Detailed Action
This action is in response to the RCE filed on 01/15/2026 for the amended claims filed 01/15/2026 for application 17/553,239, in which:
Claims 1 and 12 are independent claims.
Claims 1, 6-7, and 12 are currently amended.
Claim 9 is canceled.
Claims 1-8 and 11-21 are currently pending.
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 01/15/2026 has been entered.
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 .
Regarding the 35 USC § 112(b) Rejections:
Applicant's amendments to Claims 6 and 7 overcomes the previous rejections. The 35 USC § 112(b) Rejections have been withdrawn.
Response to Arguments
Applicant's arguments filed 01/15/2026 have been fully considered but they are not persuasive.
Regarding the 35 USC § 101 Rejections:
Applicant's arguments regarding the 35 U.S.C. 101 rejections of the previous office action have been fully considered, but are unpersuasive.
Applicant has amended the claims (Pages 7-9), to add more details to the specifics of the first/second neural network to be able to differentiate and highlight the two different separate neural networks and how they also work together via the encoding … , classifying … ,and … adapting … limitations. Thus, the Applicant's amended claims make clear that the first neural network is pretrained to identify and extract primary and semantic features using neural network weights and that the first neural network is used to encode input information into data vectors and extracting particular data vectors representing features within the input information. The amended, independent claims, further make clear that the second neural network is pretrained to match features to exemplars and is used to classify the particular data vectors. Applicant further supports their assertions by noting that the independent claims amount to significantly more due to the classification by the second neural network, at least one layer of the first neural network or second neural network by adapting layers to improve classification of particular data vectors. Thus, the limitations within the independent claim contain additional elements which amount to significantly more when viewed separately and in combination. Applicant further notes the claimed invention solving the problem that edge devices have limited processing power; where this invention resolves the limitations of edge devices.
Examiner respectfully disagrees. 35 U.S.C. § 101 rejections for the amended claims are directed to an abstract idea (Step 2A Prong 1) and do not integrate the abstract idea into a practical application (Step 2A Prong 2). The claims recites abstract ideas a-c (encoding … , classifying … ,and … adapting … claims); where the abstract ideas are evaluations/judgements that can be performed in the human mind (or by a human using pen and paper). The independent claim is no more detailed than using a device to encode specific features (mental process) via a specific neural network (additional element), classify data vectors (mental process) via a specific neural network (additional element), and adapting a specific layer to better classify (mental process). The additional elements noted within Step 2A Prong 2 are unable to amount to significantly more than the judicial exception (when evaluated individually and holistically) as they are merely applying the abstract idea on a computer or restricting the abstract idea to a specific technological environment. Thus, the additional elements are not able to integrate the abstract ideas in a practical application as they fall within MPEP 2106.05. The claims are directed towards the improvement of an abstract idea. Therefore, the claims do not integrate the judicial exception into a practical application. For the reasons given above and in the rejections below, the rejection to all Claims (including Claim 1, similar independent claims, and all dependent Claims) are maintained. More specific details are discussed below within the 35 USC § 101 Rejections.
Applicant asserts (Pages 9-10), that the independent claims are similar to the claims within Ex parte GUILLAUME DESJARDINS, RAZVAN PASCANU, RAIA THAIS HADSELL, JAMES KIRKPATRICK, JOEL WILLIAM VENESS, and NEIL CHARLES RABINOWITZ. Thus, the independent claims contain additional element(s) (or combination of elements) that integrate any judicial exception of the claim into a practical application; as the additional elements provide an improvement of computer/technology/technology field.
Examiner respectfully disagrees. The claims are directed towards the improvement of an abstract idea. Improvements to an abstract idea are still considered to an abstract idea. Additionally, the Claims does not reflect alleged improvement in the functioning of a computer or hardware processor rather the additional elements merely use a generic computer component to perform the abstract idea or restricting the abstract idea to a particular technological environment. Therefore, the claims do not integrate the judicial exception into a practical application nor amount to significantly more. The office action establishes a proper and well-supported prima facie case as the claims are explained to be not patentable via the Patent Subject Matter Eligibility steps within MPEP 2106; thus, the additional elements noted within Step 2A Prong 2 are unable to amount to significantly more than the judicial exception (when evaluated individually and holistically). The rejections follow the steps of the analysis laid out within the MPEP which was followed for the previous and current examination (see MPEP 2106). The rejection also follows the steps of the analysis as laid out in the MPEP which was followed for the previous and current examination (see MPEP 2106). The newly amended limitations within the amended claims do indeed contain additional elements (and abstract ideas) but the additional elements are unable to integrate the judicial exception as the training/pretraining by a neural network is not an abstract idea and the limitation is directed to merely applying an abstract idea using a generic computer as a tool (see MPEP 2106.05(f)(2), 2106.04(d)). The limitation is unable to provide the alleged improvement of classification as training/pretraining is done to accomplish the abstract idea(s). The training is applied which is encoding/classifying/adapting the abstract idea; which is currently broad without details of how to achieve the alleged improvement noted within the specification. The limitations are unable to provide improvement as they are currently being evaluated as either abstract idea(s) or additional elements that fall within MPEP 2106.05. The claim is not patent eligible. Although the Claims are interpreted in light of the specification, limitations from the specification are not read into the Claims.
MPEP 2106.05(a) recites:
After the examiner has consulted the specification and determined that the disclosed invention improves technology, the claim must be evaluated to ensure the claim itself reflects the disclosed improvement in technology … the claim must include the components or steps of the invention that provide the improvement described in the specification
…
It is important to note, the judicial exception alone cannot provide the improvement. The improvement can be provided by one or more additional elements. See the discussion of Diamond v. Diehr, 450 U.S. 175, 187 and 191-92, 209 USPQ 1, 10 (1981)) in subsection II, below.
Applicant fails to show how any alleged technical improvement would be provided by anything more than the judicial exception on its own. Additionally, applicant fails to show how the claim includes components or steps that would provide the alleged improvement described in the specification. By MPEP 2106.05(f)(1), "the claim recites only the idea of a solution or outcome, i.e. the claim fails to recite details of how a solution to a problem is accomplished". Moreover, the examiner maintains that the Claim does not impose any meaningful limits on the judicial exception. As noted in the rejection, the Claim does not include additional elements that are sufficient to amount to an integration of the identified abstract idea(s) into a practical application, thus the claim is directed to an abstract idea.
