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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 01/06/2023 and 04/01/2024 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Drawings
The drawings are objected to as failing to comply with 37 CFR 1.84(p)(5) because they include the following reference character(s) not mentioned in the description: Fig. 1 reference number 1 and Fig. 2 reference number 1’ (see annotated drawings below).
PNG
media_image1.png
703
748
media_image1.png
Greyscale
PNG
media_image2.png
668
803
media_image2.png
Greyscale
Corrected drawing sheets in compliance with 37 CFR 1.121(d), or amendment to the specification to add the reference character(s) in the description in compliance with 37 CFR 1.121(b) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either “Replacement Sheet” or “New Sheet” pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
Specification
Applicant is reminded of the proper content of an abstract of the disclosure.
A patent abstract is a concise statement of the technical disclosure of the patent and should include that which is new in the art to which the invention pertains. The abstract should not refer to purported merits or speculative applications of the invention and should not compare the invention with the prior art.
If the patent is of a basic nature, the entire technical disclosure may be new in the art, and the abstract should be directed to the entire disclosure. If the patent is in the nature of an improvement in an old apparatus, process, product, or composition, the abstract should include the technical disclosure of the improvement. The abstract should also mention by way of example any preferred modifications or alternatives.
Where applicable, the abstract should include the following: (1) if a machine or apparatus, its organization and operation; (2) if an article, its method of making; (3) if a chemical compound, its identity and use; (4) if a mixture, its ingredients; (5) if a process, the steps.
Extensive mechanical and design details of an apparatus should not be included in the abstract. The abstract should be in narrative form and generally limited to a single paragraph within the range of 50 to 150 words in length.
See MPEP § 608.01(b) for guidelines for the preparation of patent abstracts.
Applicant is reminded of the proper language and format for an abstract of the disclosure.
The abstract should be in narrative form and generally limited to a single paragraph on a separate sheet within the range of 50 to 150 words in length. The abstract should describe the disclosure sufficiently to assist readers in deciding whether there is a need for consulting the full patent text for details.
The language should be clear and concise and should not repeat information given in the title. It should avoid using phrases which can be implied, such as, “The disclosure concerns,” “The disclosure defined by this invention,” “The disclosure describes,” etc. In addition, the form and legal phraseology often used in patent claims, such as “means” and “said,” should be avoided.
The abstract of the disclosure is objected to because is 168 words in length. A corrected abstract of the disclosure is required and must be presented on a separate sheet, apart from any other text. See MPEP § 608.01(b).
The disclosure is objected to because of the following informalities: paragraph [0029] recites “RGB-information” and [0046] recites “RGB data” but the specification does not provide a definition for what the term “RGB” stands for. Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-15 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 1 recites the limitation in line 15 “storing visual data samples from said visual data stream in the memory buffer, processing both visual data samples of the visual data stream and visual data samples from the memory buffer...”, however it’s not clear what is the difference between the samples of visual data stream and visual data samples from the memory buffer?. The specification does not provide a definition for the terms and how they are differ from each other. Further, the claim limitation as presented for examination, discloses visual data samples are stored from visual data stream which appears to be the same source. For examination purposes, the samples of visual data stream and visual data samples are the same data type, therefore it will be view as storing visual data samples in the memory buffer and processing visual data samples from memory buffer to learn implicit and explicit knowledge. In addition, Claim 1 recites the following limitation: “the architecture...” in line 8. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, examiner is interpreting these limitations as “the artificial cognitive architecture”.
Claim 2 recites the limitation “the second module” in lines 4 & 7. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, examiner is interpreting these limitations as “the second neural network module”.
Claim 5 recites “wherein the step of transforming and sharing learned knowledge representations from the second neural network module into the first neural network module comprises...” (emphasis added), however is not clear how the steps of “transforming and sharing learned knowledge representations” is achieved as claim 1, does not recites how this is done. Claim 1, only recites “transforming and sharing information between the first neural network module and the second neural network module”, not “transforming and sharing learned knowledge representations”. In addition, claim 5 recites the limitation “the step of” in line 1.There is insufficient antecedent basis for this limitation in the claim. For examination purposes, examiner is interpreting this limitations as “a step of”.
