Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Detailed Action
This action is in response to the claims filed 04/10/2023:
Claims 1 – 20 are pending.
Claims 1, 12, and 20 are independent.
Claim Objections
Claim 2 objected to because of the following informalities: "The obtain" should read "the obtaining". 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 1-20 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.
Regarding claims 1, 2, 6, 8, 12, 13, 17, and 20, "the given signal sequence" lacks antecedent basis. Claim 1 introduces "at least one given signal sequence" such that it would be unclear to one of ordinary skill in the art which of the at least one given signal sequence corresponds to "the given signal sequence". It would further be unclear how to determine the scope of the claim in the case that a plurality of given signal sequences were used (if "the given signal sequence" should remain singular or should refer to a plurality of given signal sequences).
Regarding claims 1, 12, and 20, "the metadata representations" in “based on the metadata representations” lacks antecedent basis. Specifically, it's unclear if "the metadata representations" refers to the "one or more metadata representations" or the "at least one respective metadata representation" or both, previously introduced in claim 1.
The remaining claims are rejected with respect to their dependence on the rejected claims.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
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 1-7, 9, 12-16, 18, and 20 are rejected under U.S.C. §103 as being unpatentable over the combination of Yao (“AUTOMATED RELATIONAL META-LEARNING”, 2020) and Jin (US11514329B2).
Regarding claim 1, Yao teaches obtain: (a) one or more metadata representations, each representing metadata relating to a signal type of one or more signal types;([p. 2 §1] "To learn the meta-knowledge graph at meta-training time, for each task, we construct a prototype-based relational graph for each class, where each vertex represents one prototype" [p. 4] "For each task Ti, ARML first builds a prototype-based relational structure Ri by mapping the training samples Dtri into prototypes, with each prototype represents one class" Prototype interpreted as metadata relating to a signal type (class) of one or more signal types)
(b) one or more signal sequences, each signal sequence is an ordered sequence of values associated with a given signal type of the one or more signal types; and([p. 2 §3] "Considering a task Ti, the goal of few-shot learning is to learn a model with a dataset Di = Dtr i Dts i , where the labeled training set Dtr i = xtr j ytr j [1Ntr] only has a few samples and Dts i represents the corresponding test set" [p. 3] "In meta-learning, a sequence of tasks {T1,...TI} are sampled from a task-level probability distribution p(T), where each one is a few-shot learning task" [p. 4] "For each task Ti, ARML first builds a prototype-based relational structure Ri by mapping the training samples Dtri into prototypes, with each prototype represents one class" Task interpreted as synonymous with signal associated with a plurality of signal types (classes) in an ordered sequence of 1 to I. Alternatively the task dataset can reasonably be interpreted as an ordered signal sequence of values associated with a given signal type)
(c) at least one new signal sequence, being a new ordered sequence of values, each associated with a label, the at least one new signal sequence is associated with corresponding at least one new signal type, not included in the one or more signal types;([p. 2 §3] "Considering a task Ti, the goal of few-shot learning is to learn a model with a dataset Di = Dtr i Dts i , where the labeled training set Dtr i = xtr j ytr j [1Ntr] only has a few samples and Dts i represents the corresponding test set" [p. 8] "all datasets are divided into meta-training, meta validation and meta-testing classes" meta-testing subset of task dataset interpreted as a new signal sequence being a new ordered sequence of values associated with a label. The meta-testing subset is not included in the task set, meta-training class set, or meta-validation class set.)