Applicant asserts (Page 11) that claims 1-8 and 11-21 fully satisfy the requirements of 35 U.S.C. §101 and are patentable thereunder because amended claims 1-8 and 11-21 are directed to statutory subject matter. The Applicant respectfully requests that the present rejection of such claim be withdrawn.
Examiner respectfully disagrees. Applicant’s arguments regarding the other independent and dependent claims rely upon the same assertions as with respect to Claim 1, and are thus likewise unpersuasive. Therefore, for the reasons given above and in the rejections below, the rejection to all Claims (including Claim 1, similar independent claims, and all dependent Claims) are maintained. More specific details are discussed below within the 35 USC § 101 Rejections.
Regarding the 35 USC § 103 Rejections:
Applicant's arguments regarding the 35 U.S.C. 103 rejections of the previous office action have been fully considered, but are unpersuasive.
Applicants traverses the rejections of the pending claims (Pages 11-12) due the failure of an asserted combination to teach or suggest each and every feature of a claim remains fatal to an obviousness rejection under 35 U.S.C. §103. Applicant further supports their asserts In re Royka, In re Wada/Murphy, In re Cohia, and In re Fine to note the responsibility to establish a prima facie case of obviousness.
Examiner respectfully disagrees. The office action establishes a proper and well-supported prima facie case as the claims are explained to be not patentable over the prior art(s) under the guidance of the MPEP; where each limitation is explicitly disclosed via combination of the references. More explained below within the response to arguments and within the office action.
Applicant asserts (Pages 12-13), that the combination within the 103 rejections fails to teach or suggest all limitations of the newly amended independent claims and recites the encoding …, … adapting … and classifying … limitations. The applicant disagrees with Frenkel teaching a first neural network pretrained to identify and extract primary and semantic features using neural network weights, input information into data vectors and extracting particular data vectors as the FC layer of Frenkel is just that; where an FC layer of a single neural network is not a separate second neural network trained for different purposes as the first neural network that is claimed. And even if Frenkel teaches a first and second neural network… Frenkel absolutely fails to teach or suggests the newly amended limitations and notes the encoding …, … adapting … and classifying … limitations; specifically, the first neural network … using neural networks weights; … using a second neural network pretrained to match features to exemplars …; and based on the classification, … better match features to exemplars to … .
Examiner respectfully disagrees. The encoding …, … adapting … and classifying … limitations are taught by Frenkel and noted within the office action; please note, that no arguments for disagreement are provided. Applicant's arguments fail to comply with 37 CFR 1.111(b) because they amount to a general allegation that the claims define a patentable invention without specifically pointing out how the language of the claims patentably distinguishes them from the references. Examiner disagrees with the specific limitation noted by the applicant where Frenkel only teaches an FC layer which is just a single neural network without a separate second neural network trained for different purposes. In response to applicant's argument that the references fail to show certain features of the invention, it is noted that the features upon which applicant relies (i.e., “separate second neural network”) are not recited in the rejected claim(s). Although the claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns, 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993). However, applicant’s arguments with respect to the independent claim(s) reciting the limitation “… pretrained to match features to exemplars …” have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. Please note that the limitations recited again for the encoding …, … adapting … and classifying … limitations do not provide any arguments and merely note failing to be taught by Frenkel; thus, Applicant's arguments fail to comply with 37 CFR 1.111(b) because they amount to a general allegation that the claims define a patentable invention without specifically pointing out how the language of the claims patentably distinguishes them from the references.
Applicant asserts (Pages 13-14), that Parhi also fails to disclose the newly amended limitations as noted above for Frenkel. As such, the Applicant submits that any allowable combination of Frenkel and Parhi also fail to teach or suggest the newly amended limitations within the independent claims. Thus, the Applicant submits that the Final Office Action fails to present a prima facie case of anticipation in view of the deficiencies of Frenkel and Parhi. Therefore, the Applicant submits that for at least the reasons recited above, at least the Applicant's independent claims are not rendered obvious by the teachings of Frenkel and Parhi, alone or in any allowable combination, and, as such, fully satisfies the requirements of 35 U.S.C. § 103 and is patentable thereunder. Furthermore, the dependent claims depending on the independent claim should be also considered patentable due to dependency.
As noted above, Applicant's arguments fail to comply with 37 CFR 1.111(b) because they amount to a general allegation that the claims define a patentable invention without specifically pointing out how the language of the claims patentably distinguishes them from the references. Further, they do not show how the amendments avoid such references or objections. The office action establishes a proper and well-supported prima facie case as the claims are explained to be not patentable over the prior art(s) under the guidance of the MPEP; where each limitation is explicitly disclosed via combination of the references. Applicant’s arguments regarding the other independent and dependent claims rely upon the same assertions as with respect to Claim 1, and are thus likewise unpersuasive. Therefore, for the reasons given above and in the updated rejections below, the rejection to all Claims (including Claim 1, analogous independent Claims, and all dependent Claims) are maintained and updated as necessitated by Claim amendments. More specific details are discussed below within the 35 USC § 103 Rejections.
Claim Objections
Claim 6 is objected to because of the following informalities: Claim 6 recites “… both of the other apparatus to include … ” which appears to be a typographical error. Appropriate correction is required. For the purposes of examination, the claim is being interpreted as “… both of the other apparatuses to include …”.
