Detailed Action
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 . This action is in response to an application filed on
January 6th, 2023. Claims 1-24 are pending in this application.
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.
Regarding claim 3, the phrase "such as" renders the claim indefinite because it is unclear whether the limitations following the phrase are part of the claimed invention. Therefore, the claim is rejected under 35 U.S.C. 112(b). See MPEP § 2173.05(d).
Regarding claim 7, the claim recites “the first extractor and the second extractor”. The first and second extractor are not mentioned in a previous claim, therefore, the claim is rejected under 35 U.S.C. 112(b) for a lack of antecedent basis. See MPEP § 2173.05(d).
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 1-24 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1 Under Step 1 of the Subject Matter Eligibility Test of Products and Processes, the claim is directed towards a device, which is interpreted as a machine, which is one of the four statutory categories.
Next, under a Step 2A Prong 1 Analysis, the claim mentions “calculate a vector in a latent space called latent vector (hk), from the set of data“, “determine a task for each data, by: evaluating, for each component, a likelihood score”, and “assigning… the task associated with the component with the highest likelihood score among the plurality of evaluated scores.” As drafted, the “calculate a vector in a latent space called latent vector (hk), from the set of data“ is a process that, under the broadest reasonable interpretation, falls under the “mathematical grouping” grouping of abstract ideas, and “determine a task for each data, by: evaluating, for each component, a likelihood score”, and “assigning… the task associated with the component with the highest likelihood score among the plurality of evaluated scores” are processes that, under the broadest reasonable interpretation, fall under the “mental processes” grouping of abstract ideas.
Therefore, we have to examine the claim under Step 2A prong 2, which considers the additional elements within the claim. The claim’s additional elements are:
An electronic data processing device configured to process a set of data the set of data
one or more signals captured by a sensor
an acquisition module configured to acquire the set(s) of data to be processed
a plurality of components, each associated with a respective task each component being configured to implement a reversible neural network
and if the evaluated likelihood score is inconsistent for the component associated with the assigned task, modifying the assigned task to an unknown task.
“An electronic data processing device configured to process a set of data the set of data”, and “a plurality of components, each associated with a respective task each component being configured to implement a reversible neural network” are limitations that merely indicate the field of use and technological environment, and “generally link” a device and components to the judicial exception. (See MPEP 2106.05(h)) The “one or more signals captured by a sensor” , and “an acquisition module configured to acquire the set(s) of data to be processed” is considered to be insignificant extra-solution activity. (See MPEP 2106.05(g)) Lastly, “and if the evaluated likelihood score is inconsistent for the component associated with the assigned task, modifying the assigned task to an unknown task” is interpreted to be mere instruction to apply the exception, as it instructs how to use the likelihood score within the judicial exception. (See MPEP 2106.05(f)) Therefore, these additional elements do not integrate the abstract idea into a practical application. The claim is directed to an abstract idea.
Under a Step 2B analysis, the claim’s addition elements do not amount to significantly
more than the judicial exception as explained above in Step 2A prong 2. Additionally, “one or more signals captured by a sensor” , and “an acquisition module configured to acquire the set(s) of data to be processed” is considered to be well-understood, routine, and conventional, as it is receiving or transmitting data. (See MPEP 2106.05(d)) Therefore, the claim is ineligible.
Regarding claim 11, Under Step 1 of the Subject Matter Eligibility Test of Products and Processes, the claim is directed towards a method, which is one of the four statutory categories.
Next, under a Step 2A Prong 1 Analysis, the claim mentions “calculating a vector in a latent space, called latent vector hk for each component”, “determining a task for each data, by: evaluating, for each component, a likelihood score from the corresponding latent vector hk”, and “assigning… the task associated with the component with the highest likelihood score among the plurality of evaluated scores;” As drafted, the “calculating a vector in a latent space called latent vector (hk), for each component and from the set of data“ is a process that, under the broadest reasonable interpretation, falls under the “mathematical grouping” grouping of abstract ideas, and “determine a task for each data, by: evaluating, for each component, a likelihood score”, and “assigning… the task associated with the component with the highest likelihood score among the plurality of evaluated scores” are processes that, under the broadest reasonable interpretation, fall under the “mental processes” grouping of abstract ideas.
Therefore, we have to examine the claim under Step 2A prong 2, which considers the additional elements within the claim. The claim’s additional elements are:
A set of data corresponding to one or more signals captured by a sensor
The method being implemented by an electronic processing device
acquiring the set of data to be processed;
each component being associated with a respective task
and if the evaluated likelihood score is inconsistent for the component associated with the assigned task, modifying the assigned task to an unknown task.