Claim 6 recites the limitation “the step of” in line 1 and “the first module” in line 3.There is insufficient antecedent basis for these limitation in the claim. For examination purposes, examiner is interpreting these limitations as “a step of” and “first neural network module”.
Claim 7 recites the limitation “the first module” in line 1. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, examiner is interpreting this limitations as “the first sub-module” in line 1.
Claim 9 recites the limitations “The method according to claim 1, “the step of”, “the first and second module” in line 1. There is insufficient antecedent basis for these limitation in the claim. For examination purposes, examiner is interpreting these limitations as “a step of” and “the first neural network module and second neural network module”.
Claim 10 recites the limitation “the objective” in lines 1-2.There is insufficient antecedent basis for this limitation in the claim. For examination purposes, examiner is interpreting this limitations as “an objective” in lines 1-2.
Claim 11 recites the limitation “the step of” in line 1. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, examiner is interpreting this limitations as “a step of”.
Claim 13 recites the limitations “the cognitive architecture” in line 4, “said second neural network” in line 6, and “the first module” in line 7. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, examiner is interpreting these limitations as “the artificial cognitive architecture”, “said second neural network module” , “the first neural network module”.
Claims 2-15 are dependent on claim 1, and thus are rejected for reasons set forth in the rejection of claim 1.
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.
Claim 12 is 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 Claim 12 defines a computer program product comprising instructions embodying functional descriptive material. However, the claim does not define a non-transitory computer readable medium ( or recording medium) and therefore, under the broadest reasonable interpretation (BRI) the “computer program product” of claim 1 encompasses signal per se. As stated in MPEP 2106.03 (II) “the BRI of machine readable media can encompass non-statutory transitory forms of signal transmission, such as a propagating electrical or electromagnetic signal per se”.
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, and 11-12 are rejected under 35 U.S.C. 103 as being unpatentable over Ron Sun A Tutorial on CLARION 5.0 (hereinafter Sun) in further view of in further view of Wu et al. Label-Efficient Online Continual Object Detection in Streaming Video (hereinafter Wu).
Regarding claim 1:
A computer-implemented method for continual task learning in an artificial cognitive architecture comprising: ( Sun pg. 5, para. 1, teaches a cognitive architecture named Connectionist Learning with Rule Induction ON-line (CLARION)).
a first neural network module for encoding explicit knowledge representations, ( Sun pg. 62, sec. 3.1.1 The Top Level, para. 1, and Fig. 1.1 teaches a top level module that encodes explicit knowledge representations. Specifically, Sun pg. 22, sec: 2.2.2.4 External instructions and top-down assimilation, para. 2, teaches the top level, serve as a teacher, thus the “top level module” can be view as a neural network (e.g., first neural network module)).
a second neural network module for encoding implicit knowledge representations itself comprising a plurality of neural network sub-modules for mutually different implicit functions, and ( Sun pg. 15, sec. 2.2 The Bottom Level, para. 1, & Fig. 1.1 teaches a bottom level module or “implicit decision network” (i.e., second neural network module ) that captures implicit representations and pg. 67, sec. 3.1.2 The bottom level, para. 1, teaches the bottom level module can be used for encoding implicit knowledge. In addition, Sun pg. 63, para. 1, teaches a bottom level module can be divided into multiple networks (plurality of neural network sub-modules) each for different kind of information (i.e., for mutually different implicit functions) ).
a memory buffer, ( Sun pg. 17, para. 2 teaches a current state buffer (memory buffer)).