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 1-8 and 11-21 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Claim 1:
Subject Matter Eligibility Analysis Step 1:
Claim 1 recites an apparatus, thus a machine, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
However, Claim 1 further recites the apparatus comprising of:
encoding … identify and extract primary and semantic features … , input information into data vectors and extracting particular data vectors representing features within the input information (a human being can mentally apply evaluation to encode input information into vectors and extract features from the vectors with the aid of pen and paper by identifying and extracting specific features)
classifying … the particular data vectors (a human being can mentally apply evaluation to classify particular data vectors)
based on the classification, adapting at least one layer in the first neural network, second neural network or both by performing at least one of (1) altering weights, (2) adding layers, (3) deleting layers, (4) reordering layers to better match features to exemplars to improve classification of the particular data vectors (a human being can mentally apply evaluation to classify vectors where a human being can make a judgement to adapt specific neural networks by updating weights/model structures to improve classification)
Claim 1 thus recites an abstract idea (that falls into the “mental processes” group of abstract ideas).
Subject Matter Eligibility Analysis Step 2A Prong 2:
This judicial exception is not integrated into a practical application because the additional elements recited consists of:
An apparatus comprising: a processor; and a memory coupled to the processor, the memory having stored therein at least one of programs or instructions executable by the processor to configure the apparatus to: (to perform a mental process and the performance of an abstract idea on a computer is no more than instructions to “apply it” on a computer, by MPEP 2106.05(f))
… using a first neural network pretrained to … using neural network weights … wherein the first neural network comprises at least one encoder layer and at least one adaptor layer (to perform a mental process and the performance of an abstract idea on a computer is no more than instructions to “apply it” on a computer, by MPEP 2106.05(f))
… using a second neural network pretrained to match features to exemplars … (to perform a mental process and the performance of an abstract idea on a computer is no more than instructions to “apply it” on a computer, by MPEP 2106.05(f))
wherein the first neural network, the second neural network or both are trained using gradient-free training (which is restricting the abstract idea to a Particular Technological Environment, by MPEP 2106.05(h))
Subject Matter Eligibility Analysis Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements recited, alone or in combination, do not provide significantly more than the abstract idea itself. Additional elements a-c are merely applying the abstract idea on a computer (MPEP 2106.05(f)) which cannot provide significantly more. Additional element d is only restricting the abstract idea to a Particular Technological Environment (MPEP 2106.05(h)) which cannot provide significantly more. Thus, the claim is subject-matter ineligible.
Regarding Claim 2:
Subject Matter Eligibility Analysis Step 1:
Dependent Claim 2 recites the apparatus of Claim 1. Claim 1 is an apparatus, thus a machine, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
However, Claim 2 further recites:
… the particular data vectors are extracted using feature parameters defined by the first neural network (a human being can mentally apply evaluation to extract vectors using feature parameter defined by a neural network)
… the particular data vectors are classified using classification exemplars defined within the second neural network (a human being can mentally apply evaluation to classify vectors using classification exemplars defined by a neural network)
Claim 2 thus recites an abstract idea (that falls into the “mental processes” group of abstract ideas).
Subject Matter Eligibility Analysis Step 2A Prong 2:
This judicial exception is not integrated into a practical application because there are no new additional elements recited.
Subject Matter Eligibility Analysis Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because there are no new additional elements recited. The judicial exception alone does not provide significantly more than the abstract idea itself. Thus, the claim is subject-matter ineligible.
Regarding Claim 3:
Subject Matter Eligibility Analysis Step 1:
Dependent Claim 3 recites the apparatus of Claim 2. Claim 2 is an apparatus, thus a machine, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
However, Claim 3 does not recite any additional abstract ideas and only inherits the abstract ideas from Claim 2. Claim 3 thus recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
This judicial exception is not integrated into a practical application because the sole additional element consists of the feature parameters are pre-defined parameters, learned parameters, or a combination of predefined parameters and learned parameters and classification exemplars are pre-defined exemplars, learned exemplars, or a combination of predefined exemplars and learned exemplars (which is restricting the abstract idea to a Particular Technological Environment, by MPEP 2106.05(h)).
Subject Matter Eligibility Analysis Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the new sole additional element recited, alone or in combination, does not provide significantly more than the abstract idea itself. The additional element is only restricting the abstract idea to a Particular Technological Environment (MPEP 2106.05(h)) which cannot provide significantly more. Thus, the claim is subject-matter ineligible.
Regarding Claim 4:
Subject Matter Eligibility Analysis Step 1:
Dependent Claim 4 recites the apparatus of Claim 1. Claim 1 is an apparatus, thus a machine, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
However, Claim 4 does not recite any additional abstract ideas and only inherits the abstract ideas from Claim 1. Claim 4 thus recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
This judicial exception is not integrated into a practical application because the sole additional element consists of the apparatus is further configured to generate hyperdimensional data vectors representing features within the input information (to perform a mental process and the performance of an abstract idea on a computer is no more than instructions to “apply it” on a computer, by MPEP 2106.05(f)).
Subject Matter Eligibility Analysis Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the new sole additional element recited, alone or in combination, does not provide significantly more than the abstract idea itself. The additional element is merely applying the abstract idea on a computer (MPEP 2106.05(f)) which cannot provide significantly more. Thus, the claim is subject-matter ineligible.
Regarding Claim 5:
Subject Matter Eligibility Analysis Step 1:
Dependent Claim 5 recites the apparatus of Claim 1. Claim 1 is an apparatus, thus a machine, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
However, Claim 5 does not recite any additional abstract ideas and only inherits the abstract ideas from Claim 1. Claim 5 thus recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
This judicial exception is not integrated into a practical application because the sole additional element consists of the first neural network, second neural network or both are capable of being retrained using gradient-free training (which is restricting the abstract idea to a Particular Technological Environment, by MPEP 2106.05(h)).
Subject Matter Eligibility Analysis Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the new sole additional element recited, alone or in combination, does not provide significantly more than the abstract idea itself. The additional element is only restricting the abstract idea to a Particular Technological Environment (MPEP 2106.05(h)) which cannot provide significantly more. Thus, the claim is subject-matter ineligible.
Regarding Claim 6:
Subject Matter Eligibility Analysis Step 1:
Dependent Claim 6 recites the apparatus of Claim 1. Claim 1 is an apparatus, thus a machine, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
However, Claim 6 does not recite any additional abstract ideas and only inherits the abstract ideas from Claim 1. Claim 6 thus recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
This judicial exception is not integrated into a practical application because the sole additional element consists of the apparatus is further configured to share one or more exemplars or feature parameters with at least one other apparatus to enable the first neural network, second neural network, or both of the other apparatus to include the one or more shared exemplars or feature parameters (which is restricting the abstract idea to a Particular Technological Environment, by MPEP 2106.05(h)).