The “each component being associated with a respective task” and “the method being implemented by an electronic processing device” are limitations that merely indicate the field of use and technological environment, and “generally link” a device and components to the judicial exception. (See MPEP 2106.05(h)) The “set of data corresponding to one or more signals captured by a sensor” , and “acquiring the set of data to be processed” is considered to be insignificant extra-solution activity. (See MPEP 2106.05(g)) Lastly, “if the evaluated likelihood score is inconsistent for the component associated with the assigned task, modifying the assigned task to an unknown task.” is interpreted to be mere instruction to apply the exception, as it instructs how to use the likelihood score within the judicial exception. (See MPEP 2106.05(f)) Therefore, these additional elements do not integrate the abstract idea into a practical application. The claim is directed to an abstract idea.
Under a Step 2B analysis, the claim’s addition elements do not amount to significantly
more than the judicial exception as explained above in Step 2A prong 2. Additionally, “set of data corresponding to one or more signals captured by a sensor” , and “acquiring the set of data to be processed” is considered to be well-understood, routine, and conventional, as it is receiving or transmitting data. (See MPEP 2106.05(d)) Therefore, the claim is ineligible.
Regarding claim 2, the claim recites feedback module configured to store each unknown task data in a buffer memory and to trigger the creation of a new task if the number of data stored in the buffer memory is greater than a predefined number; the calculation module then being configured to include a new component associated with the new task; the learning of the new component being performed from said data stored in the buffer memory.
The “feedback module configured to store each unknown task data in a buffer memory”, and “calculation module then being configured to include a new component associated with the new task” are limitations that merely indicate the field of use and technological environment, and “generally link” the module and their functionality to the judicial exception, (See MPEP 2106.05(h)) and “to trigger the creation of a new task if the number of data stored in the buffer memory is greater than a predefined number”, and “the learning of the new component being performed from said data stored in the buffer memory” is interpreted to be mere instruction to apply the exception, as it instructs how to trigger the new task creation and implement learning. (See MPEP 2106.05(f)) Therefore, the claim is rejected on the same basis as claim 1.
Regarding claim 3, the claim recites the reversible neural network of each component includes parameters, such as weights; and said parameters being optimized via a maximum likelihood method. The “reversible neural network of each component includes parameters, such as weights” is a limitation that merely indicate the field of use and technological environment, and “generally link” parameters, such as, weight, to the judicial exception, (See MPEP 2106.05(h)) and “said parameters being optimized via a maximum likelihood method” is considered to be, under the broadest reasonable interpretation, a “mathematical concept”, which is a grouping of abstract idea. Therefore, the claim is rejected on the same basis as claims 1.
Regarding claim 4, the claim recites a feature extraction module connected between the acquisition module and the calculation module, the extraction module being configured to implement at least one neural network to convert the set(s) of data into a simplified representation, by extracting one or more features common to the plurality of tasks. The “feature extraction module connected between the acquisition module and the calculation module”, is a limitation that merely indicate the field of use and technological environment, and “generally link” feature extraction, to the judicial exception, (See MPEP 2106.05(h)) and “the extraction module being configured to implement at least one neural network to convert the set(s) of data into a simplified representation, by extracting one or more features common to the plurality of tasks” is interpreted to be mere instruction to apply the exception, as it instructs on how to use the feature extraction module. (See MPEP 2106.05(f)) Therefore, the claim is rejected on the same basis as claim 1.
Regarding claim 5, the claim recites the determination module is further configured to generate a vector of random or pseudo-random numbers corresponding to the distribution of the latent space of one of the components, and then to propagate said vector in an inverse manner via the corresponding reversible neural network, in order to create an artificial example of data, a task identifier associated with this artificial example being an identifier of said component. To “generate a vector of random or pseudo-random numbers corresponding to the distribution of the latent space of one of the components” is considered to be, under the broadest reasonable interpretation, a “mathematical concept”, which is a grouping of abstract idea, and “then to propagate said vector in an inverse manner via the corresponding reversible neural network, in order to create an artificial example of data,” is interpreted to be mere instruction to apply the exception, as it instructs on how to create an artificial example of data and when to do so. (See MPEP 2106.05(f)) Lastly, “a task identifier associated with this artificial example being an identifier of said component” is considered to be, under the broadest reasonable interpretation, a “mental process”, which is a grouping of abstract idea. Therefore, the claim is rejected on the same basis as claim 1.