train a meta learner autoencoder, capable of mapping at least one given signal sequence and at least one respective metadata representation, being the metadata representation representing the signal type of the given signal sequence into a meta representation vector, ([p. 5] "To encourage such stability, we introduce two reconstructions by utilizing two auto-encoders. There are two collections of parameters, i.e, CRi and CRi" [p. 6] "Aggregate CRi in equation 8 and CRi in equation 9 to get the representations qi, ti and reconstruction loss Lq, Lt" [p. 6] "The reconstruction errors in Equations 8 and 9 pose an extra constraint to enhance the training stability, leading to improvement of task representation learning." Sequence of C^_Ri from i to I interpreted as new task representation sequence. See algorithm 1 lines 8-12, especially line 12 where reconstruction loss is explicitly used to update parameters. See also Eqn. 1, Eqn. 8, Eqn. 9, and Eqn. 11. Specifically, qi and ti are necessarily vectors as the output of mean pooling is definitionally a vector. Performing gradient descent process interpreted as synonymous with training)
wherein the trained meta learner autoencoder comprises a meta learner encoder and a meta learner decoder;([p. 7] "For the aggregated function in autoencoder structure (AGt, AGt dec AGq, AGq dec), we use the GRU as the encoder and decoder in this structure")
determine, based on the metadata representations, a predicted metadata representation representing the at least one new signal type; and([p. 2] "we construct a prototype-based relational graph for each class, where each vertex represents one prototype. The prototype-based relational graph not only captures the underlying relationship behind samples, but alleviates the potential effects of abnormal samples. The meta-knowledge graph is then learned" [p. 5] "we construct a super-graph Si by connecting the prototype-based relational graph Ri with the meta-knowledge graph G for each task Ti" [p. 5 §4.2] "After constructing the super-graph Si, we are able to propagate the most relevant knowledge from meta-knowledge graph G to the prototype-based relational graph Ri by introducing a Graph Neural Networks (GNN)")
train, by utilizing the meta representation vector mapped by the meta learner encoder from the at least one new signal sequence and the predicted metadata representation, a new task model, capable of receiving one or more unlabeled signal sequences associated with the at least one new signal type, and predicting, utilizing the predicted metadata representation and the meta learner encoder, for each of the unlabeled signal sequences, a corresponding label.([p. 2] "A learning model (a.k.a., base model) f with parameters are used to evaluate the effectiveness on Dts i byminimizing the expected empirical loss" [p. 6] "After getting the task representation qi and ti, the modulating function is then used to tailor the task-specific information to the globally shared initialization 0" [p. 6] "Then, the corresponded task representation qi of CRi is summarized by applying a mean pooling operator over prototypes on the encoded dense representation [...] Similarly, we reconstruct CRi and get the corresponded task representation ti [...] For each task Ti, we perform the gradient descent process from 0i and reach its optimal parameter i. Combining the reconstruction loss Lt and Lq with the meta-learning loss defined in equation 1" [p. 8] "The traditional N-way K-shot settings are used to split training and test set for each task" base model f interpreted as new task model capable of receiving one or more unlabeled signal sequences (task level test set/query set) associated with the at least one new signal type (class dataset samples are interpreted as being unlabeled at prediction, validation is used to verify the labels and determine accuracy)).
However, Yao does not explicitly teach A system comprising a processing circuitry configured to.
Jin, in the same field of endeavor, teaches A system comprising a processing circuitry configured to: ("a system can comprise a memory that stores computer executable components and a processor that executes the computer executable components stored in the memory").
Yao as well as Jin are directed towards machine learning with autoencoders. Therefore, Yao as well as Jin are reasonably pertinent analogous art. It would have been obvious before the effective filing date of the claimed invention to adopt Jin's low-dimensional feature extraction via autoencoder as the basis for Yao's prototypes/task representation, yielding a meta representation vector in lower dimension than the original signal sequence. Jin provides as additional motivation for combination ([Col. 7 l. 43-63] "System 100 and/or one or more components of system 100 accomplish these goals by using a semi-supervised feature extraction method to extract low dimensional features that are the most important to the DNN model purpose (for example, classification), and tailoring application of the DNN model in that domain to improve the generalization"). This motivation for combination also applies to the remaining claims which depend on this combination.
