Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first
inventor to file provisions of the AIA .
Response to Arguments
Applicant's arguments filed 12/05/2025 have been fully considered but they are not persuasive.
Regarding applicants arguments for 101 in page 10-12 of applicant’s remarks, the applicant states “Claims 1-20 were rejected under 35 U.S.C. § 101 as being directed to an abstract idea without significantly more.
With respect to Step 2A, Prong 1, the Office Action asserted that the claims recite mental processes and, with respect to Step 2A, Prong 2, that none of the additional independent claim elements integrate the alleged judicial exception into a particular application. Without agreeing that the claims are directed to mental processes, to advance prosecution, Applicant has amended independent claims 1, 8, and 15 to further emphasize patent-eligible features…
Applicant respectfully asserts that the amended independent claims incorporate the alleged judicial exceptions into a practical application (see MPEP §2106.04(d)(l))… For example, the amended independent claims claim improved techniques for handling sequential data transmitted via a network between devices comprising a data handling server and a data source device. More specifically, the amended independent claims claim improved techniques for filling gaps in the sequential data using the data handling server.” The applicant argues how the amended limitations to claim 1 and analogous claims 8 and 15 solve the technical issue mentioned in the claim in paragraph 2 of the specification. However it is noted how the amended limitations only recite receiving data to identify the gaps and using previous sections to fill the data gap. The claims lack specificity in regards to how the modifying dependent data is used to fill the data gap and is generically recited to fill in the data gap using predictions. Claim 1 and analogous claims 8 and 15 do not explain any system glitches resulting in packet loss, outage, etc. Furthermore, the limitation mentioned in the applicant’s remarks are regarding claim 1 limitations that were amended. The amended limitation has not been examined and the argument not convincing. The 101 rejection was updated regarding the amended limitations.
Regarding applicants arguments for 103 in page 12-14 the applicant argues “Claims 1, 2, 4-9, 11-16, and 18-20 were rejected under 35 U.S.C. §103 as being unpatentable over Rawassizadeh, Lujic, and Dempster. Claims 3, 10, and 17 were rejected under § 103 as being unpatentable over Rawassizadeh, Lujic, Dempster, and Pantiskas. In response, Applicant has amended independent claims 1, 8, and 15 to incorporate additional limitations that Applicant asserts serve to clarify and further distinguish the independent claims from the teachings of the cited references… Applicant relies upon the arguments above and respectfully asserts that Dempster and Pantiskas, alone or combined, do not serve to cure the deficiencies in the teachings of Rawassizadeh, Lujic, and Dempster. Based at least on the amendments to the independent claims, Applicant respectfully requests that the §103 rejections be withdrawn.” Applicant argues how the amended limitations overcome the prior art of record. However the amended limitations have not been examined and thus the argument is considered moot and not convincing.
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 1-20 rejected under 35 U.S.C. 101 because the claimed invention is directed to abstract idea without significantly more. The claim(s) recite(s) significantly more. The subject matter eligibility test for products and process is describe below for claim 1 in view of dependent claims.
Regarding claim 1 analogous claims 8 and 15 :
Step 1: Is the claim to a process machine manufacture or composition of matter?
Yes – Claim 1 recites a method, which a method falls under the statutory categories
Step 2A Prong 1: Does the claim recite an abstract idea, law of nature, or natural phenomenon?
Yes – The claim recites the following:
“detecting a ” – The limitation recites a mental process of detecting a data gap. (see MPEP 2106.04(a)(2)III).
“determining, in response to the detecting, a sliding window associated with the data gap,” The limitation recites a mental process of determining a sliding window. (see MPEP 2106.04(a)(2)III).
Step 2 Prong 2: Does the claim recite additional elements that integrate the judicial exception into a particular application? No –
The claim includes the additional element(s):
“A computer-implemented method for handling transmitted via a network between devices comprising a data handling server and a data source device, the method comprising:” - The additional elements fall under “apply it” as using generic computer to handle a data gap in sequential data (See MPEP 2106.05(f)).
