DETAILED ACTION
This Action is responsive to Claims filed 08/08/2023.
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 .
Priority
Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55.
Drawings
Receipt of Drawings filed 08/08/2023 is acknowledged. These Drawings are acceptable.
Status of the Claims
Claims 1-10 are currently pending.
Claim Objections
Claim 10 objected to because of the following informalities:
“…wherein an input variable of the input nodes comprise…” One of these words should be plural. “wherein input variables of the input nodes comprise…” or “…wherein an input variable of the input nodes comprises…”
Appropriate correction is required.
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.
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 limitation(s) is/are:
An input module (Claim 1)
An initialization module (Claim 1)
A screening module (Claim 1)
An adjustment module (Claim 1)
A cramming module (Claim 1)
A reorganization module (Claim 1)
A regularization module (Claim 8)
A determining module (Claim 8)
Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/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 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-10 rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claim(s) does/do not fall within at least one of the four categories of patent eligible subject matter because the recitation of "An adaptive learning algorithm..." lacks structure precluding the claimed program from being software per se (See MPEP 2106.03(I) second list).
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claim(s) 1-3 and 8 is/are rejected under 35 U.S.C. 103 as being unpatentable over Han et al. (An adaptive growing and pruning algorithm for designing recurrent neural network, 2017), hereinafter Han; Bengio et al. (Curriculum Learning, 2009), hereinafter Bengio; Mikler et al. (Comparing Rewinding and Fine-tuning in Neural Network Pruning Reproducibility Challenge 2021, Published April 2022), hereinafter Mikler; and Dai et al. (Incremental Learning Using a Grow-and-Prune Paradigm with Efficient Neural Networks, 2019), hereinafter Dai.
(Examiner’s Note: Given the highly general nature of some of the forthcoming limitations, the Examiner interprets many of these steps/limitations to be inherent to machine learning or adaptive learning systems, in the absence of more structural detail or implementation differentiating them from typical machine learning structures. Such limitations or features will be made known before a prior art mapping is made.)
In regards to claim 1: The present invention claims: “An adaptive learning algorithm, comprising the following steps: S1: providing all training data by an input module;” (Examiner’s Note: The “providing…” limitation is being interpreted broadly to indicate that training data is accessible to the adaptive learning algorithm, which the Examiner submits is inherent to the functioning of such an algorithm). Han teaches “…a recurrent self-organizing neural networks (RSONN), using an adaptive growing and pruning algorithm (AGPA), is proposed for im- proving their performance in this paper.” (Abstract, mapping to an adaptive learning algorithm).
“S2: performing a linear regression operation on first m+1 training data by an initialization module to establish an initial single-layer neural network, wherein the initial single-layer neural network comprises an initial weight parameter;” (Examiner’s note: Given the generality of the “performing a linear regression…” portion of the limitation, the Examiner interprets this broadly and as generally inherent to single-layer networks, as demonstrated in Brouwn et al. “The optimization and analysis of single-layer network models are tractable problems, since single-layer networks constitute linear regression models” (Abstract), included here not as part of the Rejection, but to demonstrate the well-known nature of the limitation) Han teaches a single hidden layer neural network (Fig. 1) with connection weights “An RNN has a set of neurons and connection weights between the neurons. The input neurons are set by the environment and the output neurons are computed using the connection weights and the hidden neurons” (Introduction)
“S3: selecting a quantity of training data corresponding to each round by a screening module using a selection mechanism which substitutes all the training data into an acceptable single-layer neural network obtained in a previous round for prediction, calculates a residual sum of squares between an actual value and a predicted value of each training data…” (Examiner’s Note: The Examiner interprets the “selecting…” portion of the limitation of S3 to mean selecting training data for input into a neural network; and the “calculates…” portion broadly as part of performing the linear regression on the training data in S2) Han reads on training an existing neural network (Fig. 3, first two steps of the flow chart).
“S4: substituting the selected training data into the initial single-layer neural network by the screening module, and determining whether a learning goal of the training data has been achieved, if yes, accepting the initial single-layer neural network and entering step S7, if no, continuing to step S5;” (Examiner’s Note: The Examiner interprets this portion of the limitation of S3 to mean inputting the selected training data into a neural network) Han reads on training an existing neural network (Fig. 3, first two steps of the flow chart). Han also teaches, in Fig. 3, the flow chart of their algorithm. The Examiner submits, based on the description of S7 and S8 below, that Han’s model would go from an initial state, into checking the need for a growth state (analogous to S5 and S6 below, Page 53), into checking the need for a pruning state (analogous to S7 below, Page 53).