Regarding claim 2, the combination of Yao and Jin teaches The system of claim 1, wherein: (a) the obtain further includes obtaining a knowledge graph embedding of a knowledge graph, (Yao [p. 4 §4.2] "we construct and maintain a meta-knowledge graph. The vertices represent different types of meta-knowledge (e.g., the common contour between aircrafts and birds) and the edges are automatically constructed to reflect the relationship between meta-knowledge" [p. 4 §4.2] "Assuming the representation of an vertex g is given by hg Rd, we define the meta-knowledge graph as G = (HGAG), where HG = hj j [1G] RG d and AG = AG(hj hm) jm [1G] RG Gdenote the vertex feature matrix and vertex adjacency matrix, respectively. To better explain the construction of the meta-knowledge graph, we first discuss the vertex representation HG. During meta-training, tasks arrive one after another in a sequence and their corresponding vertices" H_G interpreted as knowledge graph embedding vector representation comprising one or more vector representations)
wherein (i) the knowledge graph comprises a plurality of nodes, each node representing metadata relating to a signal type of one or more signal types, and a plurality of edges, each connecting two given nodes of the nodes and each representing a relationship between the two given nodes, and (Yao [p. 2] "each vertex represents one type of meta-knowledge (e.g., the common contour between birds and aircrafts) […] each vertex represents one prototype" [p. 4 §4.2] "Assuming the representation of an vertex g is given by hg Rd, we define the meta-knowledge graph as G = (HGAG), where HG = hj j [1G] RG d and AG = AG(hj hm) jm [1G] RG Gdenote the vertex feature matrix and vertex adjacency matrix, respectively Adjacency matrix interpreted as a plurality of edges connecting two given nodes of the nodes and each representing a relationship between the two given nodes)
(ii) the knowledge graph embedding comprises one or more vector representations, each representing a node of the nodes or an edge of the edges;(Yao [p. 4 §4.2] "Assuming the representation of an vertex g is given by hg Rd, we define the meta-knowledge graph as G = (HGAG), where HG = hj j [1G] RG d and AG = AG(hj hm) jm [1G] RG Gdenote the vertex feature matrix and vertex adjacency matrix, respectively. To better explain the construction of the meta-knowledge graph, we first discuss the vertex representation HG. During meta-training, tasks arrive one after another in a sequence and their corresponding vertices" H_G interpreted as knowledge graph embedding vector representation comprising one or more vector representations (vertices, which Yao explicitly defines as vectors that exist in vector space of d dimensions))
(b) the one or more metadata representations are the corresponding one or more vector representations; (c) the at least one respective metadata representation is the at least one respective vector representation being the vector representation representing the node related to the signal type of the given signal sequence; and(Yao [p. 4] "we denote the prototype-based relational graph as Ri = (CRi ARi), where CRi = cj i j [1K] RK d represent a set of vertices, with each one corresponds to the prototype from a class" Yao explicitly states that Cri is a vector representation corresponding to a prototype (metadata) for each class (signal type))
(d) the predicted metadata representation is a predicted vector representation representing the at least one new signal type and determined based on the knowledge graph.(Yao [p. 5] "After constructing the super-graph Si, we are able to propagate the most relevant knowledge from meta-knowledge graph G to the prototype-based relational graph Ri by introducing a Graph Neural Networks (GNN) […] we get the information-propagated feature representation for the prototype-based relational graph Ri as the top-K rows of H(L) i , which is denoted as CRi" See also Eqn. 7 such that C^_Ri is interpreted as predicted (propagated) metadata representations determined based on the knowledge graph).
Regarding claim 3, the combination of Yao and Jin teaches The system of claim 1, wherein the processing circuitry is further configured to train a new task autoencoder, capable of mapping the at least one new signal sequence into a new task representation sequence, wherein the trained new task autoencoder comprises a new task encoder and a new task decoder; and wherein the training of the new task model further utilizes the new task representation sequence mapped by the new task encoder from the at least one new signal sequence.(Yao [p. 5] "To encourage such stability, we introduce two reconstructions by utilizing two auto-encoders. There are two collections of parameters, i.e, CRi and CRi" [p. 6] "Aggregate CRi in equation 8 and CRi in equation 9 to get the representations qi, ti and reconstruction loss Lq, Lt" [p. 6] "The reconstruction errors in Equations 8 and 9 pose an extra constraint to enhance the training stability, leading to improvement of task representation learning." Sequence of C^_Ri from i to I interpreted as new task representation sequence. See algorithm 1 lines 8-12, especially line 12 where reconstruction loss is explicitly used to update parameters.).
Regarding claim 4, the combination of Yao and Jin teaches The system of claim 1, wherein the processing circuitry is further configured to: receive one or more unlabeled signal sequences associated with the at least one new signal type; and(Yao [p. 2 §3] "Considering a task Ti, the goal of few-shot learning is to learn a model with a dataset Di = Dtr i Dts i , where the labeled training set Dtr i = xtr j ytr j [1Ntr] only has a few samples and Dts i represents the corresponding test set" test set interpreted as unlabeled signal sequences associated with the at least one new signal type (class corresponding to signal (Task Ti)))
predict, for each of the unlabeled signal sequences, a corresponding predicted label.(Yao [p. 8] "For each dataset, the performance accuracy with 95% confidence interval is reported" Accuracy is percent of predicted labels from the unlabeled signal sequence that match the ground truth labels. See Table 1 where the accuracy for each respective class is given.).