Subject Matter Eligibility Analysis Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the new sole additional element recited, alone or in combination, does not provide significantly more than the abstract idea itself. The additional element is only restricting the abstract idea to a Particular Technological Environment (MPEP 2106.05(h)) which cannot provide significantly more. Thus, the claim is subject-matter ineligible.
Regarding Claim 7:
Subject Matter Eligibility Analysis Step 1:
Dependent Claim 7 recites the apparatus of Claim 6. Claim 6 is an apparatus, thus a machine, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
However, Claim 7 does not recite any additional abstract ideas and only inherits the abstract ideas from Claim 6. Claim 7 thus recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
This judicial exception is not integrated into a practical application because the sole additional element consists of the exemplar or feature parameters sharing occurs to enable the other apparatus to perform at least one of extracting or classifying a new feature (to perform a mental process and the performance of an abstract idea on a computer is no more than instructions to “apply it” on a computer, by MPEP 2106.05(f)).
Subject Matter Eligibility Analysis Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the new sole additional element recited, alone or in combination, does not provide significantly more than the abstract idea itself. The additional element is merely applying the abstract idea on a computer (MPEP 2106.05(f)) which cannot provide significantly more. Thus, the claim is subject-matter ineligible.
Regarding Claim 8:
Subject Matter Eligibility Analysis Step 1:
Dependent Claim 8 recites the apparatus of Claim 1. Claim 1 is an apparatus, thus a machine, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
However, Claim 8 does not recite any additional abstract ideas and only inherits the abstract ideas from Claim 1. Claim 8 thus recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
This judicial exception is not integrated into a practical application because the sole additional element consists of the first neural network is initially defined using a predefined model (which is restricting the abstract idea to a Particular Technological Environment, by MPEP 2106.05(h)).
Subject Matter Eligibility Analysis Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the new sole additional element recited, alone or in combination, does not provide significantly more than the abstract idea itself. The additional element is only restricting the abstract idea to a Particular Technological Environment (MPEP 2106.05(h)) which cannot provide significantly more. Thus, the claim is subject-matter ineligible.
Regarding Claim 10:
Subject Matter Eligibility Analysis Step 1:
Dependent Claim 10 recites the apparatus of Claim 1. Claim 1 is an apparatus, thus a machine, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
However, Claim 10 further recites wherein the apparatus is further configured to the apparatus is further configured to adjust the second neural network to create additional exemplars based on changes in classification requirements (a human being can mentally apply evaluation to adjust a neural network to create additional exemplars based on changes in requirements). Claim 10 thus recites an abstract idea (that falls into the “mental processes” group of abstract ideas).
Subject Matter Eligibility Analysis Step 2A Prong 2:
This judicial exception is not integrated into a practical application because there are no new additional elements recited.
Subject Matter Eligibility Analysis Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because there are no new additional elements recited. The judicial exception alone does not provide significantly more than the abstract idea itself. Thus, the claim is subject-matter ineligible.
Regarding Claim 11:
Subject Matter Eligibility Analysis Step 1:
Dependent Claim 11 recites the apparatus of Claim 1. Claim 1 is an apparatus, thus a machine, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
However, Claim 11 does not recite any additional abstract ideas and only inherits the abstract ideas from Claim 1. Claim 11 thus recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
This judicial exception is not integrated into a practical application because the sole additional element consists of the apparatus is further configured to alter the first neural network when a local environment changes for the apparatus (to perform a mental process and the performance of an abstract idea on a computer is no more than instructions to “apply it” on a computer, by MPEP 2106.05(f)).
Subject Matter Eligibility Analysis Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the new sole additional element recited, alone or in combination, does not provide significantly more than the abstract idea itself. The additional element is merely applying the abstract idea on a computer (MPEP 2106.05(f)) which cannot provide significantly more. Thus, the claim is subject-matter ineligible.
Regarding Claims 12-21:
Claims 12-21 incorporate substantively all the limitations of Claims 1-8 and 10-11 in an method (thus a process) and further recites a new additional element training a first neural network … training a second neural network … (these claim limitations appear to perform a mental process and the performance of an abstract idea on a computer is no more than instructions to “apply it” on a computer, by MPEP 2106.05(f)) and does not appear to integrate the abstract idea into a particular application; thus, the claim is subject -matter ineligible as it does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements, alone or in combination, do not provide significantly more than the abstract idea itself); thus, Claims 12-21 are rejected for reasons set forth in the rejections of Claims 1-8 and 10-11, respectively.
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.
Claims 1-5, 8, 12-16, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Frenkel et al., “A 28-nm Convolutional Neuromorphic Processor Enabling Online Learning with Spike-Based Retinas”, in view of Parhi et al., “Brain-Inspired Computing: Models and Architectures”, in view of Snell et al., “Prototypical Networks for Few-Shot Learning”.
Regarding Claim 1:
Frenkel teaches:
An apparatus comprising: a processor: and a memory coupled to the processor, the memory having stored therein at least one of programs or instructions executable by the processor to configure the apparatus to:
(Frenkel, Fig. 2; Abstract, “In order to leverage the data sparsity of spike-based neuromorphic retinas for adaptive edge computing and vision applications, we follow a top-down approach and propose SPOON…”; Page 1, Column 2, “SPOON (standing for spiking online-learning convolutional neuromorphic processor), an event-driven CNN (eCNN) for adaptive edge computing”. SPOON is a processor which is integrated within event-based vision/retina such as the AER retina system shown within Fig. 2. Thus, the apparatus is the SPOON processor (the examiner interpret an apparatus as a hardware device that implements machine learning algorithms in proximity with where data is generated/accumulated) which is connected to an event-based image sensor to gather real-time data; which implies the adaptive edge computing neuromorphic designed apparatus with a SPOON processor contains a memory component with programs or instructions for SPOON event-driven CNN and neuromorphic design method).