the method comprising the steps of: providing a [sensory] data stream to the architecture; ( Sun teaches the input for the architecture include three groups of information, including sensory input (see pg. 16, sec: 2.2.1 Representation, para 1 & pg. 25, sec: 2.3.1 Representation which show top level and bottom level receive similar input). Someone ordinary skilled in the relevant art, will recognize that “sensory input” includes data such as vision to enable a machine learning model to perceive, understand and interacts with its observed environment).
storing [sensory] data samples from said [sensory] data stream in the memory buffer , ( Sun pg. 17, para. 2 teaches a current state buffer (memory buffer) that hold state information used by the top and bottom level module, this include the sensory input such as visual data).
processing both [sensory] data samples of the [sensory] data stream and visual data samples from the memory buffer using the first neural network module for learning explicit knowledge representations; ( Sun pg. 17, para. 2 teaches the three groups of information, including sensory input is being stored in the current state buffer (memory buffer) which will be used by the bottom level as well as the top level. Thus, the sensory input will be used by the top level module (i.e., first neural network module) for learning explicit knowledge representations (pg. 67, sec. 3.1.2 The bottom level, para. 1)).
processing both samples of said [sensory] data stream and [sensory] data samples from the memory buffer using the second neural network module for learning implicit knowledge representations; ( Sun pg. 17, para. 2 teaches the three groups of information, including sensory input is being stored in the current state buffer (memory buffer) which will be used by the bottom level as well as the top level. Thus, the sensory input will be used by the bottom level module (i.e., second neural network module) for learning implicit knowledge representations (pg. 62, sec. 3.1.1 The Top Level, para. 1)).
transforming and sharing information between the first neural network module and the second neural network module, as well as transforming and sharing information between sub-modules of the second neural network (
PNG
media_image3.png
374
656
media_image3.png
Greyscale
Sun pg. 12, para. 3 teaches once explicit knowledge is established at the top level (i.e., first neural network module) and it can be assimilated into the bottom level (i.e., second neural network module), this assimilation process can be done using supervised learning in which the top level serves as the teacher to the bottom level (pg. 22, sec: 2.2.2.4, para. 1-2), thus suggesting the assimilation process allows for knowledge to be shared and transformed between the top level module and bottom level module. Furthermore, Sun pg. 13, Figure 1.1 teaches the architecture of CLARION and teaches how the top level module and bottom level module transforms and share information (which is represented by the up and down arrows). In addition, Sun teaches the bottom level module comprises of multiple network (Sun pg. 63, para. 1), thus transformation and information between the multiple network (i.e., sub-modules) will also occur).
Sun does not explicitly teaches providing “visual data stream” to the architecture. Nonetheless, Wu teaches the following:
A computer-implemented method for continual task learning in an artificial cognitive architecture comprising: ( Wu Abstract, para. 1, teaches a method for continual leaning and Fig. 2 proposed an artificial cognitive architecture named “Efficient-CLS”. Further, Wu Appendix A.1.1 teaches training of the architecture is being carried out on 2 NVIDIA RTX 3090 GPUs, therefore such method will require a computer).
the method comprising the steps of: ( Wu Fig. 2 teaches a method comprising steps).
providing a visual data stream to the architecture; ( Wu pg. 4, Fig. 2 teaches providing continuous data that includes video frames (i.e., visual data stream) to the architecture of Efficient-CLS ).
storing visual data samples from said visual data stream in the memory buffer, ( Wu pg. 4, Fig. 2 and sec. 3 Efficient-CLS: Efficient Complementary Learning System. Para 2, teaches an external episodic memory (i.e., memory buffer) as a replay buffer “to store exemplars that can be retrieved for replays alongside ongoing video stream”).
processing both visual data samples of the visual data stream and visual data samples from the memory buffer using the first neural network module... ( Wu Fig. 2 teaches the episodic memory stores the visual frames ( i.e., visual data samples of the visual data stream and visual data samples) and teaches how the fast learner (i.e., first neural network module) uses the visual frames from the episodic memory).
processing both samples of said visual data stream and visual data samples using the second neural network module... ( Wu Fig. 2 teaches the slow learner (second neural network module) processing unlabeled frames (i.e., both samples of said visual data stream and visual data samples)).
