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 .
Claim Objections
Claim 19 objected to because of the following informalities:
“determine, by the first neural network model, the first set of data patterns and the second set of data patterns based on the second reconstructed dataset”
“the second reconstructed dataset” is ambiguous as the reconstructed datasets were never numbered beforehand and it will be interpreted as “the reconstructed dataset of the second dataset”
“determine, by the first neural network model, a third set of data patterns based on the first reconstructed dataset”
“the first reconstructed dataset” is ambiguous as the reconstructed datasets were never numbered beforehand and it will be interpreted as “the reconstructed dataset of the first dataset”
Appropriate correction is required.
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 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Claim 1
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 1 is a system claim thus it falls into one of the four categories of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding independent claim 1, following limitations recite a judicial exception:
“determining, by the first neural network model, a data distribution representative of the second dataset”
[Mathematical Calculations] – determining a data distribution representative of a dataset requires to compute the current dataset into a lower dimension which requires mathematical computations that recites to an abstract idea
“determining, by the first neural network model, a third dataset corresponding to a subset of data in the second dataset based on applying a threshold to the data distribution, wherein the subset of data corresponds to new data patterns in the second dataset indicative of including different characteristics than data patterns in the first dataset”
[Mathematical Calculations] – determining a dataset by applying a threshold such that the distribution comprises a dataset that have different characteristics requires mathematical computations which recites to an abstract idea.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 1, the claim recites additional elements of
“a processor” and “a non-transitory computer readable media having stored thereon instructions executable by the processor to cause the system to perform operations”
The processor and the media are recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
“applying a first neural network model to a second dataset, the first neural network model being trained using a first dataset”
Applying a dataset is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
“first neural network model” and “second neural network model”
These models are recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
“obtaining a second neural network model trained using the first dataset”
Obtaining the model is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1] is considered merely computer components that are just to store and execute code-based instructions which are considered a mere instruction to apply an exception and amount to storing and receiving information in memory, which is well-understood, routine, conventional activity (See MPEP 2106.05(d), subsection II).
The additional elements [2,4] are considered an insignificant extra solution activity and at best the equivalent of a mere data gathering recited at a high level of generality and amount to receiving or transmitting data over a network, which is well-understood, routine, conventional activity (See MPEP 2106.05(d), subsection II).
The additional element [3] is considered a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
These limitations remain a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity, which cannot provide an inventive concept.
Regarding Claim 2
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 2 is a dependent claim of 1, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
As Claim 2 does not have any abstract idea by itself, thus uses all the limitations of Claim 1.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 2, the claim recites additional elements of
“the first dataset corresponds to data generated in the network”
The data generated in the network is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
“the second dataset corresponds to data generated in the network”
The data generated in the network is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional elements [1,2] are considered an insignificant extra solution activity and at best the equivalent of a mere data gathering recited at a high level of generality and amount to receiving or transmitting data over a network, which is well-understood, routine, conventional activity (See MPEP 2106.05(d), subsection II).
These limitations remain an insignificant extra-solution activity even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity, which cannot provide an inventive concept.
Regarding Claim 3
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 3 is a dependent claim of 2, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
As Claim 3 does not have any abstract idea by itself, thus uses all the limitations of Claim 2.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 3, the claim recites additional elements of
“apply the second neural network model to new data generated in the network to classify data patterns in the new data based on the first dataset and the third dataset”
Applying a dataset is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
“second neural network model”
The model is recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1] is considered an insignificant extra solution activity and at best the equivalent of a mere data gathering recited at a high level of generality and amount to receiving or transmitting data over a network, which is well-understood, routine, conventional activity (See MPEP 2106.05(d), subsection II).
The additional element [2] is considered a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
These limitations remain a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity, which cannot provide an inventive concept.
Regarding Claim 4
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 4 is a dependent claim of 1, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding dependent claim 4, following limitations recite a judicial exception:
“determining, by the first neural network model, a data distribution representative of the first dataset”
[Mathematical Calculations] – determining a data distribution representative of a dataset requires to go through series of steps of making the dataset into a lower dimension representation and map them back higher dimension requires mathematical computation which recites to an abstract idea
“determining, by the first neural network model, a first set of data patterns in the second dataset based on a similarity with the data distribution of the first dataset”
[Mathematical Calculations] – determining a set of data patterns based on similarity within a data distribution requires to compute the error scores to compare between the numbers that requires mathematical computations which recites to an abstract idea
“determining, by the first neural network model, a second set of data patterns in the second dataset based on a difference with the data distribution of the first dataset”
[Mathematical Calculations] - determining a set of data patterns based on similarity within a data distribution requires to compute the error scores to compare between the numbers that requires mathematical computations which recites to an abstract idea
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 4, the claim recites additional elements of
“the first neural network model”
The model is recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1] is a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
This limitation remains a mere instruction to apply an exception to the generic computer components even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components, which cannot provide an inventive concept.