encoding, using a first neural network … input information into data vectors and
(Frenkel, Fig. 2; Page 2, Column 1, Paragraph 1, “… (AER) buses [23] are used for event-driven handling of input sensor spikes and of output inferences … All weights and parameters can be programmed and readback with an SPI bus. As neuromorphic vision sensors send spikes encoding temporal contrast … to efficiently extract this information, we use time-to-first-spike encoding (i.e. timing code) [25] in the convolutional layers, which are handled in the CONV core (Section II-A)”; Page 2, Column 1, Paragraph 2, “Input AER events from the sensor are encoded …”. The apparatus comprises two cores which are considered to be two separate neural networks that are coupled together within the system. The first neural network is considered the CONV core which is a Convolutional Neural Network that extracts and converts/encodes the neuromorphic vision sensor spike (input information); thus, a Spiking CNN (i.e. SCNN)) where the CONV core comprises a first neural network for encoding input information into data vectors; where the spike-based inputs are considered raw spike events (a type of data vector) captured by the AER retina shown in Fig. 2).
extracting particular data vectors representing features within the input information, wherein the first neural network comprises at least one encoder layer and at least one adaptor layer;
(Frenkel, Fig. 2; Page 2, Column 1, Paragraph 1, “… All weights and parameters can be programmed and readback with an SPI bus. As neuromorphic vision sensors send spikes encoding temporal contrast … to efficiently extract this information, we use time-to-first-spike encoding (i.e. timing code) [25] in the convolutional layers, which are handled in the CONV core (Section II-A)”. The SCNN which is utilized for feature extraction of the neuromorphic spike input information (the CONV core comprises the SCNN (first neural network) for encoding input information into data vectors). The AER retina input information shown in Fig. 2 is encoded into structured feature representations using the convolutional layers via spike-time encoding which finally outputs spikes that represents the extracted features (a type of data vectors) representing feature maps; thus, teaching extracting particular data vectors representing features within the input information. The SCNN contains an encoder layer (where the spike events are transformed into feature maps for FC core (the second neural network)); the adaptor layer (interpreted as a layer that can adapt the neural based on data received) is where the second neural network (classifier) is able to send feedback for updates/altering of the weights for the first neural network’s convolutional layers via reading back the feedback utilizing the SPI bus)).
classifying, using a second neural network … , the particular data vectors; and
based on the classification, adapting at least one layer in the first neural network, second neural network or both by performing at least one of (1) altering weights … to improve classification of particular data vectors; and
(Frenkel, Fig. 2 & 5 & 6; Page 2, Column 1, Paragraph 1, “Fully-connected layers are handled in the FC core (Section II-B), which uses a combination of frame based and event-driven processing for maximum data reuse and efficient handling of DRTP updates”; Page 3, Column 1, Paragraph 2, “The FC core consists of a 128-neuron fully-connected hidden layer followed by a 10-neuron output layer, both with 8-bit programmable weights … As highlighted in Fig. 2, the hidden layer output … is computed with a conventional frame-based approach as all the inputs are immediately available when receiving the CONV_DONE trigger, where … x is the input from the CONV core (Fig. 5). The hidden neurons are evaluated sequentially and inputs are processed by batch … Once the weighted sum of inputs … has been computed, output layer processing is triggered in an event-driven fashion to ensure maximum data reuse…”; Page 4, Column 2, “… SPOON reaches a test-set accuracy of 93.8% with offline-trained weights and of 90.2% (one epoch) or 93.0% (100 epochs) using on-chip online training, while consuming 665nJ per inference”. The apparatus comprises two cores which are considered to be two separate neural networks that are coupled together within the system. The second neural network is considered the FC core (fully connected layers) which is utilized as an adaptive classifier. The adaptive classifier is utilized for receiving the encoded feature vectors from the CONV Core (via input once receiving the CONV_DONE trigger), perform the classifying process on the particular data vectors received (spikes) using the fully connected layers (FC core comprises the second neural network), and applies DRTP (Direct Random Target Projection) to update via adaptive learning and no backpropagation (shown within Fig. 5(b)). Fig. 2 shows the FC core as the final core which does the final classification on the input data and where the DRTP weight update (shown in Figure 2 & 6 which are label based; thus, interpreted by the examiner as based on classification) allows the altering of weights on both the first (via the adaptor layer) and second neural network via the FC core as it is coupled with the CONV core to improve classification accuracy with more quantized training epochs and DRTP weight updates and is also depicted which is shown in Fig. 8).
wherein the first neural network, the second neural network or both are trained using gradient-free training.
(Frenkel, Fig. 5; Page 1, Column 2, Paragraph 2, “SPOON embeds online learning at low power and area overheads with the biologically-plausible direct random target projection (DRTP) algorithm”; Page 3, Column 2, Paragraph 1, “… DRTP is a low-cost algorithm suitable for deployment at the edge. It relies only on feedforward and local computation (Fig. 5) and estimates the hidden layer loss gradient … as a projection of the target vector”. The examiner interprets gradient-free training as a learning method that updates neural network weights without propagating gradients using backpropagation. Both CONV Core and FC Core do not backpropagate and the weights are updated for training based on DRTP where the error is randomly projected (not computed) back to the CONV core; thus, the DRTP algorithm does not rely on calculating gradients of the objective function. The CONV Core updates weights using the projection which allows the adaptive learning. Thus, both neural networks are trained using gradient-free training as they are trained with the DRTP algorithm which estimates/randomly projects the hidden layer loss gradient to update the weights with no backpropagation).
While Frenkel teaches the encoding for generating spike vectors that represent features within the input information… Frenkel does not explicitly disclose the first neural network pretrained to identify and extract primary and semantic features.