Wu is also in the same field of endeavor as Sun (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of visual data frames being used by the cognitive architecture as being disclosed and taught by Wu, in the system taught by Sun to yield the predictable results of minimizing cost and reduce model retraining time as well as providing minimal forgetting (pg. 3, bullet points 1-2).
Regarding claim 11:
Sun and Wu tech The method according to claim 1. Wu specifically teaches further comprising the step of: using the second neural network module for decision making based on the visual data stream ( Wu Fig. 2, teaches the slow learner (second neural network module) generates pseudo labels, therefore the slow learners is used for decision making based on the unlabeled frames (i.e., visual data stream)).
Regarding claim 12:
Wu, Sun, Riemer and Aslan tech the method of claim 1. Wu specifically teaches A computer program product comprising instructions which, when the program is executed by a computer, causes the computer to carry out the method of claim 1 (Wu Appendix A.1.1 teaches training of the architecture is being carried out on 2 NVIDIA RTX 3090 GPUs, therefore such method will require a computer with a computer program product comprising instructions).
Claims 2-3, and 5-7 are rejected under 35 U.S.C. 103 as being unpatentable over Wu, Sun, in further view of Geirhos et al. ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness (hereinafter Geirhos).
Regarding claim 2:
Sun and Wu teach The method according to claim 1. Sun specifically teaches wherein the plurality of neural network sub-modules comprise a first and second sub-module wherein the method comprises the steps of: (Sun pg. 63, para. 1, teaches a bottom level module (i.e., second neural network module ) can be divided into multiple networks (plurality of neural network sub-modules) each for different kind of information (i.e., for mutually different implicit functions). Thus, this implies the multiple networks will comprise of a first and second sub-module).
consolidating knowledge from the first neural network module in a first sub-module of the plurality of neural network sub-modules of the second module as an implicit memory; and ( Sun pg. 13, Figure. 1.1 teaches the CLARION architecture that comprises of the top level module (i.e., first neural network module ) and the bottom level module (i.e., second neural network module) in which knowledge/information is being shared therefore knowledge will inheritably by consolidated an implicit memory. Further, since Sun pg. 63, para. 1, teaches the bottom level module comprises multiple networks (i.e., plurality of neural network sub-modules) someone ordinary skilled in the art will recognize that knowledge will also be consolidated into the multiple networks of the bottom level module ).
processing both the [sensory] data stream and [sensory] data samples from the memory buffer in a second sub-module of the plurality of neural network sub-modules of the second module ( Sun pg. 17, para. 2 teaches the three groups of information, including sensory input is being stored in the current state buffer (memory buffer) which will be used by the bottom level (i.e., second neural network module) which comprises multiple networks (pg. 63, para. 1) such as a “second sub-module”).
While Sun teaches sensory input that can include visual data samples from memory buffer, Sun does not explicitly teach processing an implicit inductive bias using both the visual data stream and visual data samples from the memory buffer... .
Nevertheless, Wu teaches the following:
processing both the visual data stream and visual data samples from the memory buffer... ( Wu pg. 4, Fig. 2 and sec. 3 Efficient-CLS: Efficient Complementary Learning System. Para 2, teaches an external episodic memory (i.e., memory buffer) as a replay buffer “to store exemplars that can be retrieved for replays alongside ongoing video stream”).
Neither Sun or Wu teaches processing an implicit inductive bias using both the visual data stream and visual data samples.
Nevertheless, Geirhos teaches the following:
processing an implicit inductive bias using both the visual data stream and visual data samples.... ( Geirhos pg. 9, para. 2, teaches introducing Stylized-ImageNet (SIN) in order to reduce the texture bias of convolutional neural networks (CNNs) and thereby forcing CNNs to go beyond texture recognition such that the texture bias is not by design but induced by ImageNet training data. In particular, pg. 4, Fig. 3 teaches how SIN introduces implicit inductive bias using shape as a solution to visual data stream such as an image of a ring-tailed lemur).