“receiving, at the data handling server and via the network, the sequential data from the data source device; and” - The additional elements fall under “insignificant extra-solution activity” mere data gathering (See MPEP 2106.05(g))
“in response to the receiving, processing the sequential data by the data handling server, wherein the processing comprises:” - The additional elements fall under “apply it” as using generic computer to process the sequential data (See MPEP 2106.05(f)).
“wherein the sliding window being based on the timestamp includes a set of the sequential data for a duration of time preceding the timestamp, the set of the sequential data sliding window including dependent data to be used as dependent data for filling the data gap;” The additional elements fall under “insignificant extra-solution activity” mere data gathering as viewed as a whole by including data that precedes a timestamp (See MPEP 2106.05(g))
“as a result of the dependent data of the sliding window including at least one window data gap, masking the at least one window data gap, wherein the masking comprises: generating feature maps based on the dependent data in the sliding window; extracting patterns based on the feature maps; and” The additional elements fall under “apply it” as using generic computer to generate feature maps based on the dependent data and extracting patterns (See MPEP 2106.05(f)).
“modifying the dependent data in the sliding window to include the extracted patterns ” - The additional elements fall under “apply it” as using generic computer to modify the dependent data (See MPEP 2106.05(f)).
“filling the data gap using a prediction generated based on the modified dependent data.” The additional elements fall under “apply it” as using generic computer to fill in data gaps using generate predictions (See MPEP 2106.05(f)).
Step 2B: Does the claim recite additional elements that amount to significantly more than the judicial exception?
No - The claim does not include additional elements that are sufficient to amount to a significantly more than the judicial exemption. As an order whole, the claim is directed to a mental process. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements fall under data gathering and apply it and do not limit the claim. The method does not improve on the function of a computer, transforms an article into another article, nor is it applied by a particular machine, making the claim not patent eligible.
Regarding claim 2 and analogous claims 9 and 16:
Step 2A Prong 2, Step 2B: The additional element(s):
“wherein the extracted patterns are generated based on a random convolutional kernel transform algorithm.”
The additional elements fall under “apply it” as using generic computer to use convolutional kernel transform algorithm to generate extracted patterns (See MPEP 2106.05(f)).
Regarding claim 3 and analogous claims 10 and 17:
Step 2A Prong 2, Step 2B: The additional element(s):
“wherein the random convolutional kernel transform algorithm is a multi-variate time-series classification model” - The additional element falls under the “insignificant extra-solution activity”. The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application.
Regarding claim 4 and analogous claims 11 and 18:
Step 2A Prong 2, Step 2B: The additional element(s):
“wherein the random convolutional kernel transform algorithm performs a convolution with kernels on the dependent data to generate the feature maps, and wherein the extracted patterns correspond to extracted features determined from the feature maps” - The additional elements fall under “apply it” as using generic computer to perform a convolution of kernels to generate features (See MPEP 2106.05(f)). The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application.
Regarding claim 5 and analogous claims 12 and 19:
Step 2A Prong 2, Step 2B: The additional element(s):
“wherein the duration of time of the sliding window immediately precedes the timestamp.” - The additional element falls under the “insignificant extra-solution activity”. The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application.
Regarding claim 6 and analogous claims 13 and 20:
Step 2A Prong 1,
“validating the prediction based on an acceptable threshold for subsequent processing of the sequential data with the prediction;” – The limitation recites a mental process of validating the prediction. (see MPEP 2106.04(a)(2)III).
Step 2A Prong 2, Step 2B: The additional element(s): “as a result of the prediction not being validated, performing a feedback by determining a modified operation in determining a further prediction.” The additional elements fall under “apply it” as using generic computer to perform a feedback and determine a modified operation (See MPEP 2106.05(f)). The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application.