“S5… adjusting the current neural network weight parameter by an adjustment module, and determining whether the adjusted neural network is in an acceptable state, and if the adjusted neural network is in an acceptable state, entering step S7, if not, continuing to step S6;” Han teaches “Step 2) For the input sample x( t ), the center, radius, feedback weights and output weights of RNN are adjusted using the adaptive second-order algorithm.” (Page 54) before the RNN entering a growth phase (Page 53), where new hidden neurons are added (analogous to S6 below). Fig. 3 shows a growth stage check, followed by a pruning phase check (analogous to moving to S7 below). Han also teaches “Step 3) Compute the competitiveness of hidden neurons as Eq. (5) . If the condition meets the growing phase criteria, go to step 4). If the condition meets the pruning phase criteria, go to step 5). Otherwise, go to step 6).” (Page 54),
“S6…adding three hidden nodes by a cramming module to obtain a newly accepted single-layer neural network as an acceptable single-layer neural network and enter step S7;” Han teaches the RNN entering a growth phase where a new hidden neuron is added to the hidden layer (Page 53 and Fig. 3), before entering a pruning phase (analogous to S7 below). Han also teaches “Step 4) Split the jth hidden neuron and insert new hidden neurons. The initial parameters of the new neurons can be given as Eq. (9) .” (Page 54)
“S7: accessing the acceptable single-layer neural network and a weight parameter by a reorganization module to check all hidden nodes in the network and deleting invalid nodes;” Han teaches the RNN going through a pruning phase after the growth phase, where redundant neurons are removed (Page 53 and Fig. 3)
“S8: returning to step S3 in which the acceptable single-layer neural network is used as an acceptable initial single-layer neural network for a next round…” Han Fig. 3 shows the method looping back after an structure adjustment (growth and pruning) phase is complete if termination criterion is not met, skipping the initial create of the RNN and instead using the adjusted RNN.
Han fails to explicitly teach:
“S3… performs sorting, and selects a corresponding quantity of training data which is sorted into an ascending order;” (Examiner’s Note: The Examiner interprets these portions of the limitation broadly as a form of curriculum learning, using the RSS as a difficulty metric, as further research and consideration seems to indicate this is the scenario where one would model the training data (linear regression), sort the data in ascending order (least difficult to most difficult), and input it into a model (as curriculum learning functions). The concept of curriculum learning was taught in Bengio as early as 2009.
Bengio highlights the speed at which a model can learn using curriculum learning (Introduction). A person of ordinary skill in the art at the time of the Applicant’s filing would have been aware of curriculum learning, and would have been motivated to use it in the single layer network of Han in order to improve the learning speed of the model.
The combination of Han and Bengio fails to teach:
“S5…storing a current neural network weight parameter…” and
“S6…restoring to the current neural network weight parameter stored in step S5…”
The Examiner interprets these limitations broadly to indicate some form of weight rewinding, which is a means for resetting a model’s weights during training in the event of performance degradation. This concept is taught for pruning networks in Mikler “Weight rewinding restores the network’s weights from a previous point (possibly beginning) in the training history and 119 then continues training from this point using the original training schedule – in our case a piecewise constant decaying 120 learning rate schedule.” (Page 5).
Mikler highlights the effectiveness of weight parameter rewinding versus finetuning when changing the structure of a network, such as when pruning (Introduction, Page 2). A person of ordinary skill in the art at the time of the Applicant’s filing would have been aware of the benefits of weight rewinding when adjusting the structure of a network and would have reasonably implemented it in a system such as a combination of Han and Bengio in order to improve the model’s accuracy.
The combination of Han, Bengio, and Mikler fails to teach:
“S8…, and adding a quantity of training data for the next stage of training until all the training data are trained.”
Dai, in a similar field of endeavor of training a grow-and-prune algorithm reads on “S8…, and adding a quantity of training data for the next stage of training until all the training data are trained.” by teaching “While effective, this may be too idealized for many real-world scenarios where training data and their associated labels may be collected in a continuous and incremental manner, and only some data instances may be used initially to obtain the first trained model.” (Introduction) and shows the inclusion of new data into the training of model iteratively following growth and pruning stages (Fig. 1) and “We first grow and prune a model with the initial data. When new data arrive, the network undergoes a growth phase (first, based on new data and then on all available data) that increases its size to accommodate new data and knowledge. Then, we employ a pruning phase to remove redundant parameters to obtain a compact inference model.” (Page 2).