Regarding claim 6, the claim recites a retraining module configured to receive the vector generated by the determination module and to provide at least one artificial example of data and its identifier to the component(s) of the calculation module associated with the same identifier said component(s) to be re-trained, the re-training module including a copy of each component to be re-trained. The “retraining module configured to receive the vector generated by the determination module” and “re-training module including a copy of each component to be re-trained” is a limitation that merely indicate the field of use and technological environment, and “generally link” retraining with copies of components, to the judicial exception, (See MPEP 2106.05(h)) and “to provide at least one artificial example of data and its identifier to the component(s) of the calculation module associated with the same identifier said component(s) to be re-trained” is interpreted to be mere instruction to apply the exception, as it instructs on how to use the retraining module and what to use it on. (See MPEP 2106.05(f)) Therefore, the claim is rejected on the same basis as claim 5.
Regarding claim 7, the claim recites a feature extraction module connected between the acquisition module and the calculation module, the extraction module being configured to implement at least one neural network to convert the set(s) of data into a simplified representation, by extracting one or more features common to the plurality of tasks; wherein when the extraction module includes the first extractor and the second extractor, the retraining module further includes a copy of the second extractor, the retraining module then being further configured to provide at least one artificial example of data to the second extractor of the extraction module
The ”feature extraction module connected between the acquisition module and the calculation module” is a limitation that merely indicate the field of use and technological environment, and “generally link” feature extraction, to the judicial exception, (See MPEP 2106.05(h)), the “extraction module being configured to implement at least one neural network to convert the set(s) of data into a simplified representation, by extracting one or more features common to the plurality of tasks”, “wherein when the extraction module includes the first extractor and the second extractor, the retraining module further includes a copy of the second extractor”, and “the retraining module then being further configured to provide at least one artificial example of data to the second extractor of the extraction module” is interpreted to be mere instruction to apply the exception, as it instructs on how to use the extraction and retraining module. (See MPEP 2106.05(f)) Therefore, the claim is rejected on the same basis as claim 6.
Regarding claim 8, the claim recites wherein the device is configured to perform unsupervised task learning, each component of the calculation module being configured to calculate a vector in the latent space for each new datum, the latent space then including latent vectors for that new datum, an identifier of the component further being associated with each calculated latent vector.
The “device is configured to perform unsupervised task learning” is a limitation that merely indicate the field of use and technological environment, and “generally link” unsupervised task learning, to the judicial exception, (See MPEP 2106.05(h)) the “each component of the calculation module being configured to calculate a vector in the latent space for each new datum”, and “latent space then including latent vectors for that new datum,” is interpreted, under the broadest reasonable interpretation, as a “mathematical concept”, which is a grouping of abstract idea, and “an identifier of the component further being associated with each calculated latent vector” is interpreted, under the broadest reasonable interpretation, to be a “mental process”, which is a grouping of abstract idea. Therefore, the claim is rejected on the same basis as claim 1.
Regarding claim 9, the claim recites the determination module is further configured to modify the identifiers of components from a batch of identified examples, a respective identifier being associated with each example, by assigning for each example its identifier to the component presenting the highest likelihood score, the component or components not having an assigned identifier after taking into account all the examples of the batch being ignored. The “determination module is further configured to modify the identifiers of components from a batch of identified examples” is a limitation that merely indicate the field of use and technological environment, and “generally link” unsupervised task learning, to the judicial exception, (See MPEP 2106.05(h)) the “a respective identifier being associated with each example, by assigning for each example its identifier to the component presenting the highest likelihood score” is considered to be, under the broadest reasonable interpretation, to be a “mental process”, which is a grouping of abstract idea, and “the component or components not having an assigned identifier after taking into account all the examples of the batch being ignored.” is interpreted to be mere instruction to apply the exception, as it instructs to not assign an identifier. (See MPEP 2106.05(f)) Therefore, the claim is rejected on the same basis as claim 8.
Regarding claim 10, the claim recites an electronic system for detecting objects, the system comprising a sensor, and an electronic processing device for processing data connected to the sensor, wherein the electronic processing device is according to claim 1, and each data to be processed is an element present in a scene captured by the sensor. The limitation, as drafted, is interpreted to be mere instruction to apply the exception, as it instructs to process data related to an element present in a scene captured by the sensor. (See MPEP 2106.05(f)) Therefore, the claim is rejected on the same basis as claim 1.
Regarding claim 12, the claim recites a non-transitory computer-readable medium including a computer program including software instructions that, when executed by a computer, implement a method according to claim 11. The limitation, as drafted, is interpreted to be mere instruction to apply the exception, as it instructs to implement claim 11’s method into a program and load it into a non-transitory computer-readable medium. (See MPEP 2106.05(f)) Therefore, the claim is rejected on the same basis as claim 11.
