rate, and task model quality against their respective thresholds. This limitation is interpreted to cover the foregoing structure and its disclosed procedure, and equivalents thereof.
(5) a knowledge base updater deployed in a cloud or an edge side network and configured to update, based on the target task attribute and the inference model, the at least one task attribute and the corresponding first task model, recited in claim 16. The term "updater" is a generic placeholder, and the modifier "knowledge base" identifies its task rather than implementing structure. The recitation "deployed in a cloud or an edge side network" specifies where the placeholder is deployed rather than structure for performing the entire recited function. The claimed function is to update, based on the target task attribute and the inference model, the at least one task attribute and the corresponding first task model. The corresponding structure is the processor 1602 of the device 1600 executing the program code of the knowledge base update module 225 stored in the memory 1604, as described in [0206] and [0211]-[0213], programmed with the procedure of [0023] and [0104]-[0110], namely adding the attribute value of the target task attribute and the inference model to the knowledge base when the attribute value differs from that of an existing stored task attribute, and alternatively replacing the corresponding stored task attribute and task model with the target task attribute and the inference model, together with the associated training samples, tasks, task relationships, and task groups. This limitation is interpreted to cover the foregoing structure and its disclosed procedure, and equivalents thereof.
Because these claim limitations are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, they are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
Claim Rejections - 35 U.S.C. 112(b)
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.
Claim 17 is rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding claim 17, the claim recites wherein when the inference model comprises a plurality of models, and wherein the task processing apparatus is further configured to, followed by the two operations of performing inference on the input sample using all models of the plurality of models to obtain inference results that are output by all the models of the plurality of models and determining the target inference result from the inference results. The quoted language is indefinite. The clause introduced by "wherein when" states a condition, but it is joined to the following "wherein" clause by the conjunction "and", so the conditional clause is left with no consequent of its own and the claim is open to two readings of materially different scope.
Under the first reading, the words "when the inference model comprises a plurality of models" state a condition on the two operations that follow, so that claim 17 does not require the inference model to comprise a plurality of models and requires the two operations only when that condition is satisfied. Under the second reading, the conjunction "and" makes the two "wherein" clauses coordinate, so that the inference model is positively required to comprise a plurality of models and the two operations are required unconditionally. Neither claim 17, nor claim 1 from which it depends, nor the remainder of the claim set selects between these readings, and the two readings are not of the same scope. A person of ordinary skill in the art is therefore not informed with reasonable certainty of the scope of claim 17. See Nautilus, Inc. v. Biosig Instruments, Inc., 572 U.S. 898, 901, 910 (2014); In re Packard, 751 F.3d 1307 (Fed. Cir. 2014); MPEP 2173.02.
For purposes of examination the quoted limitation is interpreted as conditional: claim 17 does not require that the inference model comprise a plurality of models, and it requires only that the task processing apparatus be configured, in the case where the inference model comprises a plurality of models, to perform inference on the input sample using all of those models and to determine the target inference result from the resulting inference results. That interpretation is grounded in the specification at paragraphs [0024], [0038] and [0199], which describe performing inference on the input sample using all the models and determining the target inference result from the inference results that are output by all the models when the inference model includes a plurality of models, and describe using the output of the single model as the final inference result when the inference model includes one model. The specification informs this interpretation but is not read into the claim. See MPEP 2111.
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 1-3, 5-6, 8-9, 13-14, and 18-20 are rejected under 35 U.S.C. 103 as being unpatentable over Ghani et al. (hereinafter Ghani), US 2013/0018825 A1, in view of Aljundi et al. (hereinafter Aljundi), "Expert Gate: Lifelong Learning with a Network of Experts" (2016).
Regarding independent claim 1, Ghani teaches a task learning system, comprising (Ghani: Abstract and [0030], "The apparatus 500 comprises a statistical determination component 502 operatively connected to a storage component 504 and a comparator 512"; the apparatus 500 (a task learning system) is the machine learning arrangement that settles which of a plurality of learned models is to serve as the basis for a model built for a new domain): a knowledge base, implemented in a hardware memory, configured to store first task attributes and task models corresponding to the first task attributes (Ghani: [0031], "this same statistical characterization is performed by the statistical determination component 502 on the prior text 506 used to establish each of the prior learned models stored in the storage component 504"; the storage component 504 (a knowledge base configured to store) holds the prior learned models (task models corresponding to) together with the statistical characterization of the prior text 506 (first task attributes) that established each of them; [0020], "The device 300 may be used to implement, for example, the processing illustrated below with regard to FIG. 4 and/or to implement one or more components of the apparatus 500 illustrated in FIG. 5"; the storage component 504 is one of the components of the