Regarding claim 5, the combination of Yao and Jin teaches The system of claim 4, wherein the prediction of the corresponding predicted label by the new task model can be one or more of: anomaly detection prediction or classification prediction.(Yao [p. 9 §5.3] "Table 1: Overall few-shot classification results (accuracy 95% confidence) on Plain-Multi dataset.").
Regarding claim 6, the combination of Yao and Jin teaches The system of claim 1, wherein the meta representation vector has a lower dimension than the given signal sequence.(Jin [Col. 4 l. 24-40] "the first step involves extracting a set of low-dimensional, related data features from the training data set (or reference data set) that was used to train the DNN model. The set of low-dimensional features extracted from the training/reference data set is referred to herein as first training data features" [Col. 4 l. 50-67] "At 414, a second degree of correspondence can then be determined (e.g., by the model analysis component 116) between the second training data features 406 and the second target data features 412 (e.g., using Z-score analysis on each dimension, Mahalanobis distance analysis on multi-dimensions, t-SNE analysis on lower dimensions to get similarity distances" [Col. 8 l. 59-Col. 9 l. 6] "the data feature extraction component 104 can employ a suitable feature extraction method that provides for automatically extracting a set of features or feature vectors for both data sets that reduce the dimensionality of the respective data sets to a smaller set of representative feature vectors.").
Regarding claim 7, the combination of Yao and Jin teaches The system of claim 3, wherein the new task representation sequence has a lower dimension than the at least one new signal sequence.(Jin [Col. 8 l. 59-Col. 9 l. 6] "the data feature extraction component 104 can employ a suitable feature extraction method that provides for automatically extracting a set of features or feature vectors for both data sets that reduce the dimensionality of the respective data sets to a smaller set of representative feature vectors.").
Regarding claim 9, the combination of Yao and Jin teaches The system of claim 2, wherein the determination of the predicted vector representation is performed by: (a) adding a new node representing metadata relating to the at least one new signal type to the knowledge graph, (Yao [p. 2] "each vertex represents one type of meta-knowledge (e.g., the common contour between birds and aircrafts) […] each vertex represents one prototype" [p. 4 §4.2] "Assuming the representation of an vertex g is given by hg Rd, we define the meta-knowledge graph as G = (HGAG), where HG = hj j [1G] RG d and AG = AG(hj hm) jm [1G] RG Gdenote the vertex feature matrix and vertex adjacency matrix, respectively Adjacency matrix interpreted as a plurality of edges connecting two given nodes of the nodes and each representing a relationship between the two given nodes)
(b) adding one or more new edges, each new edge of the new edges connecting the new node with one of the nodes in the knowledge graph, and (Yao [p. 4 §4.2] "Assuming the representation of an vertex g is given by hg Rd, we define the meta-knowledge graph as G = (HGAG), where HG = hj j [1G] RG d and AG = AG(hj hm) jm [1G] RG Gdenote the vertex feature matrix and vertex adjacency matrix, respectively. To better explain the construction of the meta-knowledge graph, we first discuss the vertex representation HG. During meta-training, tasks arrive one after another in a sequence and their corresponding vertices" H_G interpreted as knowledge graph embedding vector representation comprising one or more vector representations (vertices, which Yao explicitly defines as vectors that exist in vector space of d dimensions))
(c) determining the predicted vector representation using similarities of the new node to the nodes.(Yao [p. 5] "The link between prototype cj i in prototype-based relational graph and vertex hm in meta knowledge graph is weighted by the similarity between them. More precisely, for each prototype cj i , the link weight AS(cj i hm) is calculated by applying softmax over Euclidean distances between cj i and hm m [1G]").
Regarding claims 12-16 and 18, claims 12-16 and 18 are directed towards the method performed by the system of claims 1-5 and 9, respectively. Therefore, the rejection applied to claims 1-5 and 9 also apply to claims 12-16 and 18.
Regarding claim 20, claim 20 is substantially similar to claim 1. Therefore, the rejection applied to claim 1 also applies to claim 20.
Claims 8 and 17 are rejected under U.S.C. §103 as being unpatentable over the combination of Yao, Jin, and Rombach (“High-Resolution Image Synthesis with Latent Diffusion Models”, 2022).
Regarding claim 8, the combination of Yao and Jin teaches The system of claim 1.