Regarding claim 13, the claim recites the parameters are weights. The limitation, as drafted, merely indicates the field of use and technological environment, and “generally link” parameters, such as weight, to the judicial exception. (See MPEP 2106.05(h)) Therefore, the claim is rejected on the same basis as claim 3.
Regarding claim 14, the claim recites the learning of said network is performed via a backpropagation algorithm for the calculation of the gradient of each parameter. The limitation, as drafted, under the broadest reasonable interpretation, is considered to be a “mathematical process”, which is a grouping of abstract idea. Therefore, the claim is rejected on the same basis as claim 3.
Regarding claim 15, the claim recites the learning of said network is continuous. The limitation, as drafted, is a limitation that merely indicates the field of use and technological environment, and “generally link” the continuous learning, to a device that can run it. (See MPEP 2106.05(h)) Therefore, the claim is rejected on the same basis as claim 3.
Regarding claim 16 the claim recites the learning of said network is carried out after each data processing. The limitation, as drafted, is interpreted to be mere instruction to apply the exception, as it instructs on when to perform learning. (See MPEP 2106.05(f)) Therefore, the claim is rejected on the same basis as claim 15.
Regarding claim 17, the claim recites each neural network of the extraction module is invertible. The limitation, as drafted, is a limitation that merely indicates the field of use and technological environment, and “generally link” invertible neural networks to the judicial exception. (See MPEP 2106.05(h)) Therefore, the claim is rejected on the same basis as claim 4. Regarding claim 18, the claim recites the extraction module includes a first extractor configured to implement a neural network with fixed weights following the training of said network and a second extractor configured to implement a neural network with trainable weights via continuous training. The limitation, as drafted, is a limitation that merely indicate the field of use and technological environment, and “generally link” feature extraction, to the judicial exception, (See MPEP 2106.05(h)) Therefore, the claim is rejected on the same basis as claim 4.
Regarding claim 19, the claim recites the training is carried out after each processing of data. The limitation, as drafted, is interpreted to be mere instruction to apply the exception, as it instructs on when to perform training. (See MPEP 2106.05(f)) Therefore, the claim is rejected on the same basis as claim 18.
Regarding claim 20, the claim recites the training is carried out via an inverse propagation algorithm. The limitation, as drafted, is interpreted as a “mathematical concept”, which is a grouping of abstract idea. Therefore, the claim is rejected on the same basis as claim 18.
Regarding claim 21, the claim recites said vector is back propagated to the calculation module. The limitation, as drafted, is interpreted as a “mathematical concept”, which is a grouping of abstract idea. Therefore, the claim is rejected on the same basis as claim 5.
Regarding claim 22, the claim recites said vector is back propagated to a retraining module distinct from the calculation module. The limitation, as drafted, is interpreted as a “mathematical concept”, which is a grouping of abstract idea. Therefore, the claim is rejected on the same basis as claim 5.
Regarding claim 23, the claim recites the sensor is chosen from among the group consisting in: an image sensor, a sound sensor and an object detection sensor. The limitation, as drafted, is a limitation that merely indicate the field of use and technological environment, and “generally link” a particular sensor, to the judicial exception. (See MPEP 2106.05(h)) Therefore, the claim is rejected on the same basis as claim 10.
Regarding claim 24, the claim recites each data to be processed is an object detected in an image. The limitation, as drafted is a limitation that merely indicate the field of use and technological environment, and “generally link” object detection in an image, to the judicial exception. (See MPEP 2106.05(h)) Therefore, the claim is rejected on the same basis as claim 10.
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 (i.e., changing from AIA to pre-AIA ) 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.