Regarding claim 7 and analogous claim 14:
Step 2A Prong 2, Step 2B: The additional element(s):
“wherein the modified operation is one of using a different feature extraction method, using a different method for prediction, and estimating at least one of a floor and ceiling accuracy.” - The additional elements fall under “apply it” as using generic computer to perform one of different feature extraction method, using a different method for prediction, and estimating at least one of a floor and ceiling accuracy (See MPEP 2106.05(f)). The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application.
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.
Claim(s) 1, 5-8, 12-15, 19, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over I. Lujic, V. De Maio and I. Brandic, "Resilient Edge Data Management Framework," in IEEE Transactions on Services Computing, vol. 13, no. 4, pp. 663-674, 1 July-Aug. 2020, (“Lujic”) and Finn, Chelsea, Ian Goodfellow, and Sergey Levine. "Unsupervised learning for physical interaction through video prediction, (2016), (“Finn”).
Regarding claim 1 and analogous claims 8 and 15, Lujic teaches A computer-implemented method for handling transmitted via a network between devices comprising a data handling server and a data source device, the method comprising (Lujic Page 664 2 Architecture Model Overview para 2-4, Gathering layer transmits IoT measurements to the edge layer to reduce communication costs, save bandwidth and meet latency requirements in distributed sensor networks. Gateways at this layer can aggregate sensor data sending them in an appropriate format and size to the monitoring component. In step (1) data are collected from smart buildings and then in step (2) transferred to the edge layer.
Edge layer manages data through different stages of EDMFrame, to perform accurate and timely analytics. It is composed of edge nodes, e.g., edge servers and micro data centers [10], aiming to perform data processing closer to data sources. EDMFrame includes the following elements [A computer-implemented method for handling a data gap in sequential data]:
Monitoring Component. This component (i) receives and analyses data to detect outliers and missing values, (ii) notifies mediator component about incomplete data, (iii) prepares data for the data recovery mechanism and (iv) triggers IoT actuators based on local edge analytics. It can also extrapolate data characteristics for further analytics [transmitted via a network between devices comprising a data handling server and a data source device,]):
receiving, at the data handling server and via the network, the sequential data for from the data source device; and
(para 664 Fig. 2,
PNG
media_image1.png
301
609
media_image1.png
Greyscale
2 Architecture Model Overview para 3, Edge layer manages data through different stages of EDMFrame, to perform accurate and timely analytics. It is composed of edge nodes, e.g., edge servers and micro data centers [10], aiming to perform data processing closer to data sources [at the data handling server and via the network].
Page 666 Fig 3,
PNG
media_image2.png
316
590
media_image2.png
Greyscale
[the sequential data from the data source device;])
in response to the receiving, processing the sequential data by the data handling server, wherein the processing comprises:
detecting a selecting the data gap in the sequential data, the data gap being at a timestamp (Lujic page 665 Fig 2.
PNG
media_image3.png
368
555
media_image3.png
Greyscale
[in response to the receiving, processing the sequential data by the data handling server],
Page 665, 3.1 Data Preparation, Missing values can occur for different reasons, like system or sensor failures. Once the system/sensor is recovered, the next received data point is stored right after the last generated timestamp. Therefore, to identify a gap, it is necessary to check timestamps. We propose a solution where the monitoring component receives data and stores either corresponding data value or NA for each created timestamp (lines 4-13) [detecting a selecting the data gap in the sequential data, the data gap being at a timestamp;]);
determining, in response to the detecting, a sliding window associated with the data gap, wherein the sliding window includes a set of the sequential data for a duration of time preceding the timestamp, the set of the sequential data as dependent data for filling the data gap (para 365 3.1 Data Preparation para 1 line 11-17, Missing values can occur for different reasons, like system or sensor failures. Once the system/sensor is recovered, the next received data point is stored right after the last generates timestamp. Therefore, to identify a gap, it is necessary to check timestamps. We propose a solution where the monitoring component receives data and stores either corresponding data value or NA for each created timestamp (lines 4-13).