apparatus 500 that the processing device 300 implements, so the store holding that content is the storage component 304; [0020], "the storage component 304 may include one or more devices such as volatile or nonvolatile memory including but not limited to random access memory (RAM) or read only memory (ROM)"; that storage is random access memory or read only memory (implemented in a hardware memory)); and a task processing apparatus, comprising at least one processor, coupled to the knowledge base, and configured to (Ghani: [0020], "the processing device 300 includes a processor 302 coupled to a storage component 304"; the processing device 300 (a task processing apparatus) carries a processor 302 (at least one processor) coupled to the storage component 304 that implements the storage component 504 mapped above (a knowledge base)) : obtain an input sample (Ghani: [0031], "new text input (corresponding to a new domain, as described above) is provided to the statistical determination component 502"; the new domain input (an input sample) is provided to the component that takes it in for processing); generate, when an inference task corresponding to the input sample is an unknown task, an inference model for the unknown task based on at least one task attribute of the first task attributes and a corresponding first task model of the task models (Ghani: [0019], "There may be instances, however, where none of the existing learned models in the taxonomy 100, 200 represents an identical match (or even nearly so) for the domain of a given set of input data"; the domain of the given set of input data (an inference task) for which no existing learned model is an identical match is the new domain (is an unknown task); [0019], "the instant disclosure describes techniques for identifying such existing learned models to serve as the basis for a new domain model"; [0031], "new text input (corresponding to a new domain, as described above) is provided to the statistical determination component 502"; the processing of FIG. 5 is run on that new domain (when an inference task corresponding to the input sample is an unknown task); [0027], "at least one learned model is identified as the basis for the new domain model when the new feature statistical characteristics compare favorably with the statistical characteristics of features in prior input data corresponding to the at least one learned model", [0032], "the output of the comparator 512 is one or more identifications of learned models to serve as the basis for the new domain model"; the stored model is singled out by the statistical characteristics of its own prior input (at least one task attribute of the first task attributes); [0033], "In addition to the identification(s) of the at least one learned model identified by the comparator 512, the training component 516 also takes as input the at least one learned model 518 thus identified", [0033], "the training component 516 creates the new domain model 520 by augmenting (or retraining) the at least one learned model 518 based on the new text"; the new domain model 520 (an inference model) is created (generate) from the identification produced by that attribute comparison (based on at least one task attribute) and from the learned model 518 so identified (a corresponding first task model)), wherein the at least one task attribute corresponds to the input sample (Ghani: [0027], "at least one learned model is identified as the basis for the new domain model when the new feature statistical characteristics compare favorably with the statistical characteristics of features in prior input data corresponding to the at least one learned model"; the statistical characteristics of the prior input (wherein the at least one task attribute) that single out the identified model are the ones that compare favorably with the characteristics taken from the new domain input, so they are the ones that answer to that input (corresponds to the input sample)), and wherein the inference model is a machine learning model (Ghani: [0033], "Using whatever learning algorithm that has been previously employed to develop the plurality of learned models stored in the storage component 504"; the model built for the new domain (and wherein the inference model) is produced by the same learning algorithm that produced the stored learned models (machine learning models)).
Ghani does not expressly teach and perform inference on the input sample using the inference model to obtain a target inference result.
However, Aljundi teaches and perform inference on the input sample using the inference model to obtain a target inference result. (Aljundi: page 1, Abstract, "We introduce a gating autoencoder that learns a representation for the task at hand, and is used at test time to automatically forward the test sample to the relevant expert"; the test sample (an input sample) is itself routed to the expert model (an inference model) that the gate selects for it; page 5, Algorithm 1 sets out the test phase of the Expert Gate system as taking the sample as its input and returning a prediction as its output, the expert being selected from that sample and then activated on that same sample; the prediction so returned (a target inference result) is the output of the selected expert model operating on the very sample that selected it).
Because Ghani and Aljundi are analogous art with both addressing the selection and reuse of previously learned models when a machine learning system meets data belonging to a task it has not yet modelled, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Aljundi's test-time gating deployment to Ghani's task learning system apparatus 500, whose newly built new domain model 520 is returned to the store from which the models are drawn (Ghani: [0033], "the resulting new domain model may then be provided back to the storage component 504 for later recall"), with a reasonable expectation of success, so that the new domain input is forwarded to the model that answers to it, to teach and perform inference on the input sample using the inference model to obtain a target inference result. This modification would have been motivated by the desire to forward the sample at hand automatically to the model relevant to it while keeping only one expert network loaded into memory at any given time (Aljundi: page 1, Abstract).