Claims 1, 4, 5, 10, 11, 12, 17, 21 and 24 are rejected under 35 U.S.C. 103 as being unpatentable over Da Ren et al. (Herein referred to as Ren) (A multi-encoder neural conversation model) in view of Kathy Lee et al. (Herein referred to as Lee) (Twitter Trending Topic Classification) and in further view of Zavesky et al. (Herein referred to as Zavesky) (U.S. Patent Application No. US 20200169717 A1)
Regarding claim 1, Ren teaches an electronic data processing device configured to process a set of data, the device comprising: an acquisition module configured to acquire the set(s) of data to be processed (“The Topic Generation Module generates a topic distribution based on input sequences”, pg. 3, in the bulleted list in “3. Model”) (The module acquires data and generates a topic distribution based on the data.) a calculation module including a plurality of components, each associated with a respective task each component being configured to implement a reversible neural network to calculate a vector in a latent space (“The Encoder Selection Module selects one or more than one encoders (depending on selection methods) to process input sequences... The encoder encodes an input sequence into a fixed length vector”, pg. 3, in the bulleted list in “3. Model”; pg. 5, left column, second paragraph of “3.3. Response generation module”) (The set of encoders are components that are each associated with a respective task and calculates a vector based on an input sequence.) a determination module configured to determine a task for each data, (“After getting a topic distribution, MENC divides conversation data into different clusters in encoder selection module. Encoder selection module selects one or more than one encoders to process input. In this article, we propose three methods to choose encoders according to the distribution.”, pg. 3, right column, under “3.2. Encoder selection module”) (One of three encoder are selected by the encoder selection module.) by: evaluating, for each component, a likelihood score from the corresponding latent vector and assigning, to said data, the task associated with the component with the highest likelihood score among the plurality of evaluated scores (“An input sequence is processed by the encoder whose corresponding topic’s probability is the largest.”, pg. 4, right column, last paragraph before Algorithm 3)
However, Ren does not teach the set of data corresponding to one or more signals captured by a sensor that if the evaluated likelihood score is inconsistent for the component associated with the assigned task, modifying the assigned task to an unknown task.
Lee teaches a multinomial classifier that selects a topic based on a likelihood, and that if the evaluated likelihood score is inconsistent for the component associated with the assigned topic, modifying the assigned topic to an unknown topic. (“We identified 18 classes for topic classification. The classes are art & design, books, charity & deals, fashion, food & drink, health, humor, music, politics, religion, holidays & dates, science, sports, technology, business, tv & movies, other news, and other… the news related to political events are classified as politics. If the topic is about news that is not in any of the categories, it is classified as other news. If the trend definition or tweet text is gibberish or if it is in a language other than English, then we classify the topic as other category”, pgs. 3 and 4, under “B. Labeling”) (The “other” topic acts as out unknown topic. Since the topics are determined by a multinomial distribution, it would also be easy to add topics.) Since the encoders of Ren are specified according to their topics, and they are multi-task encoders, with data specific to their topics, (“To make use of the data in different clusters, we design a Seq2Seq models with multi-encoder. Each encoder is trained with the data in a specific cluster. In this way, MENC can learn and generate topic related patterns rather than dull and meaningless ones.”, pg. 3, left column, under “3. Model”) and since Ren has a functionality for determining and assigning a task based on likelihood, with the multinomial and “other” topic of Lee, the combination teaches if the evaluated likelihood score is inconsistent for the component associated with the assigned task, modifying the assigned task to an unknown task.
Therefore, it would have been considered obvious to one of ordinary skill in the art,
prior to the current application’s filing date, to combine the device and multi-encoder of Ren, with the other category as disclosed by Lee. One would be motivated to combine the two teachings, prior to the filing date of the current application, as the classifying of topics into general classes allows for easier retrieval of information, as described in Lee. (“we defined 18 general classes: arts & design, books, business, charity & deals, fashion, food & drink, health, holidays & dates, humor, music, politics, religion, science, sports, technology, tv & movies, other news, and other. Our goal is to aid users searching for information… by classifying topics into general classes (e.g., sports, politics, books) for easier retrieval of information., pg. 2, left column, first paragraph)
However, the combination still does not explicitly teach the set of data corresponding to one or more signals captured by a sensor
Zavesky teaches the set of data corresponding to one or more signals captured by a sensor (“The types of features from which 2D object detection/recognition models may be derived may include visual features from 2D video. For instance, the visual features may include low-level invariant image data, such as colors (e.g., RGB (red-green-blue) or CYM (cyan-yellow-magenta) raw data (luminance values) from a CCD/photo-sensor array), shapes, color moments, color histograms, edge distribution histograms, etc.”, Paragraph 22)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the device and multi-encoder of Ren, as modified by Lee, with the signal capturing of Zavesky. One could combine the two teachings, according to known methods of gathering data, and the combination would yield predictable results.