However, the combination of Yao and Jin doesn't explicitly teach, wherein the training of the meta learner autoencoder is performed by: (a) adding noise to at least part of the given signal sequence giving rise to a given noised signal sequence and/or masking at least part of the given signal sequence giving rise to a given masked signal sequence,
(b) linking the given noised signal sequence and/or the given masked signal sequence with the predicted metadata representation,
and (c) reconstructing the given signal sequence.
Rombach, in the same field of endeavor, teaches The system of claim 1, wherein the training of the meta learner autoencoder is performed by: (a) adding noise to at least part of the given signal sequence giving rise to a given noised signal sequence and/or masking at least part of the given signal sequence giving rise to a given masked signal sequence, ([p. 4] "These models can be interpreted as an equally weighted sequence of denoising autoencoders (xt t); t = 1 T, which are trained to predict a denoised variant of their input xt, where xt is a noisy version of the input x" See also Eqn. 1)
(b) linking the given noised signal sequence and/or the given masked signal sequence with the predicted metadata representation, ([p. 4] "Similar to other types of generative models [56, 83], diffusion models are in principle capable of modeling conditional distributions of the form p(zy). This can be implemented with a conditional denoising autoencoder (zt t y) and paves the way to controlling the synthesis process through inputs y such as [...] semantic maps")
and (c) reconstructing the given signal sequence.([p. 4] "samples from p(z) can be decoded to image space with a single pass through D" Decoded interpreted as synonymous with reconstructed).
The combination of Yao and Jin as well as Rombach are directed towards machine learning with autoencoders. Therefore, the combination of Yao and Jin as well as Rombach are reasonably pertinent analogous art. It would have been obvious before the effective filing date of the claimed invention to substitute the generic autoencoder in Yao with the LDM autoencoder in Rombach. Rombach provides as additional motivation for combination ([Abstract] “Our latent diffusion models (LDMs) achieve new state-of-the-art scores for image inpainting and class-conditional image synthesis and highly competitive performance on various tasks, including text-to-image synthesis, unconditional image generation and super-resolution, while significantly reducing computational requirements compared to pixel-based DMs”).
Regarding claim 17, claim 17 is directed towards the method performed by claim 8. Therefore, the rejection applied to claim 8 also applies to claim 17.
Claims 10, 11, and 19 are rejected under U.S.C. §103 as being unpatentable over the combination of Yao and Jin and Sallab (“Meta learning Framework for Automated Driving”, 2017).
Regarding claim 10, the combination of Yao and Jin teaches The system of claim 1.
However, the combination of Yao and Jin doesn't explicitly teach wherein the signal sequences and the at least one new signal sequence are read from sensors associated with a physical entity.
Sallab, in the same field of endeavor, teaches The system of claim 1, wherein the signal sequences and the at least one new signal sequence are read from sensors associated with a physical entity.([p. 3 ] "two cameras has been added in addition to the central camera to augment the training states with different situations supported with the correct action to take in each").
The combination of Yao and Jin as well as Sallab are directed towards meta-learning. Therefore, the combination of Yao and Jin as well as Sallab are analogous art in the same field of endeavor. It would have been obvious before the effective filing date of the claimed invention to combine the teachings of the combination of Yao and Jin with the teachings of Sallab by applying the meta-learning system in Yao to autonomous vehicles as suggested by Sallab. Sallab provides as additional motivation for combination ([p. 5 §6] “based on the principles of Meta learning and data set aggregation. The proposed algorithm MetaDAgger is shown to be able to generalize on unseen test tracks, achieving much less training time and better sample efficiency”).
Regarding claim 11, the combination of Yao, Jin, and Sallab teaches The system of claim 10, wherein the physical entity is a vehicle.(Sallab [p. 1 §1] "Automated driving development has radically changed during the past few years, driven by advances in Artificial Intelligence (AI), and specifically Deep Learning (DL). Developing an efficient and safe driving policy is in the heart of reaching high level of autonomy in a robot car [...] real car deployment").
Regarding claim 19, claim 19 is directed towards the method performed by the system of claim 10. Therefore, the rejection applied to claim 10 also applies to claim 19.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Yang (US20230377748A1) is directed towards an autoencoder meta-learner system for learning graph neural network embeddings.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SIDNEY VINCENT BOSTWICK whose telephone number is (571)272-4720. The examiner can normally be reached M-F 7:30am-5:00pm EST.
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, Miranda Huang can be reached on (571)270-7092. 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.
/SIDNEY VINCENT BOSTWICK/Examiner, Art Unit 2124