Para 666 3.4 Forecasting Process, After corresponding missing indexes are stored by the preparation component and the first gap identified by the gap identification component, the data processor analyzes predecessor data before the gap. Selected forecasting technique is then applied for the recovering process [determining, in response to the detecting, a sliding window associated with the data gap, wherein the sliding window being based on the timestamp includes a set of the sequential data ]… If seasonality occurs in time series, by checking periodicity, the data processor can forward that information to the next component. Users can also specify additional information about the data, such as a monitoring frequency, e.g., if temperature data are collected every five minutes, then the seasonal parameter value 288, representing the expected daily seasonality (12 * 24), is included in the forecasting procedure. Once all necessary parameters are forwarded from the data processor, the forecasting process can start. Missing values are replaced in the original dataset, and their indexes are removed from the vector v. Once the current gap is recovered, the next gap (if exists) is considered in a new cycle. The recovering process stops when no more missing values are left in nom. [for a duration of time preceding the timestamp, the set of the sequential data as dependent data for filling the data gap;]);
Lujic does not explicitly teach as a result of the dependent data of the sliding window including at least one window data gap, masking the at least one window data gap, wherein the masking comprises: generating feature maps based on the dependent data in the sliding window; extracting patterns based on the feature maps; and modifying the dependent data in the sliding window to include the extracted
However Finn teaches
as a result of the dependent data of the sliding window including at least one window data gap, masking the at least one window data gap, wherein the masking comprises (Finn Page 1 1 Introduction para 2 line 5-10, Such models follow a paradigm of reconstructing future frames from the internal state of the model. In our approach, we propose a method which does not require the model to store the object and background appearance. Such appearance information is directly available in the previous frame. We develop a predictive model which merges appearance information from previous frames with motion predicted by the model. As a result, the model is better able to predict future video sequences for multiple steps, even involving objects not seen at training time [as a result of the dependent data of the sliding window including at least one window data gap].
Page 3 3 Motion-Focused Predictive Models,
In order to learn about object motion while remaining invariant to appearance, we introduce a class of video prediction models that directly use appearance information from previous frames to construct pixel predictions. Our model computes the next frame by first predicting the motions of image segments, then merges these predictions via masking. In this section, we discuss our novel pixel transformation models, and propose how to effectively merge predicted motion of multiple segments into a single next image prediction. The architecture of the CDNA model is shown in Figure 1. Diagrams of the DNA and STP models are in Appendix B [masking the at least one window data gap, wherein the masking comprises:]):
generating feature maps based on the dependent data in the sliding window; extracting patterns based on the feature maps (Finn Page 3,
PNG
media_image4.png
261
760
media_image4.png
Greyscale
[extracting patterns based on the feature maps;]
Figure 1: Architecture of the CDNA model, one of the three proposed pixel advection models. We use convolutional LSTMs to process the image, outputting 10 normalized transformation kernels from the smallest middle layer of the network and an 11-channel compositing mask from the last layer (including 1 channel for static background). The kernels are applied to transform the previous image into 10 different transformed images, which are then composited according to the masks. The masks sum to 1 at each pixel due to a channel-wise softmax. Yellow arrows denote skip connections.
Page 5 3.3 Action-conditioned Convolutional LSTMs para 2 line 1-4,
The model architecture is displayed in Figure 1 and detailed in Appendix B. In an interactive setting, the agent’s actions and internal state (such as the pose of the robot gripper) influence the next image. We integrate both into our model by spatially tiling the concatenated state and action vector across a feature map, and concatenating the result to the channels of the lowest-dimensional activation map [generating feature maps based on the dependent data in the sliding window;]);
and modifying the dependent data in the sliding window to include the extracted (Finn page 4 3.2 Composing Object Motion Predictions para 3, For each model, including DNA, we also include a “background mask” where we allow the models to copy pixels directly from the previous frame. Besides improving performance, this also produces interpretable background masks that we visualize in Section 5. Additionally, to fill in previously occluded regions, which may not be well represented by nearby pixels, we allowed the models to generate pixels from an image, and included it in the final masking step [modifying the dependent data in the sliding window to include the extracted patterns].);
and filling the data gap using a prediction generated based on the modified dependent data (Finn Page 4 3.2 Composing Object Motion Predictions para 1-3
PNG
media_image5.png
144
758
media_image5.png
Greyscale
In practice, our experiments show that the CDNA and STP models learn to mask out objects that are moving in consistent directions. The benefit of this approach is two-fold: first, predicted motion transformations are reused for multiple pixels in the image, and second, the model naturally extracts a more object centric representation in an unsupervised fashion, a desirable property for an agent learning to interact with objects. The DNA model lacks these two benefits, but instead is more flexible as it can produce independent motions for every pixel in the image.