Regarding claim 11 Ren teaches a method for processing a set of data, the method being implemented by an electronic processing device and comprising: acquiring the set of data to be processed; (“The Topic Generation Module generates a topic distribution based on input sequences”, pg. 3, in the bulleted list in “3. Model”) (The module acquires data and generates a topic distribution based on the data. The data being acquired could be acquired by any generic sensor) calculating, via the implementation of a reversible neural network for each component of a plurality of components, a vector in a latent space, for each component and from the set of data, each component being associated with a respective task (“The Encoder Selection Module selects one or more than one encoders (depending on selection methods) to process input sequences... The encoder encodes an input sequence into a fixed length vector”, pg. 3, in the bulleted list in “3. Model”; pg. 5, left column, second paragraph of “3.3. Response generation module”) (The set of encoders are components that are each associated with a respective task and calculates a vector based on an input sequence.) determining a task for each data, (“After getting a topic distribution, MENC divides conversation data into different clusters in encoder selection module. Encoder selection module selects one or more than one encoders to process input. In this article, we propose three methods to choose encoders according to the distribution.”, pg. 3, right column, under “3.2. Encoder selection module”) (One of three encoder are selected by the encoder selection module.) by: evaluating, for each component, a likelihood score from the corresponding latent vector and assigning, to said data, the task associated with the component with the highest likelihood score among the plurality of evaluated scores; (“An input sequence is processed by the encoder whose corresponding topic’s probability is the largest.”, pg. 4, right column, last paragraph before Algorithm 3)
However, Ren does not explicitly teach, the set of data corresponding to one or more signals captured by a sensor and that if the evaluated likelihood score is inconsistent for the component associated with the assigned task, modifying the assigned task to an unknown task.
Lee teaches a multinomial classifier that selects a topic based on a likelihood, and the if evaluated likelihood score is inconsistent for the component associated with the assigned topic, modifying the assigned topic to an unknown topic. (“We identified 18 classes for topic classification. The classes are art & design, books, charity & deals, fashion, food & drink, health, humor, music, politics, religion, holidays & dates, science, sports, technology, business, tv & movies, other news, and other… the news related to political events are classified as politics. If the topic is about news that is not in any of the categories, it is classified as other news. If the trend definition or tweet text is gibberish or if it is in a language other than English, then we classify the topic as other category”, pgs. 3 and 4, under “B. Labeling”) (The “other” topic acts as out unknown topic. Since the topics are determined by a multinomial distribution, it would also be easy to add topics.) Since the encoders of Ren are specified according to their topics, and they are multi-task encoders, with data specific to their topics, (“To make use of the data in different clusters, we design a Seq2Seq models with multi-encoder. Each encoder is trained with the data in a specific cluster. In this way, MENC can learn and generate topic related patterns rather than dull and meaningless ones.”, pg. 3, left column, under “3. Model”) and since Ren has a functionality for determining and assigning a task based on likelihood, with the multinomial and “other” topic, the combination teaches if the evaluated likelihood score is inconsistent for the component associated with the assigned task, modifying the assigned task to an unknown task.
Therefore, it would have been considered obvious to one of ordinary skill in the art,
prior to the current application’s filing date, to combine the device and multi-encoder of Ren, with the other category as disclosed by Lee. One would be motivated to combine the two teachings, prior to the filing date of the current application, as the classifying of topics into general classes allows for easier retrieval of information, as described in Lee. (“we defined 18 general classes: arts & design, books, business, charity & deals, fashion, food & drink, health, holidays & dates, humor, music, politics, religion, science, sports, technology, tv & movies, other news, and other. Our goal is to aid users searching for information… by classifying topics into general classes (e.g., sports, politics, books) for easier retrieval of information., pg. 2, left column, first paragraph)
However, the combination still does not explicitly teach the set of data corresponding to one or more signals captured by a sensor.
Zavesky teaches the set of data corresponding to one or more signals captured by a sensor (“The types of features from which 2D object detection/recognition models may be derived may include visual features from 2D video. For instance, the visual features may include low-level invariant image data, such as colors (e.g., RGB (red-green-blue) or CYM (cyan-yellow-magenta) raw data (luminance values) from a CCD/photo-sensor array), shapes, color moments, color histograms, edge distribution histograms, etc.”, Paragraph 22)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the device and multi-encoder of Ren, as modified by Lee, with the signal capturing of Zavesky. One could combine the two teachings, according to known methods of gathering data, and the combination would yield predictable results.