Dai highlights that continuously changing data often requires retraining a neural network “For example, biomedical datasets are typically updated regularly when the number of data points obtained from patients increases, or disease trends shift across time [9]. This makes it necessary to update a DNN model frequently to accommodate the new data and capture the new information effectively.” and the need to efficiently expand the learning ability of the network without excessive redundancy (Introduction). It would have been obvious to one of ordinary skill in the art at the time of the Applicant’s filing to combine the iterative increase of training data of Dai into a system such as Han in order to improve the learning ability of the overall model.
In regards to claim 2: The present invention claims: “wherein the initial single-layer neural network comprises a plurality of input nodes, a hidden node and an output node, and the weight parameter comprises a weight value between the input nodes and the hidden node, a deviation value of the hidden node, a weight value between the hidden node and the output node, and a deviation value of the output node.” Han Fig. 1 shows an input node, hidden nodes, and an output node. Section 2 (Page 52) outlines the typical weight parameters and values associated with the layers, connections, and nodes.
In regards to claim 3: The present invention claims: “wherein the weight value between the hidden node and the output node of the initial single-layer neural network is 1, and the deviation value of the output node is a minimum value of the actual value in an initial data.” The combination of Han, Bengio, Mikler, and Dai discloses the claimed invention except for the explicit values listed (“the weight value…is 1” and the deviation value) in the claims. It would have been obvious to one having ordinary skill in the art at the time the invention was made to initialize similar parameters (Han Section 2 and Page 53; Dai Sections 4.1-4.2; at least), since it has been held that discovering an optimum value of a result effective variable involves only routine skill in the art. In re Boesch, 617 F.2d 272,265 USPQ 215 (CCPA 1980).
In regards to claim 8: The present invention claims: “wherein the reorganization module executes a regularization module to adjust the weight parameter, and deletes one of all hidden nodes in each operation, and then a determining module determines whether the reorganized single-layer neural network after deleting nodes is acceptable, if yes, the deleted nodes are regarded as the invalid nodes, and the next node is determined after deleting the nodes, if not, the nodes are added back again, and then determining whether the next node is an invalid node.” (Examiner’s Note: Although this claim’s wording is convoluted, the Examiner broadly maps this limitation to a method similar to Dai’s recoverable pruning (Section 4.3.2, Pages 5-6) where nodes that are pruned are not permanently removed so the model may recover lost accuracy from a previous iteration.
Claim(s) 4-7 is/are rejected under 35 U.S.C. 103 as being unpatentable over Han, Bengio, Mikler, and Dai as applied to claim 1 above, and further in view of Ru et al. (An Improved Levenberg-Marquardt Algorithm with Adaptive Learning Rate for RBF Neural Network, 2016), hereinafter Ru.
In regards to claim 4: While a combination of Han, Bengio, Mikler, and Dai reads on adjusting the weight parameters/learning rates during an iteration of the algorithm (Han, Section 3.2, at least), the combination fails to explicitly teach: “wherein the adjustment module sets an adjustment learning rate and adjusts the initial weight parameter according to the adjustment learning rate to obtain an adjustment weight parameter, when a loss function value of the adjustment weight parameter is less than a loss function value of the initial weight parameter, determining whether a residual value is less than a preset target value, if yes, using the acceptable single-layer neural network and the adjustment weight parameter and entering step S7, if no, increasing the adjustment learning rate to re-acquire the adjustment weight parameter until a predetermined quantity of training times is reached.” (Examiner’s Note: Although the limitations of this Claim (and claim 6) are somewhat convoluted, the Examiner interprets the steps listed as similar to a fairly conventional Levenberg-Marquardt algorithm for an adaptive learning rate. Ru teaches “Besides, an important parameter which plays an important role in the update process is λ, and it is related to the trust region’s size. When λ is close to zero, the LM algorithm is equal to the Newton's method; on the contrary, when it changes to a larger number, it is close to the gradient descent method. Therefore, adjusting the learning rate can realize the combination of suitable features of both algorithms: the local convergence properties of Gauss-Newton as well as the consistent error decrease properties of gradient descent method. To fasten the convergence of the algorithm, a strategy for parameter adjustment [24] is selected to adjust λ in every iteration process.” (Page 3632) In both cases, either raising the learning rate as in claim 4, or lowering it as in claim 6, the learning rate is optimized to speed up convergence in comparison to output error or loss values.
Ru teaches “the adaptive learning rate is integrated into the improved LM algorithm in order to accelerate the convergence speed of training algorithm and improve the network performance of nonlinear system modeling.” (Abstract). It would have been obvious to on of ordinary skill in the art at the time of the Applicant’s filing to incorporate elements of Ru’s improved adaptive learning rate algorithm into a combination of Han, Bengio, Mikler, and Dai in order to speed up model convergence.