Regarding Claim 5
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 5 is a dependent claim of 4, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
As Claim 5 does not have any abstract idea by itself, thus uses all the limitations of Claim 4.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 5, the claim recites additional elements of
“the first neural network model”
The model is recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1] is a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
This limitation remains a mere instruction to apply an exception to the generic computer components even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components, which cannot provide an inventive concept.
Regarding Claim 6
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 6 is a dependent claim of 4, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding dependent claim 6, following limitations recite a judicial exception:
“determining a set of error scores based on the second set of data patterns”
[Mathematical Calculations] – determining a set of error scores based on the set of data patterns requires to compare each data’s reconstruction error that requires mathematical computation which recites to an abstract idea.
“determining the new data patterns based on the set of error scores based on the second set of data patterns”
[Mathematical Calculations] – determining the new data patterns based on the error scores requires to compare the error scores in the second data set to a threshold that requires mathematical computation which recites to an abstract idea.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
The claim 6 does not recite any additional elements other than abstract ideas, so it does not integrate into a practical application. Thus, this claim is directed to the abstract idea.
Regarding Claim 7
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 7 is a dependent claim of 1, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
As Claim 7 does not have any abstract idea by itself, thus uses all the limitations of Claim 1.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 7, the claim recites additional elements of
“first neural network model comprises an auto encoder neural network model”
The model and the auto encoder are recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1] is a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
This limitation remains a mere instruction to apply an exception to the generic computer components even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components, which cannot provide an inventive concept.
Regarding Claim 8
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 8 is a method claim thus it falls into one of the four categories of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding independent claim 8, following limitations recite a judicial exception:
“determining, by the first neural network model, a data distribution representative of the second dataset”
[Mathematical Calculations] – determining a data distribution representative of a dataset requires to compute the current dataset into a lower dimension which requires mathematical computations that recites to an abstract idea
“determining, by the first neural network model, a third dataset corresponding to a subset of data in the second dataset based on applying an error threshold to the data distribution, the subset of data corresponding to new data patterns in the second dataset indicative of including different characteristics than data patterns in the first dataset”
[Mathematical Calculations] – determining a dataset by applying an error threshold such that the distribution comprises a dataset that have different characteristics requires mathematical computations which recites to an abstract idea.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 8, the claim recites additional elements of
“a computing device”
The computing device is recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
“applying, by a computing device, a first neural network model to a second dataset, wherein the first neural network model is trained using a first dataset”
Applying a dataset is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
“first neural network model” and “second neural network model”
These models are recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1,3] is considered a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
The additional element [2] is considered an insignificant extra solution activity and at best the equivalent of a mere data gathering recited at a high level of generality and amount to receiving or transmitting data over a network, which is well-understood, routine, conventional activity (See MPEP 2106.05(d), subsection II).
These limitations remain a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity, which cannot provide an inventive concept.
Regarding Claim 9
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 9 is a dependent claim of 8, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
As Claim 9 does not have any abstract idea by itself, thus uses all the limitations of Claim 8.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 9, the claim recites additional elements of
“second neural network model” and “previous neural network model”
These models are recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1] is considered a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
This limitation remains a mere instruction to apply an exception to the generic computer components even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components, which cannot provide an inventive concept.
Regarding Claim 10
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 10 is a dependent claim of 9, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
As Claim 10 does not have any abstract idea by itself, thus uses all the limitations of Claim 9.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 10, the claim recites additional elements of
“applying the second neural network model trained using the third dataset to new data generated in the computing network to classify data patterns based on the first dataset and the third dataset”
Applying a dataset is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
“second neural network model”
The model is recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1] is considered an insignificant extra solution activity and at best the equivalent of a mere data gathering recited at a high level of generality and amount to receiving or transmitting data over a network, which is well-understood, routine, conventional activity (See MPEP 2106.05(d), subsection II).
The additional element [2] is considered a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
These limitations remain a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity, which cannot provide an inventive concept.
Regarding Claim 11
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 11 is a dependent claim of 8, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
As Claim 11 does not have any abstract idea by itself, thus uses all the limitations of Claim 8.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 11, the claim recites additional elements of
“the first dataset corresponds to data generated in the network”
The data generated in the network is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
“the second dataset corresponds to data generated in the network”
The data generated in the network is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional elements [1,2] are considered an insignificant extra solution activity and at best the equivalent of a mere data gathering recited at a high level of generality and amount to receiving or transmitting data over a network, which is well-understood, routine, conventional activity (See MPEP 2106.05(d), subsection II).
These limitations remain an insignificant extra-solution activity even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity, which cannot provide an inventive concept.