However, Parhi explicitly discloses:
… pretrained to identify and extract primary and semantic features using neural network weights …
(Parhi, Page 185, Column 2, Paragraph 1, “TensorFlow or PyTorch … have greatly facilitated transfer learning where pre-trained models learned from one dataset can be refined to learn models of another dataset; Page 198, Figure 26, Column 1, Paragraph 3, “… architecture of this network is shown in Fig. 26. At the highest level, it is a fully convolutional network comprised of two parts: (i) a contracting path and (ii) a symmetric expanding path”. Parhi teaches transfer learning to use pretrained models; where U-Net is one of the architectures to utilize the pretrained CNN (thus, using neural network weights as CNNs contain kernels/filters (which is known to one of ordinary skill in the art as weights of a convolutional neural network) ) which contains a contracting path to identify and extract features. The contracting path extracts both higher-level and lower level features (which is interpreted by the examiner as the primary and semantic features as the higher level features are the primary/important features while the semantic features are the lower level attributes/categories/functions/etc.); thus, used to identify and extract primary and semantic features).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use the apparatus of Frenkel comprising encoding and classifying utilizing CNNs, use of spike-vectors, and retraining with the explicitly pretrained CNNs and hyperdimensional encoder/vectors taught by Parhi to reduce training time, use less hardware compute resources, handle redundancy against noise, and represent the data in a more robust manner for retraining to improve accuracy (Parhi, Page 185, Column 2, Paragraph 1, “These tools have greatly facilitated transfer learning where pre-trained models learned from one dataset can be refined to learn models of another dataset. This reduces training time and improves accuracy. Due to the democratization of the tools and their availability, and transfer learning, deep learning as a research tool is now available to everyone and consumers can reap the benefits by using their edge devices”; , Page 193, Paragraph 1, “Traditional computing approaches rely on bits to encode data, and all operations are deterministic and require much hardware compute resources. However, HD computing uses a hypervector representation where the dimensionality is in the order of thousands. These ultra-wide words introduce redundancy against noise, and are, therefore, inherently robust”).
However, Frenkel/Parhi teach a second neural network and pretraining but do not explicitly disclose the second neural network being pretrained to match features to exemplars.
Nevertheless, Snell teaches:
… pretrained to match features to exemplars …
(Snell, Page 2, Paragraph 3, “Ours differs because it is a softmax over classes, rather than points, computed from Euclidean distances to each class’s prototype representation”; Abstract, “We propose a more streamlined approach, prototypical networks, that learns a metric space in which few-shot classification can be performed by computing Euclidean distances to prototype representations of each class … We further demonstrate that a similar idea can be used for zero-shot learning, where each class is described by a set of attributes ….”; Page 3, Equation 1, “
PNG
media_image1.png
51
149
media_image1.png
Greyscale
”; Page 6, Paragraph 6, “… We train these networks to specifically perform well in the few-shot setting by using episodic training …”. Equation 1 shows the embedding function fθ(x) which represents the feature embedding for each class representation (ck = prototype); where, the prototype is taking all exemplar features for class k to produce a prototype vector/embedding representation. Thus, fθ(x) is interpreted by the examiner as the feature embedding which is matched to the exemplars ck which is pretrained via episodic training to learn the metric space).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use the classification apparatus of Frenkel/Parhi with the explicitly matching of features to exemplars taught by Snell to represent each feature to an exemplar, training via episodic training, simplicity, efficiency, and performance (Snell, Page 6, Paragraph 6, “…We have proposed a simple method called prototypical networks for few-shot learning based on the idea that we can represent each class by the mean of its examples in a representation space learned by a neural network. We train these networks to specifically perform well in the few-shot setting by using episodic training. Prototypical networks are simple to implement, and computationally efficient. We showed that this approach is equivalent to predicting the weights of a linear classifier, where the weights and biases are a function of the prototypes. Prototypical networks achieve state-of-the-art results … We further showed how this approach can be adapted to the zero-shot setting by taking an embedding of an attribute vector for each class to be the prototype. This approach achieves state-of-the-art results on zero-shot classification …”).
Regarding Claim 2:
Frenkel/Parhi/Snell teach the apparatus of Claim 1 and Frenkel further teaches:
wherein the particular data vectors are extracted using feature parameters defined by the first neural network and
(Frenkel, Fig. 2 & 3; Page 2, Column 1, Paragraph 1, “… to efficiently extract this information, we use time-to-first-spike encoding (i.e. timing code) [25] in the convolutional layers, which are handled in the CONV core (Section II-A)”; Page 2, Column 1, Paragraph 2, “The CONV core consists of a convolutional layer with 10 5x5 8-bit programmable kernels followed by a stride-4 maxpooling layer … Input AER events from the sensor are encoded with an 11-bit address, which covers the pixel {x,y} coordinates for 32x32 images and an ON/OFF polarity bit … The partial sums (psums) of the feature map elements associated to kernel i and input pixel {x,y}”. The feature parameters defined by the first neural network (CONV Core) are the Convolutional Kernels/filters, the weights of the spiking neurons, timestamps, and the maxpooling layer. The CONV Core (feature extractor) extracts the particular data vectors using feature parameters defined by the first neural network by utilizing feature parameters such as the # of spiking convolutional layers, # of the kernels (with stride)/, # of maxpooling layers, which are defined by the first neural network (SCNN) to process and extract features from the spike vectors).
the particular data vectors are classified using classification exemplars defined within the second neural network.
(Frenkel, Fig. 3; Page 2, Column 1, Paragraph 1, “Fully-connected layers are handled in the FC core (Section II-B), which uses a combination of frame based and event-driven processing for maximum data reuse and efficient handling of DRTP updates”; Page 3, Column 1, Paragraph 2, “The FC core consists of a 128-neuron fully-connected hidden layer followed by a 10-neuron output layer, both with 8-bit programmable weights … As highlighted in Fig. 2, the hidden layer output … is computed with a conventional frame-based approach as all the inputs are immediately available when receiving the CONV_DONE trigger, where … x is the input from the CONV core (Fig. 5). The hidden neurons are evaluated sequentially and inputs are processed by batch … Once the weighted sum of inputs … has been computed, output layer processing is triggered in an event-driven fashion to ensure maximum data reuse…”. The classifier is utilized for receiving the encoded feature vectors from the CONV Core, then performs the classifying process on the particular data vectors received (spikes) using the fully connected layers (FC core comprises the second neural network), and applies DRTP (Direct Random Target Projection) to update and adapts both with adaptive learning and no backpropagation (shown within Fig. 5(b)). classification exemplars (which is interpreted by the examiner as a feature representation that defines a class) within the classifier are each neuron within the FC core as they represent the different learned categories/classes. The exemplars are defined by the second neural network due to the exemplars being stored in the weights as each neuron in the FC core is associated with a specific class for categorization and these weights learn over time via DRTP (random projections) to improve the final classification).