Geirhos is also in the same field of endeavor as Sun and Wu (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of implicit inductive bias for training data, as being disclosed and taught by Geirhos, in the system taught by Sun and Wu to yield the predictable results of showing how the “texture bias in standard CNNs can be overcome and changed towards a shape bias if trained on a suitable dataset”. Such that “networks with a higher shape bias are inherently more robust to many different image distortions... and reach higher performance on classification and object recognition tasks” (Geirhos pg. 3, para 1).
Regarding claim 3:
Sun, Wu and Geirhos teach The method according to claim 2. Sun specifically teaches wherein the plurality of neural network sub-modules comprises a third sub-module, and wherein the method comprises the step of: consolidating information from the second sub-module within the third sub-module ( Sun pg. 13, Figure. 1.1 teaches the CLARION architecture that comprises of the top level module (i.e., first neural network module ) and the bottom level module (i.e., second neural network module) in which knowledge/information is being shared therefore knowledge will inheritably by consolidated an implicit memory. Further, since Sun pg. 63, para. 1, teaches the bottom level module comprises multiple networks (i.e., plurality of neural network sub-modules) which can include a first, second and third sub-module for which someone ordinary skilled in the art will recognize that knowledge will also be consolidated into the multiple networks).
Regarding claim 5:
Sun, Wu and Geirhos The method according to claim 2. Wu specifically teaches wherein the step of transforming and sharing learned knowledge representations from the second neural network module into the first neural network module comprises sharing information from the first, second and third sub-modules into the first neural network module ( Sun pg. 13, Figure. 1.1 teaches the CLARION architecture which is a two levels type architecture that comprises a top level module (i.e., first neural network module ) and a bottom level module (i.e., second neural network module) in which knowledge/information is being shared, therefore knowledge will inheritably by consolidated an implicit memory. Further, since Sun pg. 63, para. 1, teaches the bottom level module comprises multiple networks (i.e., plurality of neural network sub-modules) which can include a first, second and third sub-module, information will be shared between the multiple networks and the first neural network module).
Regarding claim 6:
Sun, Wu and Geirhos teach The method according to claim 2. Sun specifically teaches wherein the step of consolidating knowledge from the first neural network module in a first sub-module occurs , and wherein information from a consolidated learning of first sub-module is transferred to both the first module and second sub-module ( Sun pg. 13, Figure. 1.1 teaches the CLARION architecture that comprises of the top level module (i.e., first neural network module ) and the bottom level module (i.e., second neural network module) in which knowledge/information is being shared therefore knowledge will inheritably by consolidated an implicit memory. Further, since Sun pg. 63, para. 1, teaches the bottom level module comprises multiple networks (i.e., plurality of neural network sub-modules) someone ordinary skilled in the art will recognize that knowledge will also be consolidated into the multiple networks which include a first and second sub-modules).
Sun does not explicitly teaches knowledge consolidation occurs at a regular interval. However, Wu teaches the following:
...wherein the step of consolidating knowledge from the first neural network module in [the second neural network] occurs at a regular interval... (Wu pg. 5, sec: 5 Synapses Consolidation via Exponential Moving Average, para. 1 teaches applying EMA to consolidate the knowledge between the fast learner (i.e., first neural network module) and the slow learner (i.e., second neural network module). In addition, Fig. 2 and sec: 3 para. 1-2 teaches consolidation of knowledge/information occurs at a regular interval as the learners are complementary to each other, forming a positive feedback loop, thus enabling consolidation of knowledge at regular interval).