Regarding dependent claim 2, Ghani, in view of Aljundi, teach the task learning system of claim 1, wherein the task processing apparatus is further configured to: determine a target task attribute of the input sample based on the input sample (Ghani: [0031], "the statistical determination component 502 establishes a distribution of the relative frequencies of words and/or phrases (i.e., features) within the new text"; the statistical determination component 502, one of the components of the apparatus 500 that the processing device 300 (the task processing apparatus) implements, establishes the new feature statistical characteristics (determine a target task attribute) from the features of the new domain input (of the input sample) itself (based on the input sample)) and a subset of the first task attributes that correspond to the input sample (Ghani: [0026], "it may be desirable to compare against only some of the available plurality of learned models"; [0032], "The statistical determination component 502 provides the statistical characteristics of the new text 508 (i.e., the new features statistical characteristics) as well as the statistical characteristics of the features in the prior text 506 for each of the learned models being compared to the comparator 512"; the new feature statistical characteristics (the target task attribute) are established in the same statistical characterization as, and are provided together with, the statistical characteristics of features in the prior input held for only those learned models taken for comparison with the new domain input (a subset of the first task attributes that correspond to the input sample), which are among the statistical characteristics of features in the prior input (the first task attributes) held in the store); and determine, based on the target task attribute, the at least one task attribute, and the corresponding first task model, that the inference task is the unknown task. (Ghani: [0032], "the comparator 512 may obtain identifications 513 of the learned models corresponding to the statistical characteristics being compared, thereby allowing the comparator 512 to identify which of the learned models being compared results in favorable comparisons"; the comparator 512 compares the new feature statistical characteristics (the target task attribute) with the statistical characteristics of features in the prior input (the at least one task attribute) of the learned model 518 (the corresponding first task model) that it identifies; [0027], "a favorable comparison results when the similarity of compared statistical characteristics exceeds a desired threshold according to a suitable metric"; [0019], "There may be instances, however, where none of the existing learned models in the taxonomy 100, 200 represents an identical match (or even nearly so) for the domain of a given set of input data"; that thresholded comparison is what establishes that no existing learned model is an identical match for the domain of a given set of input data (the inference task), that domain being the new domain (the unknown task))
Regarding dependent claim 3, Ghani, in view of Aljundi, teach the task learning system of claim 2, wherein the task processing apparatus is further configured to determine, based on a difference between the target task attribute and the at least one task attribute, that the inference task is the unknown task (Ghani: [0027], "increasingly lower distances or differences as evaluated by the metric correspond to increasingly favorable comparisons"; the comparator 512, one of the components of the apparatus 500 that the processing device 300 (the task processing apparatus) implements, evaluates the distance or difference between the new feature statistical characteristics (the target task attribute) and the statistical characteristics of features in the prior input (the at least one task attribute) of the learned model 518 it identifies; [0019], "There may be instances, however, where none of the existing learned models in the taxonomy 100, 200 represents an identical match (or even nearly so) for the domain of a given set of input data"; that distance or difference is what shows that no existing learned model is an identical match for the domain of a given set of input data (the inference task), that domain being the new domain (the unknown task)).
Regarding dependent claim 5, Ghani, in view of Aljundi, teach the task learning system of claim 2, wherein the task processing apparatus is further configured to generate the inference model for the unknown task (Ghani: [0033], "the training component 516 creates the new domain model 520 by augmenting (or retraining) the at least one learned model 518 based on the new text"; the training component 516, one of the components of the apparatus 500 that the processing device 300 (the task processing apparatus) implements, creates the new domain model 520 (the inference model) for the new domain (the unknown task)) based on the target task attribute, the at least one task attribute, and the corresponding first task model (Ghani: [0032], "the output of the comparator 512 is one or more identifications of learned models to serve as the basis for the new domain model"; [0033], "In addition to the identification(s) of the at least one learned model identified by the comparator 512, the training component 516 also takes as input the at least one learned model 518 thus identified"; the new domain model 520 is created from the identification that the comparator 512 produces by comparing the new feature statistical characteristics (the target task attribute) with the statistical characteristics of features in the prior input (the at least one task attribute), and from the learned model 518 (the corresponding first task model) so identified).
Regarding dependent claim 6, Ghani, in view of Aljundi, teach the task learning system of claim 5, wherein the knowledge base is further configured to store a task relationship that comprises one or more of a subordinate relationship or a migration relationship (Ghani: [0016], "at least some of the plurality of learned models (e.g., S1, S2, S3) may be aggregated to form another learned model (e.g., S7) based on the subjects associated with the at least some of the plurality of sentiment classifiers"; the tree taxonomy 100 (a task relationship) records, for each learned model, the more general learned model into which it is aggregated, which is a subordinate relationship; [0018], "the various taxonomies 100, 200 may be used to identify which of the plurality of learned models to use for any new input data to be analyzed"; [0020], "The storage component 304, in turn, includes stored executable instructions 316 and data 318"; the taxonomy the apparatus consults to identify learned models is data 318 held in the storage component 304 that implements the storage component 504 (the knowledge base)), and wherein the task processing apparatus is further configured to generate the inference model based on the target task attribute, the first task attributes, the task models, and the task relationship (Ghani: [0019], "domain may be provided if it is based on one of the plurality of learned models already within the taxonomy 100, 200, or even combinations of such models or subsets of such models"; the new domain model 520 (the inference model) that the processing device 300 (the task processing apparatus) creates is based on learned models drawn from the plurality of learned models (the task models) as organized in the tree taxonomy 100 (the task relationship); [0032], "The statistical determination component 502 provides the statistical characteristics of the new text 508 (i.e., the new features statistical characteristics) as well as the statistical characteristics of the features in the prior text 506 for each of the learned models being compared to the comparator 512"; the learned model so drawn is the one identified by comparing the new feature statistical characteristics (the target task attribute) with the statistical characteristics of features in the prior input (the first task attributes)).