Regarding Claim 12
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 12 is a dependent claim of 8, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding dependent claim 12, following limitations recite a judicial exception:
“determining, by the first neural network model, a data distribution representative of the first dataset”
[Mathematical Calculations] – determining a data distribution representative of a dataset requires to go through series of steps of making the dataset into a lower dimension representation and map them back higher dimension requires mathematical computation which recites to an abstract idea
“determining, by the first neural network model, a first set of data patterns in the second dataset based on a similarity with the data distribution of the first dataset”
[Mathematical Calculations] – determining a set of data patterns based on similarity within a data distribution requires to compute the error scores to compare between the numbers that requires mathematical computations which recites to an abstract idea
“determining, by the first neural network model, a second set of data patterns in the second dataset based on a difference with the data distribution of the first dataset”
[Mathematical Calculations] - determining a set of data patterns based on similarity within a data distribution requires to compute the error scores to compare between the numbers that requires mathematical computations which recites to an abstract idea
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 12, the claim recites additional elements of
“the first neural network model”
The model is recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1] is a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
This limitation remains a mere instruction to apply an exception to the generic computer components even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components, which cannot provide an inventive concept.
Regarding Claim 13
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 13 is a dependent claim of 12, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding dependent claim 13, following limitations recite a judicial exception:
“determining a set of error scores based on the second set of data patterns”
[Mathematical Calculations] – determining a set of error scores based on the set of data patterns requires to compare each data’s reconstruction error that requires mathematical computation which recites to an abstract idea.
“determining the new data patterns based on the set of error scores based on the second set of data patterns”
[Mathematical Calculations] – determining the new data patterns based on the error scores requires to compare the error scores in the second data set to a threshold that requires mathematical computation which recites to an abstract idea.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
The claim 13 does not recite any additional elements other than abstract ideas, so it does not integrate into a practical application. Thus, this claim is directed to the abstract idea.
Regarding Claim 14
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 14 is a dependent claim of 8, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
As Claim 14 does not have any abstract idea by itself, thus uses all the limitations of Claim 8.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 14, the claim recites additional elements of
“first neural network model”
The model is recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1] is a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
This limitation remains a mere instruction to apply an exception to the generic computer components even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components, which cannot provide an inventive concept.
Regarding Claim 15
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 15 is a non-transitory computer-readable medium claim thus it falls into one of the four categories of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding independent claim 15, following limitations recite a judicial exception:
“determine, by the first neural network model, a first set of data patterns in the second dataset based on a similarity with a data distribution of the first dataset”
[Mathematical Calculations] – determining a set of data patterns based on similarity within a data distribution requires to compute the error scores to compare between the numbers that requires mathematical computations which recites to an abstract idea
“determine, by the first neural network model, a second set of data patterns in the second dataset based on differences with the data distribution of the first dataset”
[Mathematical Calculations] - determining a set of data patterns based on similarity within a data distribution requires to compute the error scores to compare between the numbers that requires mathematical computations which recites to an abstract idea
“determine, by the first neural network model, an error threshold to identify new data patterns in the second dataset, wherein the new data patterns in the second dataset are indicative of including different characteristics that data patterns in the first dataset”
[Mathematical Calculations] – determining an error threshold to identify data patterns requires the use mean squared error which recites to an abstract idea.
“determine, by the first neural network model, a third dataset corresponding to the new data patterns in the second dataset based on applying the error threshold to the first set of data patterns and the second set of data patterns”
[Mathematical Calculations] – determining a dataset by applying the error threshold requires comparing the each sample’s reconstructed error to the threshold that requires mathematical computations which recites to an abstract idea.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 15, the claim recites additional elements of
“apply, a first neural network model to a second dataset, wherein the first neural network model is trained using a first dataset”
Applying a dataset is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
“first neural network model” and “second neural network model”
These models are recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
“obtaining a second neural network model trained using the first dataset”
Obtaining the model is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional elements [1,3] are considered an insignificant extra solution activity and at best the equivalent of a mere data gathering recited at a high level of generality and amount to receiving or transmitting data over a network, which is well-understood, routine, conventional activity (See MPEP 2106.05(d), subsection II).
The additional element [2] is a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
These limitations remain a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity, which cannot provide an inventive concept.
Regarding Claim 16
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 16 is a dependent claim of 15, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
As Claim 16 does not have any abstract idea by itself, thus uses all the limitations of Claim 15.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 16, the claim recites additional elements of
“the first dataset corresponds to data generated in the network”
The data generated in the network is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
“the second dataset corresponds to data generated in the network”
The data generated in the network is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional elements [1,2] are considered an insignificant extra solution activity and at best the equivalent of a mere data gathering recited at a high level of generality and amount to receiving or transmitting data over a network, which is well-understood, routine, conventional activity (See MPEP 2106.05(d), subsection II).
These limitations remain an insignificant extra-solution activity even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity, which cannot provide an inventive concept.