Regarding claim 4, Ren, as modified by Lee and Zavesky, teaches the device according to claim 1, wherein the device further comprises a feature extraction module connected between the acquisition module and the calculation module, the extraction module being configured to implement at least one neural network to convert the set(s) of data into a simplified representation, by extracting one or more features common to the plurality of tasks (“a predefined set of generic classes such as news, events, opinions, deals, and private messages based on author information and domain specific features extracted from tweets such as presence of shortening of words and slangs, time-event phrases, opinionated words, emphasis on words, currency and percentage signs”, pg. 2, right column, under “Related Works” (Lee)) (A classification neural network can be easily configured to extract text and classify the data based on features within, as there exists something in Lee with a similar functionality. (“there are 20 classes in the 20 newsgroups dataset. After removing the 100 most frequent words and the words appear less than 10 times, we select top 100 topic words in each clusters. The embedding size of topic words is set to be 100 and the embedding of topic words are initialized from a uniform distribution”, pg. 7, left column, the third bullet in the list (Ren)))
Regarding claim 5, Ren, as modified by Lee and Zevesky, teaches the device according to claim 1, wherein the determination module is further configured to generate a vector of random or pseudo-random numbers corresponding to the distribution of the latent space of one of the components (“k random numbers R = {r0, . . .,rk−1} are generated. Each random number ri is in [0, 1). ri is related to a certain topic probability pi and an encoder Ei.”, pg. 4, left column, second paragraph (Ren)) and then propagate said vector in an inverse manner via the corresponding reversible neural network, in order to create an artificial example of data, (“there is an encoder and a decoder. The encoder encodes an input sequence into a fixed length vector, while the decoder decodes an output sequence according to the fixed length vector.”, pg. 5, left column, under “3.3. Response generation module” (Ren)) (The decoder decodes what the encoder encoded acting as a way of propagating a vector in an inverse manner.) and a task identifier associated with this artificial example being an identifier of said component (“Each encoder is trained with the data in a specific cluster. In this way, MENC can learn and generate topic related patterns rather than dull and meaningless ones.”, pg. 3, left column, under “3. Model” (Ren)) (The quotation implies the topic specific nature of the encoders, which uses the topic specific data to determine which encoder would be best trained to deal with new examples.)
Regarding claim 10, Ren, as modified by Lee, teaches an electronic system for detecting objects, the system comprising a sensor and an electronic processing device for processing data connected to the sensor (“…we propose a Multi-Encoder Neural Conversation (MENC) model. MENC generates topic distributions based on input sequences. Then, it divides conversation data into different clusters according to their topic distributions.”, pg. 3, left column, under “3. Model”, see also Fig. 1 (Ren)) (The dataset comprising messages and responses would require a way to capture data corresponding to signals, such as the sensors of Zavesky.) wherein the electronic processing device is according to claim 1, but does not explicitly teach each data to be processed is an element present in a scene captured by the sensor
Zavesky teaches each data to be processed is an element present in a scene captured by the sensor. (“As referred to herein, a machine learning model (MLM) (or machine learning-based model) may comprise a machine learning algorithm (MLA) that has been “trained” or configured in accordance with input data (e.g., training data) to perform a particular service, e.g., to detect a type of 2D object in image and/or video content.”, Paragraph 21)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the teachings of Ren and Lee, with the image processing of Zavesky. One would be motivated to combine the teachings, prior to the filing date of the current application, as this allows for object detection within an image in topic detection algorithms, (instead of just topic detection in text) as disclosed by Zavesky. (“…the processing system identifies a “first” object in the source video. In one example, the first object may be detected in the source video in accordance with one or more 2D object detection/recognition models… the volumetric videos may be tagged with topics contained therein (manually or via application of topic detection algorithms).”, Paragraph 40 and 49)
Regarding claim 12, Ren, as modified by Lee and Zevesky, teaches a non-transitory computer-readable medium including a computer program including software instructions that, when executed by a computer, implement a method according to claim 11. (You would need a non-transitory computer-readable medium to distribute the method of claim 11)
Regarding claim 17, Ren, as modified by Lee and Zevesky, teaches the device of claim 4, wherein each neural network of the extraction module is invertible. (“…last hidden states of encoders are initial hidden states of decoders. Therefore, decoders generate responses depending on the last hidden states of encoders.”, pg. 5, left column, bottom paragraph of “3.3. Response generation module” (Ren)) (The decoder decodes what the encoder encoded acting as a way of inverting the encoder.)
Regarding claim 21, Ren, as modified by Lee and Zevesky, teaches the device according to claim 5, wherein said vector is back propagated to the calculation module. (“there is an encoder and a decoder. The encoder encodes an input sequence into a fixed length vector, while the decoder decodes an output sequence according to the fixed length vector.”, pg. 5, left column, under “3.3. Response generation module” (Ren)) (The decoder decodes what the encoder encoded acting as a way of a vector being backpropagated, and the process could be configured to go through the calculation module.)