For each model, including DNA, we also include a “background mask” where we allow the models to copy pixels directly from the previous frame. Besides improving performance, this also produces interpretable background masks that we visualize in Section 5. Additionally, to fill in previously occluded regions, which may not be well represented by nearby pixels, we allowed the models to generate pixels from an image, and included it in the final masking step [based on the modified dependent data].
page 5, 3.3 Action-conditioned Convolutional LSTMs
Most existing physics and video prediction models use feedforward architectures [17, 15] or feedforward encodings of the image [20]. To generate the motion predictions discussed above, we employ stacked convolutional LSTMs [28]. Recurrence through convolutions is a natural fit for multi-step video prediction because it takes advantage of the spatial invariance of image representations, as the laws of physics are mostly consistent across space. As a result, models with convolutional recurrence require significantly fewer parameters and use those parameters more efficiently.
The model architecture is displayed in Figure 1 and detailed in Appendix B. In an interactive setting, the agent’s actions and internal state (such as the pose of the robot gripper) influence the next image. We integrate both into our model by spatially tiling the concatenated state and action vector across a feature map, and concatenating the result to the channels of the lowest-dimensional activation map. Note, though, that the agent’s internal state (i.e. the robot gripper pose) is only input into the network at the beginning, and must be predicted from the actions in future timesteps. We trained the networks using an l2 reconstruction loss. Alternative losses, such as those presented in [17] could complement this method [filling the data gap using a prediction generated]).
Lujic and Finn are considered to be analogous to the claim invention because they are in the same field data imputation. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Lujic to incorporate the teachings of Finn and mask missing values with previous segments using feature maps. Doing so to be able to combine predictions into a single image and provide interpretable internal representations (Finn page 2 1 Introduction para 3, To merge appearance and predicted motion, we output the motion of pixels relative to the previous image. Applying this motion to the previous image forms the next frame. We present and evaluate three motion prediction modules. The first, which we refer to as dynamic neural advection (DNA), outputs a distribution over locations in the previous frame for each pixel in the new frame. The predicted pixel value is then computed as an expectation under this distribution. A variant on this approach, which we call convolutional dynamic neural advection (CDNA), outputs the parameters of multiple normalized convolution kernels to apply to the previous image to compute new pixel values. The last approach, which we call spatial transformer predictors (STP), outputs the parameters of multiple affine transformations to apply to the previous image, akin to the spatial transformer network previously proposed for supervised learning [11]. In the case of the latter two methods, each predicted transformation is meant to handle separate objects. To combine the predictions into a single image, the model also predicts a compositing mask over each of the transformations. DNA and CDNA are simpler and easier to implement than STP, and while all models achieve comparable performance, the object-centric CDNA and STP models also provide interpretable internal representations.).
Regarding claim 5 and analogous claims 12 and 19, Lujic in view of Finn teaches the computer-implemented method of claim 1.
Lujic teaches wherein the duration of time of the sliding window immediately precedes the timestamp (Lujic page 665 3.1 Data Preparation para line 12-16, Missing values can occur for different reasons, like system or sensor failures. Once the system/sensor is recovered, the next received data point is stored right after the last generated timestamp. Therefore, to identify a gap, it is necessary to check timestamps.