Regarding dependent claim 8, Ghani, in view of Aljundi, teach the task learning system of claim 5, wherein the knowledge base is further configured to store first training samples corresponding to the first task attributes (Ghani: [0031], "this same statistical characterization is performed by the statistical determination component 502 on the prior text 506 used to establish each of the prior learned models stored in the storage component 504"; the storage component 504 (the knowledge base) holds, with each stored learned model, the prior text 506 (first training samples) used to establish it, and the statistical characteristics of features in the prior input (the first task attributes) are taken from that same prior text 506; [0033], "the new domain model 520 is associated with the new text within the storage model, i.e., the new text used to establish the new domain model 520 is essentially now the prior text"; the text that establishes a model is kept in the store as the prior text 506 of that model), and wherein the task processing apparatus is further configured to: determine, based on the target task attribute, second training samples of the first training samples respectively corresponding to a plurality of candidate task models in the task models (Ghani: [0027], "where there may be multiple learned models giving rise to distance or difference values below a threshold, then it may be desirable to combine the multiple learned models to serve as the basis for the new domain model"; the comparator 512, one of the components of the apparatus 500 that the processing device 300 (the task processing apparatus) implements, picks out of the plurality of learned models (the task models) the multiple learned models (a plurality of candidate task models) whose distance or difference from the new feature statistical characteristics (the target task attribute) falls below the threshold, and so determines the prior text 506 held for each of the multiple learned models (second training samples)); retrain one or more candidate task models of the plurality of candidate task models based on the second training samples (Ghani: [0028], "where the learned models are developed according to a supervised or semi-supervised algorithm, the new domain input may be used as additional training data to re-train the at least one learned model"; the identified learned models (one or more candidate task models) are re-trained with the new domain input used as training data additional to the prior text 506 held for each of the multiple learned models (the second training samples)); and use, as the inference model, the one or more candidate task models. (Ghani: [0028], "The resulting, retrained model is then provided as the new domain model"; once retrained, the identified learned models (the one or more candidate task models) are what is provided as the new domain model 520 (the inference model)).
Regarding dependent claim 9, Ghani, in view of Aljundi, teach the task learning system of claim 5, wherein the knowledge base is further configured to store first training samples corresponding to the first task attributes (Ghani: [0031], "this same statistical characterization is performed by the statistical determination component 502 on the prior text 506 used to establish each of the prior learned models stored in the storage component 504"; the storage component 504 (the knowledge base) holds, with each stored learned model, the prior text 506 (first training samples) used to establish it, and the statistical characteristics of features in the prior input (the first task attributes) are taken from that same prior text 506; [0033], "the new domain model 520 is associated with the new text within the storage model, i.e., the new text used to establish the new domain model 520 is essentially now the prior text"; the text that establishes a model is kept in the store as the prior text 506 of that model), and wherein the task processing apparatus is further configured to: determine, based on the target task attribute, second training samples of the first training samples respectively corresponding to a plurality of candidate task models in the task models (Ghani: [0027], "where there may be multiple learned models giving rise to distance or difference values below a threshold, then it may be desirable to combine the multiple learned models to serve as the basis for the new domain model"; the comparator 512, one of the components of the apparatus 500 that the processing device 300 (the task processing apparatus) implements, picks out of the plurality of learned models (the task models) the multiple learned models (a plurality of candidate task models) whose distance or difference from the new feature statistical characteristics (the target task attribute) falls below the threshold, and so determines the prior text 506 held for each of the multiple learned models (second training samples)); perform training based on the second training samples to obtain a new task model (Ghani: [0028], "where the learned models are developed according to a supervised or semi-supervised algorithm, the new domain input may be used as additional training data to re-train the at least one learned model"; the training is taken on the new domain input as training data additional to the prior text 506 held for each of the multiple learned models (the second training samples); [0033], "the training component 516 creates the new domain model 520 by augmenting (or retraining) the at least one learned model 518 based on the new text"; what that training produces is the new domain model 520 (a new task model)); and use the new task model as the inference model (Ghani: [0036], "the new domain model 520 is then used by the analysis component 522 to analyze additional new text to provide analysis results"; the new domain model 520 (the new task model) is itself the new domain model 520 (the inference model) that the analysis component 522 thereafter applies).
Regarding dependent claim 13, Ghani, in view of Aljundi, teach the task learning system of claim 1, wherein the task processing apparatus is further configured to update, based on a target task attribute of the input sample and the inference model (Ghani: [0031], "the statistical determination component 502 establishes a distribution of the relative frequencies of words and/or phrases (i.e., features) within the new text"; [0033], "the new domain model 520 is associated with the new text within the storage model, i.e., the new text used to establish the new domain model 520 is essentially now the prior text"; the apparatus 500, implemented by the processing device 300 (the task processing apparatus), revises its store on the basis of the new feature statistical characteristics (a target task attribute) established from the new domain input (the input sample) and of the new domain model 520 (the inference model) established from that input), one or more task attributes that are stored in the knowledge base (Ghani: [0024], "The determination of the statistical characteristics of the prior input data for each of the plurality of learned models may be done at time of creation of a given learned model"; once the new text is the prior text of the new domain model 520, the statistical characteristics of features in the prior input for that model, which are the new feature statistical characteristics, are determined at its creation and join the statistical characteristics held in the storage component 504 (one or more task attributes that are stored in the knowledge base)) and one or more task models that are stored in the knowledge base (Ghani: [0033], "the resulting new domain model may then be provided back to the storage component 504 for later recall"; the new domain model 520 is written into the storage component 504, changing the plurality of learned models held in the storage component 504 (one or more task models that are stored in the knowledge base)).