Regarding claim 24, Ren, as modified by Lee and Zavesky, teaches the system according to claim 10, wherein each data to be processed is an object detected in an image. (“In one example, the processing system may update one or more 2D object detection/recognition models in accordance with user feedback or based upon user actions for improved indexing. For example, if a user has manually corrected a 3D object model to “dog” instead of “cat,” this fact may be used as a negative example/feedback to retrain the 2D detection/recognition model for “cat.””, Paragraph 15 (Zavesky))
Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over Da Ren et al. (Herein referred to as Ren) (A multi-encoder neural conversation model) in view of Kathy Lee et al. (Herein referred to as Lee) (Twitter Trending Topic Classification) in view of Zavesky et al. (Herein referred to as Zavesky) (U.S. Patent Application No. US 20200169717 A1) and in further view of Ling Chen et al. (Herein referred to as Chen) (A Knowledge-Based Semisupervised Hierarchical Online Topic Detection Framework)
Regarding claim 2, Ren, as modified by Lee and Zavesky, teaches the device of claim 1, as well as, a feedback module configured to store each unknown task data in a buffer memory (“Data Collection stage - trending topic, topic definition and tweets are downloaded to compose a document… If the trend definition or tweet text is gibberish or if it is in a language other than English, then we classify the topic as other category.”, pg. 3, Figure 2; pg. 4, left column, first paragraph (Lee)) (The data is download and stored, and if the topic is unknown, it is allocated to the “other” category.) and the learning of the new component being performed from said data stored in the buffer memory. (“MENC can learn the high frequency patterns among different topics so that it can generate more topic related and meaningful responses”, pg. 2, left column, first paragraph (Ren))
However, the combination does not explicitly teach to trigger the creation of a new task if the number of data stored in the buffer memory is greater than a predefined number; nor the calculation module then being configured to include a new component associated with the new task.
Chen teaches to trigger the creation of a new task if the number of data stored in the buffer memory is greater than a predefined number; (“A large number of novel documents suggests that there exist emerging topics. New topic nodes are added as subtopic nodes of the current node in this case, e.g., w2 is an emerging topic of the root node w3”, pg. 6, left column, second paragraph) (There would have to be a predefined number to know whether or not there were a “large number of novel documents”. A new topic node is added whenever the number of topic documents are greater than that number.) and the calculation module then being configured to include a new component associated with the new task. (“A large number of novel documents suggests that there exist emerging topics. New topic nodes are added as subtopic nodes of the current node in this case,", pg. 6, left column, second paragraph) (The new topic node acts as our new task, with the subtopic nodes acting as out new component for the calculation module to be configured to.)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the filing date of the current application, to combine the teachings of Ren, as modified by Lee and Zavesky, with the new topic generation of Chen. One would be motivated to combine the two teachings, prior to the filing date of the current application, as a threshold value helps define what is novel and requiring a new node, or not, as described by Chen. (“Only few novel documents are regarded as outliers and cannot be viewed as a sign of emerging topics, i.e., the ratio between novel documents and the documents of the current topic node should be larger than a predefined emerging topic threshold θ emerge, e.g., 20%. Note that an emerging topic is related to a certain topic bandwidth. If the bandwidth of the current topic node is large, the probability of finding novel documents is usually small, i.e., it is not likely to appear a new top-level topic”, pg. 7, left column, fourth paragraph)
Claims 3, 13, 15 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Da Ren et al. (Herein referred to as Ren) (A multi-encoder neural conversation model) in view of Kathy Lee et al. (Herein referred to as Lee) (Twitter Trending Topic Classification) in further view of Zavesky et al. (Herein referred to as Zavesky) (U.S. Patent Application No. US 20200169717 A1) and in further view of A. Trelin et al. (Herein referred to as Trelin) (Binary Stochastic Filtering: a Method for Neural Network Size Minimization and Supervised Feature Selection)
Regarding claim 3, Ren, as modified by Lee and Zavesky, teaches the device of claim 1, but does not explicitly teach components including parameters, said parameters being optimized via a maximum likelihood method.
Trelin teaches components including parameters, said parameters being optimized via a maximum likelihood method. (“Cross-entropy, a function derived 4 from the maximum likelihood estimation method is widely used as a loss… The model was optimized using Adam algorithm [22] with parameters”, pgs. 4-5)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the teachings of Ren, Lee and Zavesky with the maximum likelihood estimation and parameters of Trelin. One would be motivated to combine the two teachings, prior to the filing date of the current application, as the method of Trelin comprising maximum likelihood estimation for parameters, has high accuracy as described in Trelin. (“the method was able to select minimal number of features, surpassing literature references by the accuracy/dimensionality ratio.”, Abstract; See also Table 2 on pg. 10)
Regarding claim 13, Ren, as modified by Lee, Zavesky and Trelin, teaches the device according to claim 3, wherein the parameters are weights. (“Thus, by applying the ST estimator, weights of BSF layers could be optimized during model training…The model was optimized using Adam algorithm [22] with parameters…During training process the weights of neurons and BSF are optimized”, pg. 3, bottom paragraph; pg. 5, paragraphs 2 and 3; pg. 8, step