Page 666 3.4 Forecasting Process para 1 line 4-10, After corresponding missing indexes are stored by the preparation component and the first gap identified by the gap identification component, the data processor analyzes predecessor data before the gap. Selected forecasting technique is then applied for the recovering process [wherein the duration of time of the sliding window immediately precedes the timestamp].).
Regarding claim 6 and analogous claims 13 and 20, Lujic in view of Finn teaches the computer-implemented method of claim 1.
Lujic teaches further comprising: validating the prediction based on an acceptable threshold for subsequent processing of the sequential data with the prediction (
Lujic, 4 Edge Storage Management Page 667 Para. 3, Validation of the Specification List. This phase checks the user-defined specification list. During the execution of the proposed algorithm, users can update the specification list anytime, e.g., setting forecast accuracy threshold [based on an acceptable threshold], a new forecast horizon or different forecast methods. This list has to be checked each time a cycle starts since any changes made to it can affect the whole edge storage management [for subsequent processing of the sequential data with the prediction] (See Page 667 Fig. 4));
as a result of the prediction not being validated, performing a feedback by determining a modified operation in determining a further prediction (
Lujic 4 Edge Storage Management Page 667 Para. 2 Learning Phase. This phase derives information about data, such as time series pattern recognition, used to determine the most appropriate method for that specific pattern [28], or seasonality over a certain period, used to set up a forecast method [29]. This phase is executed only once and provides information used by all the other phases.
Lujic 4 Edge Storage Management Page 667 Para. 4, Multiple Forecast Iteration on the Available Dataset. This phase takes one of the forecasting methods (in our case ETS or ARIMA) with accuracy threshold (
f
t
h
a
c
) and forecast horizon (
f
h
) from the specification list [as a result of the prediction not being validated]. The available dataset is divided into training and test data. Test data are equal to the number of data points specified by the user in the specification list (i.e.,
f
h
)). The amount of training data is reduced in each iteration by a certain amount of data to find parts of the dataset resulting in required forecast accuracy. At the end of each iteration, forecast accuracy measures are added in the vector γ to be used in the next phase [performing a feedback by determining a modified operation in determining a further prediction]).
Regarding claim 7 and analogous claim 14, Lujic in view of Finn teaches the computer-implemented method of claim 1.
Lujic further teaches wherein the modified operation is one of using a different feature extraction method, using a different method for prediction, and estimating at least one of a floor and ceiling accuracy (Lujic 4 Edge Storage Management Page 667 Para. 4, Multiple Forecast Iteration on the Available Dataset. This phase takes one of the forecasting methods (in our case ETS or ARIMA) with accuracy threshold (
f
t
h
a
c
) and forecast horizon (
f
h
) from the specification list [using a different method for prediction]. The available dataset is divided into training and test data. Test data are equal to the number of data points specified by the user in the specification list (i.e.,
f
h
)). The amount of training data is reduced in each iteration by a certain amount of data to find parts of the dataset resulting in required forecast accuracy. At the end of each iteration, forecast accuracy measures are added in the vector γ to be used in the next phase).
Claim(s) 2, 9, 16, 11, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Lujic in view of Finn and further in view of Dempster et al. “ROCKET: Exceptionally fast and accurate time series classification using random convolutional kernels”, (2019) (“Dempster”).
Regarding claim 2 and analogous claims 9 and 16, Lujic in view of Finn teaches the computer-implemented method of claim 1.
Lujic and Finn are combinable for the same rationale as set forth above with respect to claim 1 and analogous claims 8 and 15.
Lujic does not explicitly teach wherein the extracted patterns are generated based on a random convolutional kernel transform algorithm.
However Dempster teaches wherein the extracted patterns are generated based on a random convolutional kernel transform algorithm (Dempster Page 6 3 Method Para 2.
3. In particular, Rocket [random convolutional kernel transform algorithm] makes key use of kernel dilation. In contrast to the typical use of dilation in convolutional neural networks, where dilation increases exponentially with depth (e.g., Yu and Koltun 2016; Bai et al. 2018; Franceschi et al. 2019), we sample dilation randomly for each kernel, producing a huge variety of kernel dilation, capturing patterns at different frequencies and scales, which is critical to the performance of the method (see section 4.3.4, below).