Regarding dependent claim 14, Ghani, in view of Aljundi, teach the task learning system of claim 13, wherein the task processing apparatus is further configured to add the target task attribute and the inference model to the knowledge base (Ghani: [0033], "the new text used to establish the new domain model 520 is essentially now the prior text"; [0024], "The determination of the statistical characteristics of the prior input data for each of the plurality of learned models may be done at time of creation of a given learned model"; the new feature statistical characteristics (the target task attribute), determined at the creation of the new domain model 520 from what is now its prior text, join the statistical characteristics held in the storage component 504 (the knowledge base); [0033], "the resulting new domain model may then be provided back to the storage component 504 for later recall"; the new domain model 520 (the inference model) is added to that same store by the processing device 300 (the task processing apparatus)), and wherein the knowledge base is further configured to simultaneously store the first task attributes, the target task attribute, the task models, and the inference model (Ghani: [0031], "this same statistical characterization is performed by the statistical determination component 502 on the prior text 506 used to establish each of the prior learned models stored in the storage component 504"; the storage component 504 (the knowledge base) continues to hold the plurality of learned models (the task models) and the statistical characteristics of features in the prior input (the first task attributes) while the new domain model 520 (the inference model) and the new feature statistical characteristics (the target task attribute) are added to it, so all of them are held at once).
Regarding claims 18-20, these are task learning method claims that are substantially the same as the task learning system of claims 1-3, respectively. Thus, claims 18-20 are rejected for the same reasons as claims 1-3.
Claims 7 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Ghani in view of Aljundi, as applied in the rejections of claims 5 and 1, respectively above, and further in view of Watson et al. (hereinafter Watson), US 2019/0354850 A1.
Regarding dependent claim 7, Ghani, in view of Aljundi, teach the task learning system of claim 5, wherein the task processing apparatus is further configured to: determine a plurality of candidate task models in the task models based on the target task attribute (Ghani: [0027], "where there may be multiple learned models giving rise to distance or difference values below a threshold, then it may be desirable to combine the multiple learned models to serve as the basis for the new domain model"; the comparator 512, one of the components of the apparatus 500 that the processing device 300 (the task processing apparatus) implements, picks out of the plurality of learned models (the task models) the multiple learned models (a plurality of candidate task models) whose distance or difference from the new feature statistical characteristics (the target task attribute) falls below the threshold).
Ghani and Aljundi do not expressly teach and use the plurality of candidate task models as the inference model.
However, Watson teaches and use the plurality of candidate task models as the inference model (Watson: [0080], "Wherein computational resources are available to train multiple models transferred from different sources, an ensemble was constructed. To compute the prediction of the ensemble, the scores of the models were averaged"; the multiple models transferred from different sources (the plurality of candidate task models) are held together as the ensemble (the inference model), and it is the ensemble, rather than any single member, that produces the prediction).
Because Ghani, in view of Aljundi, and Watson are analogous art with all three addressing the reuse of previously trained models for a machine learning task the system has not yet modelled, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Watson's ensemble construction to the task learning system of Ghani and the several learned models whose distance values fall below its threshold, with a reasonable expectation of success, keeping the several learned models whose distance values fall below the threshold as the operative model rather than merging them into the single new domain model Ghani otherwise creates from them, to teach and use the plurality of candidate task models as the inference model. This modification would have been motivated by the desire to obtain the best performance available from the candidate models (Watson: [0081]).
Regarding dependent claim 17, Ghani, in view of Aljundi, teach the task learning system of claim 1, wherein when the inference model comprises a plurality of models, [[and wherein]] (interpreted per the 35 U.S.C. 112(b) rejection set forth above) the task processing apparatus is further configured to (Ghani: [0027], "where there may be multiple learned models giving rise to distance or difference values below a threshold, then it may be desirable to combine the multiple learned models to serve as the basis for the new domain model"; several learned models at once, the multiple learned models (a plurality of models), may be taken as the basis for the new domain model 520 (the inference model), which is the case the recited condition addresses, and the processing device 300 (the task processing apparatus) is the apparatus so configured).
Ghani and Aljundi do not expressly teach perform inference on the input sample using all models of the plurality of models to obtain inference results that are output by all the models of the plurality of models; and determine the target inference result from the inference results.
However, Watson teaches perform inference on the input sample using all models of the plurality of models to obtain inference results that are output by all the models of the plurality of models; and determine the target inference result from the inference results (Watson: [0080], "Wherein computational resources are available to train multiple models transferred from different sources, an ensemble was constructed. To compute the prediction of the ensemble, the scores of the models were averaged."; every model in the ensemble (all models of the plurality of models) is run and returns a score, the scores of the models (inference results that are output by all the models of the plurality of models) are averaged, and the prediction of the ensemble (the target inference result) is computed from them).
Because Ghani, in view of Aljundi, and Watson are analogous art with all three addressing the reuse of previously trained models for a machine learning task the system has not yet modelled, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Watson's score-averaging ensemble prediction to the task learning system of Ghani in the case where the model it builds rests on several learned models, with a reasonable expectation of success, running each of those models on the sample and reducing the outputs they return to the one result reported, to teach perform inference on the input sample using all models of the plurality of models to obtain inference results that are output by all the models of the plurality of models; and determine the target inference result from the inference results. This modification would have been motivated by the desire to obtain the best performance available from the models taken together (Watson: [0081]).
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Ghani in view of Aljundi and further in view of Ghanta et al. (hereinafter Ghanta), US 2019/0377984 A1.