Regarding claim 7:
Sun, Wu and Geirhos teach The method according to claim 2. Wu specifically teaches wherein the first module and second sub-module learn on their own modality with a supervised cross entropy loss on both the samples of the visual data stream and the samples of the memory buffer (Wu pg. 4, sec: 3.1 Learning with Labeled Frames, teaches a supervised cross entropy loss such as a classification loss on the label video frames samples is being used).
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Sun, Wu, Geirhos in further view of Tsatsin et al. US 2017/0357896 A1 (hereinafter Tsatsin).
Regarding claim 4:
Sun, Wu and Geirhos teach The method according to claim 3. While Sun teaches the bottom level module (i.e., second neural network module) comprises multiple networks (i.e., plurality of neural network sub-modules) this can include the third sub-module (see pg. 63, para. 1). Neither Sun, Wu and Geirhos disclose the third sub-module acting as a regularizer.
Nevertheless, Tsatsin teaches the following:
wherein the third sub-module acts as a regularizer (Tsatsin [0092] neural networks (i.e., second neural network module) are trained with regularizer (i.e., third sub-module). Such regularizer can be added to prevent overfitting and for example penalize and/or adjust for larger weights Wi).
Tsatsin is also in the same field of endeavor as Sun, Wu and Geirhos (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of a regularizer in neural networks as being disclosed and taught by Tsatsin, in the system taught by Sun, Wu and Geirhos to yield the predictable results of an overall more accurate and efficient visual image discovery results (Tsatsin [0092]).
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Wu, Sun, in further view of Riemer et al. Learning to Learn without Forgetting by Maximizing Transfer and Minimizing Interference as cited in IDS dated 01/06/2023 and in further view of Arani et al. US 2023/0076893 A1 (hereinafter Arani).
Regarding claim 8:
Sun and Wu teach The method according to claim 1. While Wu teaches maintaining an external episodic memory, as a replay buffer, to store exemplars that can be retrieved for replays alongside ongoing video stream (pg. 4, sec: 3 Efficient-CLS: Efficient Complementary Learning System, para. 2).
Neither Sun or Wu teach or suggest wherein the memory buffer is continuously or intermittently supplemented with new samples from the visual data stream replacing already present samples within said memory buffer, and wherein the method comprises: applying a logit loss between present samples and new samples.
Nonetheless, Riemer teaches the following:
wherein the memory buffer is continuously or intermittently supplemented with new samples from the visual data stream replacing already present samples within said memory buffer, and wherein the method comprises: ( Riemer Algorithm 1 line 20-21, teaches a memory buffer is updated continuously with “reservoir sampling” in the for loop. In particularly, pg. 4, sec: 3.1. Experience Replay, para. 2, teaches the buffer is updated “with reservoir sampling (Appendix F). This ensures that at every time-step the probability that any of the N examples seen has of being in the buffer is equal to
M
s
i
z
e
∕
N
. The content of the buffer resembles a stationary distribution over all examples seen to the extent that the items stored captures the variation of past examples”, that is the buffer is being updated with new samples, replacing already present samples within the buffer. In addition, Riemer pg. 8-9, sec: 5 Evaluating for continual reinforcement learning, and Fig. 4 teaches the implementation of MER algorithm in two environments, specifically in arcade games such as Catcher and Flappy Bird, this suggest visual data stream (i.e., frames from the games, as shown in Fig. 4) will be used as sample in order to implement MER).
Riemer is also in the same field of endeavor as Sun and Wu (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of processing sample from memory buffer, as being disclosed and taught by Riemer, in the system taught by Sun and Wu to yield the predictable results of “a general purpose solution to continual learning problems that outperforms strong baselines for both supervised continual learning benchmarks and continual learning in non-stationary reinforcement” ( see Riemer pg. 10, sec: 7 Conclusion).
Neither Sun, Wu or Riemer teach or suggest applying a logit loss between present samples and new samples.
applying a logit loss between present samples and new samples ( Arani [0074-0075] and Equation (1) teaches applying a cross entropy loss (i.e., logit loss) between the data stream from the episodic memory (i.e., present sample from the buffer) and the data stream (i.e., new samples)).