In regards to claim 5: The present invention claims: “wherein the adjustment learning rate is increased to 1.2 times.” The combination of Han, Bengio, Mikler, Dai, and discloses the claimed invention except for the explicit values listed (“increased to 1.2 times”) in the claims. It would have been obvious to one having ordinary skill in the art at the time the invention was made to adjust the learning rate (Han Section 3.2; Bengio Section 4.2; at least), since it has been held that discovering an optimum value of a result effective variable involves only routine skill in the art. In re Boesch, 617 F.2d 272,265 USPQ 215 (CCPA 1980).
In regards to claim 6: The present invention claims: “wherein when the loss function value of the adjustment weight parameter is not less than the loss function value of the initial weight parameter, determining whether the adjustment learning rate is greater than an adjustment learning rate threshold, if yes, reducing the adjustment learning rate to re-acquire the adjustment weight parameter, if no, entering step S6.” See above how a combination of Han, Bengio, Mikler, and Dai, along with the adaptive learning algorithm of Ru would read on raising or lowering the learning rate to best speed up convergence each iteration of the algorithm. Mikler also makes reference to learning rate rewinding (Section 4.3), which while not an exact mapping, lends itself to the act of resetting or reducing weights to speed or improve convergence being known in the art at the time of the Applicant’s filing.
In regards to claim 7: The present invention claims: “wherein the adjustment learning rate is reduced to 0.7 times.” The combination of Han, Bengio, Mikler, Dai, and discloses the claimed invention except for the explicit values listed (“reduced to 0.7 times”) in the claims. It would have been obvious to one having ordinary skill in the art at the time the invention was made to adjust the learning rate (Han Section 3.2; Bengio Section 4.2; at least), since it has been held that discovering an optimum value of a result effective variable involves only routine skill in the art. In re Boesch, 617 F.2d 272,265 USPQ 215 (CCPA 1980).
Claim(s) 9 and 10 is/are rejected under 35 U.S.C. 103 as being unpatentable over Han, Bengio, Mikler, and Dai as applied to claim 1 above, and further in view of Bouqata et al (US 2013/0262349 A1), hereinafter Bouqata.
In regards to claim 9: The present invention claims: “wherein an output value of the output node is a forecast of raw material consumption after a predetermined period.” The combination of Han, Bengio, Mikler, and Dai fails to explicitly teach the particular field of endeavor recited in the Claims 9 and 10, however; Bouqata in a similar field of endeavor of adaptive learning teaches “A computer-implemented system includes an edge module and at least one input device coupled to the edge module. The at least one input device is configured to generate data input signals. The system also includes a cognitive module coupled to the edge module. The cognitive module includes a perception sub-module coupled to the edge module. The perception sub-module is configured to receive the data input signals. The cognitive module also includes a learning sub-module coupled to the perception sub-module. The learning submodule is configured to adaptively learn at least in part utilizing the data input signals.” (Abstract) and “As used herein, the term "asset management system" is intended to be representative of any computer-implemented programs and computer-based systems that facilitate, without limitation, design, planning, execution, control, and monitoring of assets. Such asset management systems include, without limitation, " supply chain management (SCM) systems." As used herein, SCM systems facilitate executing activities that include, without limitation, tracking movement and storage of raw materials, tracking work-in process inventory, and tracking finished goods from a point of origin to a point of consumption, and synchronizing supply with demand.” ([0034])
Ru highlights the need for automated intelligent systems with adaptive learning capabilities in industrial systems (Background). It would have been obvious to one of ordinary skill in the art working in the particular field of endeavor to utilize a system such as a combination of Han, Bengio, Mikler, and Dai to improve the intelligent automation systems of an industrial setting.
In regards to claim 10: The present invention claims: “wherein an input variable of the input nodes comprise raw material consumption, average consumption, raw material part number, raw material group classification, and usage period or a combination thereof during different periods.” Ru teaches “supply chain management (SCM) systems." As used herein, SCM systems facilitate executing activities that include, without limitation, tracking movement and storage of raw materials, tracking work-in process inventory, and tracking finished goods from a point of origin to a point of consumption, and synchronizing supply with demand.” ([0034]) which the Examiner submits broadly reads on raw material usage or consumption.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to GRIFFIN T BEAN whose telephone number is (703)756-1473. The examiner can normally be reached M - F 7:30 - 4:30.
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, Li Zhen can be reached at (571) 272-3768. 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.
/GRIFFIN TANNER BEAN/ Examiner, Art Unit 2121
/Li B. Zhen/ Supervisory Patent Examiner, Art Unit 2121
Prosecution Insights