Regarding dependent claim 4, Ghani, in view of Aljundi, teach the task learning system of claim 2, wherein the task processing apparatus is further configured to determine, (Ghani: [0020], "The device 300 may be used to implement, for example, the processing illustrated below with regard to FIG. 4 and/or to implement one or more components of the apparatus 500 illustrated in FIG. 5"; the processing device 300 (the task processing apparatus) implements the comparator 512 by which the determination is made) that the inference task is the unknown task (Ghani: [0027], "a favorable comparison results when the similarity of compared statistical characteristics exceeds a desired threshold according to a suitable metric"; the comparator 512 tests the compared statistical characteristics against a threshold; [0019], "There may be instances, however, where none of the existing learned models in the taxonomy 100, 200 represents an identical match (or even nearly so) for the domain of a given set of input data"; that test is what establishes that no existing learned model is an identical match for the domain of a given set of input data (the inference task), that domain being the new domain (the unknown task)).
Ghani and Aljundi do not expressly teach that the determination that the inference task is the unknown task is made based on any one or more of a confidence of performing inference on the input sample using each task model of the task models, a model migration rate, or a task model quality of the task models.
However, Ghanta teaches based on any one or more of a confidence of performing inference on the input sample using each task model of the task models, a model migration rate, or a task model quality of the task models (Ghanta: [0118], "the score module 306 calculates 406 a suitability score describing the suitability of the training data set to the inference data set as a function of the first and the second statistical data signatures"; the suitability score (a model migration rate) reports how suitable the data a machine learning model was trained on is to the data now to be analyzed, which is a measure of the appropriateness of performing inference on that data using that model; [0115], "the action module 308 sends an alert, message, notification, or the like (e.g., to an administrator or other user) that indicates that the training data set and machine learning model that was trained using the training data set is unsuitable for the inference data set in response to the suitability score not satisfying a suitability threshold"; that score is tested against a threshold, and a score that fails the threshold establishes that the model held for that data is not suitable for it). Because Ghani, in view of Aljundi, and Ghanta are analogous art with all three addressing whether a machine learning model the system already holds fits the data now presented to it, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Ghanta's suitability score and threshold test to Ghani's comparison of the new feature statistical characteristics against the statistical characteristics held for each of the plurality of learned models, with a reasonable expectation of success, so that the determination Ghani makes at block 406 is taken on a score reporting directly whether the data a stored model was trained on suits the data now to be analyzed, to teach based on any one or more of a confidence of performing inference on the input sample using each task model of the task models, a model migration rate, or a task model quality of the task models. This modification would have been motivated by the desire to monitor the suitability of a machine learning model, trained using a training data set, for an inference data set (Ghanta: [0001]).
Claims 10-12 and 15-16 are rejected under 35 U.S.C. 103 as being unpatentable over Ghani in view of Aljundi and further in view of Sharma et al. (hereinafter Sharma), US 2020/0327371 A1.
Regarding dependent claim 10, Ghani, in view of Aljundi, teach the task learning system of claim 1, wherein the task processing apparatus comprises (Ghani: [0020], "The device 300 may be used to implement, for example, the processing illustrated below with regard to FIG. 4 and/or to implement one or more components of the apparatus 500 illustrated in FIG. 5"; the processing device 300 (the task processing apparatus) implements the components of the apparatus 500 that carry out the two functions mapped below): configured to generate, when the inference task corresponding to the input sample is the unknown task, the inference model based on the at least one task attribute and the corresponding first task model (Ghani: [0031], "new text input (corresponding to a new domain, as described above) is provided to the statistical determination component 502"; [0033], "the training component 516 creates the new domain model 520 by augmenting (or retraining) the at least one learned model 518 based on the new text"; for the new domain input (the input sample), the domain of a given set of input data (the inference task) being the new domain (the unknown task), the training component 516 creates the new domain model 520 (the inference model) from the learned model 518 (the corresponding first task model) that the comparator 512 identified on the statistical characteristics of features in the prior input (the at least one task attribute)); configured to perform inference on the input sample using the inference model to obtain the target inference result (Aljundi: page 1, Abstract, "We introduce a gating autoencoder that learns a representation for the task at hand, and is used at test time to automatically forward the test sample to the relevant expert"; as set forth for claim 1, the test sample (the input sample) is forwarded to the relevant expert (the inference model), and the prediction that expert returns is the target inference result).
Ghani and Aljundi do not expressly teach that those two functions are carried out by components sited apart from one another, namely a model determiner deployed in a cloud or an edge side network and an inference performer deployed in the edge side network.
However, Sharma teaches a model determiner deployed in a cloud or an edge side network (Sharma: [0180], "a machine learning model to be deployed to and executed on the example edge platform 406, 609 may be suitably developed and trained in the cloud 412 using a model creation component 902, model training component 904, and data storage and aggregation component 573"; the model creation component 902 and model training component 904 (a model determiner) sit in the cloud 412 (a cloud) and build the model there out of the data the data storage and aggregation component 573 holds) and an inference performer deployed in the edge side network (Sharma: [0013], "Without first transmitting the first sensor data stream to the remote cloud network for processing, the machine learning model operates on the first sensor data stream and produces a stream of first inferences about a first network device in real-time"; the machine learning model on the edge computing platform (an inference performer) runs on the edge computing platform (the edge side network) and produces its inferences there without the data being sent anywhere else).
Because Ghani, in view of Aljundi, and Sharma are analogous art with all three addressing the building and the deployment of machine learning models across the parts of a distributed system, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Sharma's split of model creation from model execution to the task learning system of Ghani, with a reasonable expectation of success, siting the component that builds the model for the new domain on the cloud side and the component that runs it on the edge side where the samples arrive, to teach a model determiner deployed in a cloud or an edge side network and an inference performer deployed in the edge side network. This modification would have been motivated by the desire to produce inferences from the arriving data in real time without first transmitting that data to the remote cloud network (Sharma: [0013]).