Regarding Claim 3:
Frenkel/Parhi/Snell teach the apparatus of Claim 2 and Frenkel further teaches:
wherein the feature parameters are pre-defined parameters, learned parameters, or a combination of predefined parameters and learned parameters and
(Frenkel, Page 2, Paragraph 2, “The CONV core consists of a convolutional layer with 10 5x5 8-bit programmable kernels followed by a stride-4 maxpooling layer, kernels are randomly initialized upon reset … Input AER events from the sensor are encoded with an 11-bit address, which covers the pixel {x,y} coordinates for 32x32 images and an ON/OFF polarity bit”; Page 2, Column 1, Paragraph 1, “All weights and parameters can be programmed and readback with an SPI bus”. The feature parameters within CONV core are first pre-defined and initialized randomly and then learned upon via DRTP. Feature parameters of a CNN include # of convolutional layers, # of kernels, # of pooling layers, etc. Thus, the feature parameters are a combination of predefined parameters and learned parameters.).
classification exemplars are pre-defined exemplars, learned exemplars, or a combination of predefined exemplars and learned exemplars.
(Frenkel, Page 3, Paragraph 2, “The FC core consists of a 128-neuron fully-connected hidden layer followed by a 10-neuron output layer, both with 8-bit programmable weights that are automatically initialized to zero for online learning (Section II-C)”; Page 2, Column 1, Paragraph 1, “All weights and parameters can be programmed and readback with an SPI bus”; Page 3, Column 1, Paragraph 3, “C. On-chip online training with direct random target projection (DRTP)”.The classification exemplars within the FC core are predefined and initialized to zero and then learned upon via on-chip online training with DRTP. Thus, the classification exemplars are a combination of predefined exemplars and learned exemplars).
Regarding Claim 4:
Frenkel/Parhi/Snell teach the apparatus of Claim 1 and Frenkel further teaches:
wherein the apparatus is further configured to generate … data vectors representing features within the input information.
(Frenkel, Page 2, Paragraph 1, “As neuromorphic vision sensors send spikes encoding temporal contrast [24], … conveying useful data for edge detection. In order to efficiently extract this information, we use time-to-first-spike encoding (i.e. timing code) [25] in the convolutional layers, which are handled in the CONV core (Section II-A)”. The Spiking Convolutional Neural Network (SCNN) comprises an encoder that encodes spikes (data vectors representing features) with timing of the first spike).
While Frenkel teaches the encoder for generating spike vectors that represent features within the input information… Frenkel does not explicitly disclose the encoder being hyperdimensional.
However, Parhi explicitly discloses hyperdimensional computing with a hyperdimensional encoder:
hyperdimensional
(Parhi, Page 193, Column 2, Paragraph 1, “Hyperdimensional (HD) computing … relies on the high dimensionality, randomness, and the abundance of nearly orthogonal vectors [44]. Traditional computing approaches rely on bits to encode data … However, HD computing uses a hypervector representation where the dimensionality is in the order of thousands”. HD (hyperdimensional) computing is taught by using hypervectors within Parhi. Hypervectors are )
The motivation of Claim 1’s combination is maintained.
Regarding Claim 5:
Frenkel/Parhi/Snell teach the apparatus of Claim 1 and Frenkel further teaches:
wherein the first neural network, second neural network or both are capable of being retrained using gradient-free training.
While Frenkel teaches the gradient-free training of both the first and second neural network within Claim 1… Frenkel does not explicitly disclose retraining.
However, Parhi explicitly discloses:
retrained
(Parhi, Page 196, Column 1, Paragraph 1, “In an SNN, although the accuracy of the output is degraded initially, the accuracy improves as more inputs are processed. Even when more data are not available, the accuracy can be improved by retraining the weights”. Retraining the weights of the SNN (Spiking Neural Network) is taught by Parhi).
The motivation of Claim 1’s combination is maintained.
Regarding Claim 8:
Frenkel/Parhi/Snell teach the apparatus of Claim 1 and Frenkel further teaches:
wherein the first neural network is initially defined using a predefined model.
(Frenkel, Page 4, Column 2, Paragraph 1, “When enabling on-chip DRTP-based online learning, where the convolution kernels are initialized and fixed to random values and plastic fully-connected weights are initialized to zero …”. The CONV core (first neural network) is initialized to fixed/random values while the FC core (second neural network) neurons/weights/exemplars are initialized to zero when integrating online learning).
Regarding Claims 12-16, and 19:
Claims 12-16, and 19 recite a method to be performed by the apparatus of Claims 1-5 and 8, and thus are rejected for reasons set forth in the rejections of Claims 1-3 and 8, respectively.
Claims 6-7, and 17-18 are rejected under 35 U.S.C. 103 as being unpatentable over Frenkel et al., “A 28-nm Convolutional Neuromorphic Processor Enabling Online Learning with Spike-Based Retinas”, in view of Parhi et al., “Brain-Inspired Computing: Models and Architectures”, in view of Snell et al., “Prototypical Networks for Few-Shot Learning”, in view of Lim et al., “Federated Learning in Mobile Edge Networks: A Comprehensive Survey”.
Regarding Claim 6:
Frenkel/Parhi/Snell teach the apparatus of Claim 1 and Frenkel further teaches:
wherein the apparatus is further configured to… exemplars or feature parameters … to enable the first neural network, second neural network, or both of the other apparatus to …
While Frenkel teaches the gradient-free training of both the first and second neural network within Claim 1… Frenkel does not explicitly disclose retraining.
However, Lim teaches:
the apparatus share one or more … with at least one other similar apparatus … include the one or more shared exemplars or feature parameters.