Lujic and Dempster are considered to be analogous to the claim invention because they are in the same field of time series analysis. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Lujic to incorporate the teachings of Dempster by implementing a random convolutional kernel transform algorithm. Doing so to take advantage of state of the art accuracy and using less computational resources (Dempster Abstract, Most methods for time series classification that attain state-of-the-art accuracy have high computational complexity, requiring significant training time even for smaller datasets, and are intractable for larger datasets. Additionally, many existing methods focus on a single type of feature such as shape or frequency. Building on the recent success of convolutional neural networks for time series classification, we show that simple linear classifiers using random convolutional kernels achieve state-of-the-art accuracy with a fraction of the computational expense of existing methods).
Lujic and Dempster are considered to be analogous to the claim invention because they are in the same field of time series analysis. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Lujic to incorporate the teachings of Dempster by implementing a random convolutional kernel transform algorithm. Doing so to take advantage of state of the art accuracy and using less computational resources (Dempster Abstract, Most methods for time series classification that attain state-of-the-art accuracy have high computational complexity, requiring significant training time even for smaller datasets, and are intractable for larger datasets. Additionally, many existing methods focus on a single type of feature such as shape or frequency. Building on the recent success of convolutional neural networks for time series classification, we show that simple linear classifiers using random convolutional kernels achieve state-of-the-art accuracy with a fraction of the computational expense of existing methods.
Regarding claim 4 and analogous claims 11 and 18, Lujic in view of Finn teaches the computer-implemented method of claim 1.
Lujic and Finn are combinable for the same rationale as set forth above with respect to claim 1 and analogous claims 8 and 15.
Lujic and Dempster are combinable for the same rationale as set forth above with respect to claim 2 and analogous claims 9 and 16.
Dempster further teaches wherein the random convolutional kernel transform algorithm performs a convolution with kernels on the dependent data to generate the feature maps, and wherein the extracted patterns correspond to extracted features determined from the feature maps (
(Dempster Page 4 2.3 Convolutional Neural Networks and Convolutional Kernels para 2-4, Page Convolutional neural networks represent a different approach to time series classification than many other methods. Rather than approaching the problem with a preconceived representation, convolutional neural networks use convolutional kernels to detect patterns in the input. In learning the weights of the kernels, a convolutional neural network learns the features in time series associated with different classes (Ismail Fawaz et al. 2019a).
A kernel is convolved with an input time series through a sliding dot product operation, to produce a feature map which is, in turn, used as the basis for classification (see Ismail Fawaz et al. 2019a). The basic parameters of a kernel are its size (length), weights and bias, dilation, and padding (see generally Goodfellow et al. 2016, ch. 9) [wherein the random convolutional kernel transform algorithm performs a convolution with kernels on the dependent data]. A kernel has the same structure as the input, but is typically much smaller. For time series, a kernel is a vector of weights, with a bias term which is added to the result of the convolution operation between an input time series and the weights of the given kernel. Dilation `spreads' a kernel over the input such that with a dilation of two, for example, the weights in a kernel are convolved with every second element of an input time series (see Bai et al. 2018). Padding involves appending values (typically zero) to the start and end of input time series, typically such that the `middle' weight of a given kernel aligns with the first element of an input time series at the start of the convolution operation.
Convolutional kernels can capture many of the types of features used in other methods. Kernels can capture basic patterns or shapes in time series, similar to shapelets: the convolution operation will produce large output values where the kernel matches the input. Further, dilation allows kernels to capture the same pattern at different scales (Yu and Koltun 2016). Multiple kernels in combination can capture complex patterns
The feature maps produced in applying a kernel to a time series reflect the extent to which the pattern represented by the kernel is present in the time series. In a sense, this is not unlike dictionary methods, which are based on the frequency of occurrence of patterns in time series. [to generate the feature maps, and wherein the extracted patterns correspond to extracted features determined from the feature maps]).