Regarding dependent claim 11, Ghani, in view of Aljundi, teach the task learning system of claim 1, wherein the task processing apparatus further comprises (Ghani: [0020], "The device 300 may be used to implement, for example, the processing illustrated below with regard to FIG. 4 and/or to implement one or more components of the apparatus 500 illustrated in FIG. 5"; the processing device 300 (the task processing apparatus) implements the components of the apparatus 500 that make the two determinations mapped below): configured to determine a target task attribute of the input sample based on the input sample and the at least one task attribute (Ghani: [0031], "the statistical determination component 502 establishes a distribution of the relative frequencies of words and/or phrases (i.e., features) within the new text"; [0032], "The statistical determination component 502 provides the statistical characteristics of the new text 508 (i.e., the new features statistical characteristics) as well as the statistical characteristics of the features in the prior text 506 for each of the learned models being compared to the comparator 512"; the statistical determination component 502 establishes the new feature statistical characteristics (a target task attribute) from the features of the new domain input (the input sample), in the same statistical characterization as, and together with, the statistical characteristics of features in the prior input (the at least one task attribute)); configured to determine, based on the target task attribute, the at least one task attribute, and the corresponding first task model, that the inference task is the unknown task (Ghani: [0032], "the comparator 512 may obtain identifications 513 of the learned models corresponding to the statistical characteristics being compared, thereby allowing the comparator 512 to identify which of the learned models being compared results in favorable comparisons"; the comparator 512 compares the new feature statistical characteristics (the target task attribute) with the statistical characteristics of features in the prior input (the at least one task attribute) of the learned model 518 (the corresponding first task model) that it identifies; [0027], "a favorable comparison results when the similarity of compared statistical characteristics exceeds a desired threshold according to a suitable metric"; [0019], "There may be instances, however, where none of the existing learned models in the taxonomy 100, 200 represents an identical match (or even nearly so) for the domain of a given set of input data"; that thresholded comparison is what establishes that no existing learned model is an identical match for the domain of a given set of input data (the inference task), that domain being the new domain (the unknown task)).
Ghani and Aljundi do not expressly teach that those two determinations are made at the edge, namely by an attribute determiner deployed in an edge side network and a task determiner deployed in the edge side network.
However, Sharma teaches an attribute determiner deployed in an edge side network and a task determiner deployed in the edge side network (Sharma: [0234], "selected analytics expressions implementing logic, mathematical, statistical, or other functions, or a combination, may be executed in the CEP of the edge platform on a stream of inferences generated by a model"; the statistical functions that compute the measure are executed in the CEP of the edge platform, as the site at which the computation is executed (an attribute determiner), on the edge computing platform (an edge side network) itself, rather than at the site that built the model; [0023], "The analytics expressions define what constitutes an unacceptable level of drift or degradation of model accuracy and track the model output to determine if the accuracy has degraded beyond an acceptable limit."; those same edge-resident expressions carry the standard the deployed model is judged against and make that judgment at the edge platform, as the site at which the determination is made (a task determiner)).
Because Ghani, in view of Aljundi, and Sharma are analogous art with all three addressing whether a deployed machine learning model still fits the data a distributed system is feeding it, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Sharma's edge-resident analytics expressions to the task learning system of Ghani, with a reasonable expectation of success, siting at the edge platform, rather than at the site that built the model, both the computation of the characterization of the arriving sample and the test of that characterization against what is held for the stored models, to teach an attribute determiner deployed in an edge side network and a task determiner deployed in the edge side network. This modification would have been motivated by the desire to detect at the edge, from the model's own output stream, when the accuracy has degraded beyond an acceptable limit (Sharma: [0023]).
Regarding dependent claim 12, Ghani, in view of Aljundi, teach all the elements of claim 1.
Ghani and Aljundi do not expressly teach wherein the knowledge base is deployed in a cloud.
However, Sharma teaches wherein the knowledge base is deployed in a cloud (Sharma: [0180], "a machine learning model to be deployed to and executed on the example edge platform 406, 609 may be suitably developed and trained in the cloud 412 using a model creation component 902, model training component 904, and data storage and aggregation component 573"; the data storage and aggregation component 573 in the cloud 412 (the knowledge base) holds the data the models are built from, the cloud 412 (a cloud) being where the models are developed and trained; [0235], "On the cloud platform 412, the transferred predictions, inferences, data and analytics results, and other information can be aggregated in cloud storage 573."; the store in which that data is held is the cloud storage 573 on the cloud platform 412).
Because Ghani, in view of Aljundi, and Sharma are analogous art with all three addressing where in a distributed system the models and the data that characterize them are held, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Sharma's cloud-side siting of the data store to the task learning system of Ghani, with a reasonable expectation of success, keeping the store of models and of the characteristics held for each on the cloud side, to teach wherein the knowledge base is deployed in a cloud. This modification would have been motivated by the desire to carry out model creation and training where significant compute and storage resources are available (Sharma: [0159]).