Arani is also in the same field of endeavor as Sun, Wu and Riemer (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of applying a logit loss, as being disclosed and taught by Arani, in the system taught by Sun, Wu and Riemer to yield the predictable results of providing “improved methods and systems that train and use artificial intelligence inference models that can take advantage of the interplay between rapid instance-based learning and slow structured learning” (Arani [0006]).
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Wu, Sun, in further view of Aslan et al. US 2017/0132528 A1 (hereinafter Aslan).
Regarding claim 9:
Wu and Sun teach The method according to claim 1. Neither Wu or Sun wherein the step of sharing information between the first and second module is governed by a knowledge sharing loss objective
wherein the step of sharing information between the first and second module is governed by a knowledge sharing loss objective ( Aslan [0007] teaches an objective function (knowledge sharing loss objective) is used for jointly training the set of machine learning models. Specifically, [0032] teaches “the objective function used for joint training can be formulated in a way to effectively allow the two models 100 and 102 to collaborate and discuss their respective predictions with each other (via the path 110) to help each model learn how the other model thinks, which factors into its own training”).
Aslan is also in the same field of endeavor as Sun and Wu (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of sharing information between the models employing an objective function, as being disclosed and taught by Aslan, in the system taught by Sun and Wu to yield the predictable results of “machine learning models that are trained using the techniques described herein can perform better (in terms of the accuracy of the model output) than conventionally-trained machine learning models in some scenarios”. And “providing more flexibility in model training, as compared to current training methods” ([0008] -[0009]).
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Sun, Wu, Aslan in further view of Liu et al. US 20210142164 A1 (hereinafter Liu).
Regarding claim 10:
Sun, Wu, and Aslan teach The method according to claim 9. Neither Sun, Wu or Aslan teach wherein a minimum Mean Squared Error is employed as the objective for all the knowledge sharing losses.
Nevertheless, Liu teaches the following:
wherein a minimum Mean Squared Error is employed as the objective for all the knowledge sharing losses ( Liu [0067] teaches the distillation objective is to minimize the mean squared error (MSE) between the student network logits against the teacher’s logits. Thus, teaching a minimum MSE being used as the objective).
Liu is also in the same field of endeavor as Sun, Wu and Aslan (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of a minimum MSE, as being disclosed and taught by Liu, in the system taught by Sun, Wu and Aslan to yield the predictable results of “provide a general distillation framework or architecture that is applicable to either transformer-based models or other models as well” (Liu [0036]) such that a simple neural network model performance and inference speed can be improved (see Liu [0037]).
Claim 13 is rejected under 35 U.S.C. 103 as being unpatentable over Wu, Sun, and Arani.
Regarding claim 13:
Sun and Wu tech the method of claim 1. Specifically, Wu teaches (i) the artificial cognitive architecture according to claim 1, wherein the cognitive architecture continues to train the first neural network module, and (ii) the second neural network module, wherein said second neural network has been trained together with the first module using the method according to claim 1 (Wu Figure 2, teaches the Efficient CLS architecture (i.e., artificial cognitive architecture) and teaches the fast learner (i.e., first neural network module) and the slow learner ( i.e., second neural network module) being trained together).
Neither Sun or Wu specifically teach an at least partially autonomous driving system comprising: at least one camera designed for providing a visual data stream for visual data samples, and a computer designed for classifying and/or detecting objects using...
Nevertheless, Arani teaches the following:
An at least partially autonomous driving system comprising: ( Arani [0086] teaches a semi-autonomous vehicle (i.e., partially autonomous driving system)).
at least one camera designed for providing a visual data stream for visual data samples, and a computer designed for classifying and/or detecting objects using: ( Arani [0086] teaches the semi-autonomous vehicle include one or more sensors such as a camera and a LiDAR, to capture images of the environment surrounding the vehicle and [0098] teaches a model can be used to “recognize and classify the objects depicted in the images, and make certain predictions based on knowledge learned through the training process”).