(Lim, Fig. 3; Page 2035, Column 2, Paragraph 1, “FL allows users to collaboratively train a shared model while keeping personal data on their devices …”; Page 2048, Column 2, Paragraph 3, “One of the main objectives of FL is to protect the privacy of participants, i.e., the participants only need to share parameters of the trained model instead of sharing their actual data.”. Fig.3 shows a general training process for FL (federated learning) where the apparatus is able to share exemplars and feature parameters with at least one other apparatus as the models are shared with a global model (which contains the classification exemplars and feature parameters). Participants (interpreted as an apparatus) are able to share model parameters without sharing their actual data).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use the apparatus of Frenkel/Parhi comprising exemplars and feature parameters, with the federated learning method taught by Lim to share the exemplars/feature parameters to gather more data between edge devices, maintain privacy, and lead to rapid development in training (enabling technology to improve accuracy) (Lim, Page 2035, Column 2, Paragraph 1, “Motivated by privacy concerns among data owners, the concept of FL is introduced in [21]. FL allows users to collaboratively train a shared model while keeping personal data on their devices, thus alleviating their privacy concerns. As such, FL can serve as an enabling technology for ML model training …”; Page 2019, Column 2, Paragraph 2, “… based on an updated model shared from a participant, the server can check whether the shared model can help to improve the global model’s performance or not.”).
Regarding Claim 7:
Frenkel/Parhi/Snell/Lim teach the apparatus of Claim 6 and Frenkel further teaches:
wherein the exemplar or feature parameters sharing occurs to enable the other similar apparatus to perform at least one of extracting or classifying a new feature.
(Lim, Fig. 3; Page 2035, Column 2, Paragraph 1, “FL allows users to collaboratively train a shared model while keeping personal data on their devices …”; Page 2020, Column 1, Paragraph 1, “… the participant can extract more generalized features from the global model,”. Fig.3 shows a general training process for FL (federated learning) where apparatus are able to share exemplars and feature parameters with at least one other apparatus as the models are shared with a global model (which contains the classification exemplars and feature parameters). The other apparatus is able to extract new and updated feature parameters/exemplars from the global model).
The motivation of Claim 6’s combination is maintained.
Regarding Claims 17-18:
Claims 17-18 recite a method to be performed by the apparatus of Claims 6-7, and thus are rejected for reasons set forth in the rejections of Claims 6-7, respectively.
Claims 10-11 and 20-21 are rejected under 35 U.S.C. 103 as being unpatentable over Frenkel et al., “A 28-nm Convolutional Neuromorphic Processor Enabling Online Learning with Spike-Based Retinas”, in view of Parhi et al., “Brain-Inspired Computing: Models and Architectures”, in view of Snell et al., “Prototypical Networks for Few-Shot Learning”, in view of Lee et al., “Neuro.ZERO: A Zero-Energy Neural Network Accelerator for Embedded Sensing and Inference Systems”.
Regarding Claim 10:
Frenkel/Parhi/Snell teach the apparatus of Claim 1 and Frenkel further teaches:
wherein the apparatus is further configured to adjust the second neural network to …
Frenkel teaches the DRTP algorithm for the classifier to adjust the second neural network by adaptive learning. However, Frenkel does not explicitly disclose to create additional exemplars and only to adjust/adapt/update the exemplars.
However, Lee teaches:
create additional exemplars based on changes in classification requirements.
(Lee, Page 141, Column 2, Paragraph 1, “In such cases, an accelerator becomes necessary for the system to achieve desirable performance … generates an extended version of it by adding additional neurons to each layer”; Lee teaches being able to create additional neurons (exemplars) based on achieving desirable performance for classification by utilizing an accelerator (interpreted as changes in classification requirements such as performance/accuracy requirements)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use the apparatus of Frenkel comprising the CONV and FC Core with the learning method taught by Lee to adjust/alter the FC and CONV core by adding exemplars (neurons) or changing modes to increase efficiency (Lee, Page 141, Column 2, Paragraph 1, “A larger network having more neurons, in general, is a better classifier [67, 68, 113] … Neuro.ZERO is benefited by additional neurons … an accelerator becomes necessary for the system to achieve desirable performance … Neuro.ZERO generates an extended version of it by adding additional neurons to each layer. The newly added neurons are identical in numbers and types for each layer … extended convolutional (Conv) and two extended fully-connected (FC) layers having the same dimensions …”).
Regarding Claim 11:
Frenkel/Parhi/Snell teach the apparatus of Claim 1 and Frenkel further teaches:
wherein the apparatus alter the first neural network
Frenkel teaches the DRTP algorithm for the classifier to alter the first neural network by adaptive learning. However, Frenkel does not explicitly disclose altering when a proximate environment changes for the apparatus.
However, Lee teaches:
when a local environment changes for the apparatus.
(Lee, Page 141, Column 1, Paragraph 1, “… Neuro.ZERO can be operated (1) in the extended inference mode when the user enters an environment that requires higher-resolution images to detect objects, (2) in the expedited mode when the user is in a busy area, (3) in the ensemble mode when there are multiple cameras or a different sensor (e.g., microphone) to independently detect the same event, or (4) in the training mode when environment-specific parameter tuning is necessary to obtain better classification results”; Page 145, Column 2, Paragraph 1, “… number of weights trained at each iteration in skip-out algorithm is flexibly changed based on the energy …”. The system of Lee teaches being able to operate in different modes when a environment changes (detecting an energy change; thus, interpreted as a local change as the environment is based on energy) for the apparatus.).
The motivation of Claim 10’s combination is maintained.
Regarding Claims 15-18, and 20-21:
Claims 15-18, and 20-21 recite a method to be performed by the apparatus of Claims 4-7 and 10-11, and thus are rejected for reasons set forth in the rejections of Claims 15-18, and 20-21, respectively.
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 IBRAHIM RAHMAN whose telephone number is (703)756-1646. The examiner can normally be reached M-F 8am-5pm.
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, Kakali Chaki can be reached at (571) 272-3719. 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.
/I.R./Examiner, Art Unit 2122
/KAKALI CHAKI/Supervisory Patent Examiner, Art Unit 2122
Prosecution Insights