Regarding dependent claim 15, Ghani, in view of Aljundi, teach the task learning system of claim 13, wherein the task processing apparatus is further configured to: replace a corresponding task attribute in the knowledge base with the target task attribute (Ghani: [0020], "the processing device 300 includes a processor 302 coupled to a storage component 304", [0031], "the statistical determination component 502 establishes a distribution of the relative frequencies of words and/or phrases (i.e., features) within the new text", [0024], "The determination of the statistical characteristics of the prior input data for each of the plurality of learned models may be done at time of creation of a given learned model", [0033], "the new domain model 520 is associated with the new text within the storage model, i.e., the new text used to establish the new domain model 520 is essentially now the prior text"; the apparatus 500, implemented by the processing device 300 (the task processing apparatus), establishes new text associated with the new domain by the new feature statistical characteristics (with the target task attribute) to replace the prior text within the storage component (replace a corresponding task attribute in the knowledge base)).
Ghani and Aljundi do not expressly teach and replace a corresponding task model in the knowledge base with the inference model.
However, Sharma teaches replace a corresponding task model in the knowledge base with the inference model (Sharma: [0236], "The updated model or weight factors, or a combination, coefficient, or parameters are then re-deployed to the edge to replace the current model, weight factors, coefficients, or parameters"; [0022], "the model on the edge computing platform being updated or replaced by a modified edge-converted model from time to time"; the modified edge-converted model (the inference model) is put in the place of the current model (a corresponding task model), rather than set alongside it).
Because Ghani, in view of Aljundi, and Sharma are analogous art with all three addressing how a store of machine learning models and of the data characterizing them is carried forward as new data arrives, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Sharma's replacement of a superseded model by the model that updates it to the store the task learning system of Ghani maintains, with a reasonable expectation of success, writing the newly built model into the store in the place of the model it supersedes and, because that store holds one characterization for each model it holds, writing the newly computed characterization into the place of the characterization held for that model, the choice between overwriting a superseded entry and accumulating a further entry alongside it being a choice between two known and predictable alternatives, to teach wherein the task processing apparatus is further configured to: replace a corresponding task attribute in the knowledge base with the target task attribute; and replace a corresponding task model in the knowledge base with the inference model. This modification would have been motivated by the desire to have the inferences produced from the arriving data become more and more accurate over time as the model in use is updated or replaced (Sharma: [0022]).
Regarding dependent claim 16, Ghani, in view of Aljundi, teach the task learning system of claim 14, wherein the task processing apparatus comprises (Ghani: [0020], "The device 300 may be used to implement, for example, the processing illustrated below with regard to FIG. 4 and/or to implement one or more components of the apparatus 500 illustrated in FIG. 5"; the processing device 300 (the task processing apparatus) implements the components of the apparatus 500 that carry out the function mapped below) configured to update, based on the target task attribute and the inference model, the at least one task attribute and the corresponding first task model (Ghani: [0033], "the training component 516 creates the new domain model 520 by augmenting (or retraining) the at least one learned model 518 based on the new text"; the learned model 518 (the corresponding first task model), identified by comparing the new feature statistical characteristics (the target task attribute) with the stored statistical characteristics, is augmented or retrained into the new domain model 520 (the inference model); [0024], "where a given learned model is updated from time to time based on its prior input data, then the determination of the statistical characteristics for that given learned model may be determined each time the model is updated"; for the model so updated, the statistical characteristics of features in the prior input (the at least one task attribute) are determined anew when the model is updated).
Ghani and Aljundi do not expressly teach that the updating is carried out by a component sited apart from the rest, namely a knowledge base updater deployed in a cloud or an edge side network.
However, Sharma teaches a knowledge base updater deployed in a cloud or an edge side network (Sharma: [0235], "The transferred data and information in the cloud storage may then be used and operated on by the data mining, model update, and verification components 902, 904 to evaluate the accuracy of the edge-based model results and to further tune the edge-based model as necessary or desirable."; the data mining, model update, and verification components 902, 904 (a knowledge base updater) sit on the cloud 412 (a cloud) and work there on the data aggregated in the cloud storage 573 and on the edge-based model together).
Because Ghani, in view of Aljundi, and Sharma are analogous art with all three addressing the maintenance of a store of machine learning models across the parts of a distributed system, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Sharma's cloud-side model update component to the task learning system of Ghani, with a reasonable expectation of success, siting on the cloud side the component that revises both the stored characterization and the stored model, to teach wherein the task processing apparatus comprises a knowledge base updater deployed in a cloud or an edge side network configured to update, based on the target task attribute and the inference model, the at least one task attribute and the corresponding first task model. This modification would have been motivated by the desire to have the deployed models constantly learning and becoming better trained and more intelligent over time (Sharma: [0236]).
Response to Arguments
Applicant’s amended title is acknowledged and the objection to the title of the Specification set forth in the Office Action dated 3/18/2026 is hereby withdrawn.
Applicant’s amendments and remarks traversing the 35 U.S.C. 101 rejections are persuasive and the 35 U.S.C. 101 rejections set forth in the Office Action dated 3/18/2026 are hereby withdrawn.
Applicant’s amendments and remarks regarding the 35 U.S.C. 102/103 rejections set forth in the Office Action dated 3/18/2026 have been fully considered but are moot in view of the new grounds of rejection set forth above.
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 KUANG FU CHEN whose telephone number is (571)272-1393. The examiner can normally be reached M-F 9:00-5:30pm ET.
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, Jennifer Welch can be reached at (571) 272-7212. 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.
/KC CHEN/ Primary Patent Examiner, Art Unit 2143
Prosecution Insights