Claim(s) 3, 10, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Lujic in view of Finn and Dempster and further in view of Pantiskas et al., "Taking ROCKET on an Efficiency Mission: Multivariate Time Series Classification with LightWaveS," 2022 18th International Conference on Distributed Computing in Sensor Systems (DCOSS), Marina del Rey, Los Angeles, CA, USA, 2022, pp. 149-152 (“Pantiskas”).
Regarding claim 3 and analogous claims 10 and 17, Lujic in view of Finn and Dempster teach the computer-implemented method of claim 2.
Lujic and Finn are combinable for the same rationale as set forth above with respect to claim 1 and analogous claims 8 and 15.
Lujic and Dempster are combinable for the same rationale as set forth above with respect to claim 2 and analogous claim 9 and 16.
Lujic does not explicitly teach wherein the random convolutional kernel transform algorithm is a multi-variate time-series classification model.
However Pantiskas teaches wherein the random convolutional kernel transform algorithm is a multi-variate time-series classification model (Pantiskas Page 3 D. Wavelets Para. 2,
LightWaveS aims to combine the strong points of these works under a single generalized framework, with a focus on efficiency. We aim to bridge the gap between ROCKET and the wavelet theory, and we progress to the next logical step of wavelet scattering. We keep this approach lightweight, both in depth and paths of the scattering, so we can apply it to time series channels on a massive scale in a very short time. The arbitrary base set of wavelets can potentially be extended based on expert opinion, backed by the solid theory behind wavelets and their applications, making LightWaveS a suitable platform for experimentation on solutions for MTSC problems [a multi-variate time-series classification model]. Finally, the hierarchical feature filtering leads to the most relevant output features of the scattering coefficients being selected).
Lujic and Pantiskas are considered to be analogous to the claim invention because they are in the same field of series analysis. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Lujic to incorporate the teachings of Pantiskas and by implementing a random convolutional kernel transform algorithm and process multivariate timeseries. Doing so to take advantage of better accuracy and process multivariate timeseries. (Pantiskas Abstract, Nowadays, with the rising number of sensors in sectors such as healthcare and industry, the problem of multivariate time series classification (MTSC) is getting increasingly relevant and is a prime target for machine and deep learning approaches. Their expanding adoption in real-world environments is causing a shift in focus from the pursuit of ever-higher prediction accuracy with complex models towards practical, deployable solutions that balance accuracy and parameters such as prediction speed. An MTSC model that has attracted attention recently is ROCKET, based on random convolutional kernels, both because of its very fast training process and its state-of-the-art accuracy. However, the large number of features it utilizes may be detrimental to inference time. Examining its theoretical background and limitations enables us to address potential drawbacks and present LightWaveS: a framework for accurate MTSC, which is fast both during training and inference. Specifically, utilizing wavelet scattering transformation and distributed feature selection, we manage to create a solution that employs just 2.5% of the ROCKET features, while achieving accuracy comparable to recent MTSC models.).
Pertinent Prior Art
The prior art made of record and not relied upon is considered pertinent to applicant's
disclosure.
M. P. likowski, M. Śmieja, Ł. Struski and J. Tabor, "MisConv: Convolutional Neural Networks for Missing Data," 2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), Waikoloa, HI, USA, 2022, pp. 2917-2926, doi: 10.1109/WACV51458.2022.00297 teaches identifying missing data as using a convolution neural network and predicting expected value of the convolution filter and recreating a complete image.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALFREDO CAMPOS whose telephone number is (571)272-4504. The examiner can normally be reached 7:00 - 4:00 pm M - F.
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, Michael J. Huntley can be reached at (303) 297-4307. 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.
/ALFREDO CAMPOS/Examiner, Art Unit 2129
/MICHAEL J HUNTLEY/Supervisory Patent Examiner, Art Unit 2129