Arani is also in the same field of endeavor as Sun and Wu (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of an autonomous driving system comprising sensors such as cameras as being disclosed and taught by Arani, in the system taught by Sun and Wu to yield the predictable results of providing “improved methods and systems that train and use artificial intelligence inference models that can take advantage of the interplay between rapid instance-based learning and slow structured learning” (Arani [0006]).
Claims 14-15 are rejected under 35 U.S.C. 103 as being unpatentable over Sun, Wu and Geirhos in further view of Wang et al. Escaping Saddle Points Faster with Stochastic Momentum (hereinafter Wang).
Regarding claim 14:
Sun, Wu and Geirhos teach The method of claim 2 wherein the step of consolidating knowledge from the first neural network module in a first sub-module of the plurality of neural network sub-modules of the second module as an implicit memory is performed through stochastic updates ( Sun pg. 13, Figure. 1.1 teaches the CLARION architecture which is a two levels type architecture that comprises a top level module (i.e., first neural network module ) and a bottom level module (i.e., second neural network module) in which knowledge/information is being shared therefore knowledge will inheritably by consolidated an implicit memory. Further, since Sun pg. 63, para. 1, teaches the bottom level module comprises multiple networks (i.e., plurality of neural network sub-modules) someone ordinary skilled in the art will recognize that knowledge will also be consolidated into the multiple networks. In addition, Sun pg. 41, sec: 2.4 Integrating the outcomes of the two levels, teaches a number of method can be used for integrating (consolidating) the outcomes (information/knowledge) from two levels this include stochastic selection which can be view as a form of “stochastic update”).
Sun does not explicitly teaches stochastic momentum. Nevertheless, Wang teaches the following:
...stochastic momentum... (Wang Abstract teaches “Stochastic gradient descent (SGD) with stochastic momentum is popular in non-convex stochastic optimization and particularly for the training of deep neural networks”).
Wang is also in the same field of endeavor as Sun, Wu and Geirhos (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of Stochastic gradient descent with stochastic momentum as being disclosed and taught by Wang, in the system taught by Sun, Wu and Geirhos to yield the predictable results of adding stochastic momentum to Stochastic gradient descent such that it improves deep network training (Wang Abstract).
Regarding claim 15:
Sun, Wu and Geirhos teach The method of claim 3 wherein the step of consolidating information from the second sub-module within the third sub-module is performed through stochastic updates ( Sun pg. 13, Figure. 1.1 teaches the CLARION architecture which is a two levels type architecture that comprises a top level module (i.e., first neural network module ) and a bottom level module (i.e., second neural network module) in which knowledge/information is being shared therefore knowledge will inheritably by consolidated an implicit memory. Further, since Sun pg. 63, para. 1, teaches the bottom level module comprises multiple networks (i.e., plurality of neural network sub-modules) this can include a first, second and third sub-module for which someone ordinary skilled in the art will recognize that knowledge will also be consolidated into the multiple networks. In addition, Sun pg. 41, sec: 2.4 Integrating the outcomes of the two levels, teaches a number of method can be used for integrating (consolidating) the outcomes (information/knowledge) from two levels this include stochastic selection which can be view as a form of “stochastic update”).
Sun does not explicitly teaches stochastic momentum. Nevertheless, Wang teaches the following:
...stochastic momentum... (Wang Abstract teaches “Stochastic gradient descent (SGD) with stochastic momentum is popular in non-convex stochastic optimization and particularly for the training of deep neural networks”).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to GISEL G FACCENDA whose telephone number is (703)756-1919. The examiner can normally be reached Monday - Friday 8:00 am - 4:00 pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Abdullah Al Kawsar can be reached at (571) 270-3169. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/G.G.F./Examiner, Art Unit 2127
/ABDULLAH AL KAWSAR/Supervisory Patent Examiner, Art Unit 2127