Regarding Claim 17
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 17 is a dependent claim of 16, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
As Claim 17 does not have any abstract idea by itself, thus uses all the limitations of Claim 6.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 17, the claim recites additional elements of
“apply the second neural network model trained using the third dataset to new data generated in the network to classify data patterns in the new data based on the first dataset and the third dataset”
Applying a dataset is merely data gathering recited at a high level of generality, thus is insignificant extra-solution activity (See MPEP 2106.05(g)).
“second neural network model”
The model is recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1] is considered an insignificant extra solution activity and at best the equivalent of a mere data gathering recited at a high level of generality and amount to receiving or transmitting data over a network, which is well-understood, routine, conventional activity (See MPEP 2106.05(d), subsection II).
The additional element [2] is considered a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
These limitations remain a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components and an insignificant extra-solution activity, which cannot provide an inventive concept.
Regarding Claim 18
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 18 is a dependent claim of 15, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding dependent claim 18, following limitations recite a judicial exception:
“translate, by the first neural network model based on a first function, the second dataset into encoded representations of the second dataset”
[Mathematical Calculations] – translating the dataset into encoded representations using the first function involves mathematical computation which recites to an abstract idea
“translate, by the first neural network model based on a second function, the encoded representations of the second dataset into a reconstructed dataset of the second dataset”
[Mathematical Calculations] – translating the encoded representations into a reconstructed dataset using the second function involves mathematical computation which recites to an abstract idea
“determine one or more data patterns in the reconstructed dataset of the second dataset based on the first dataset”
[Mathematical Calculations] – determining the data patterns in the dataset requires to compute the reconstruction errors which are compared to the baseline that involves mathematical computation which recites to an abstract idea
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 11, the claim recites additional elements of
“first neural network model”
The model is recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1] is considered a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
This limitation remains a mere instruction to apply an exception to the generic computer components even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components, which cannot provide an inventive concept.
Regarding Claim 19
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 19 is a dependent claim of 18, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding dependent claim 19, following limitations recite a judicial exception:
“translate, by the first neural network model based on the first function, the first dataset into encoded representation of the first dataset”
[Mathematical Calculations] – translating the dataset into encoded representations using the first function involves mathematical computation which recites to an abstract idea
“translate, by the first neural network model based on the second function, the encoded representations of the first dataset into a reconstructed dataset of the first dataset”
[Mathematical Calculations] – translating the encoded representations into a reconstructed dataset using the second function involves mathematical computation which recites to an abstract idea
“determine one or more data patterns in the reconstructed dataset of the first dataset based on the first dataset”
[Mathematical Calculations] – determining the data patterns in the dataset requires computing the reconstruction errors where the samples are aligned according to the size of the error that represents the patterns that involves mathematical computation which recites to an abstract idea
“determine, by the first neural network model, the first set of data patterns and the second set of data patterns based on the second reconstructed dataset”
[Mathematical Calculations] – determining the data patterns in the dataset requires compute the reconstruction errors which are compared to the baseline that involves mathematical computation which recites to an abstract idea
“determine, by the first neural network model, a third set of data patterns based on the first reconstructed dataset”
[Mathematical Calculations] – determining the data patterns based on the dataset requires computing the reconstruction errors where the samples are aligned according to the size of the error that represents the patterns and the aggregation of them are the set thus it involves mathematical computation which recites to an abstract idea
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
Regarding Claim 19, the claim recites additional elements of
“the first neural network model”
The model is recited at a high level of generality and is merely adding words “apply it” to the judicial exception. (see MPEP 2106.05(f))
[Even when viewed in combination, the additional elements do no more than automate the mental processes that a person could perform, using computer components as a tool, thus the claim as a whole does not integrate into a practical application.]
Step 2B – whether the claim as a whole amount to significantly more than the judicial exception? I.e. Are there any additional elements (features/limitations/step) recited in the claim beyond the abstract idea?
The claim does not provide an inventive concept (significantly more than the abstract idea). The claim is ineligible.
As explained above, the additional element [1] is a mere instruction to apply an exception to the generic computer components or machine-learning components that simply runs mathematical calculations. (see MPEP 2106.05(f))
This limitation remains a mere instruction to apply an exception to the generic computer components even upon reconsideration. Even when considered in combination, the additional element represents a mere instruction to apply an exception to the generic computer components, which cannot provide an inventive concept.
Regarding Claim 20
Step 1 – whether the claim falls within any statutory category. See MPEP 2016.03
Claim 20 is a dependent claim of 19, thus it falls within the same category of statutory subject matter.
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding dependent claim 20, following limitations recite a judicial exception:
“determinine a first set of error scores based on the second set of data patterns”
[Mathematical Calculations] – determining a set of error scores based on the set of data patterns requires to compare each data’s reconstruction error that requires mathematical computation which recites to an abstract idea.
“determinine a second set of error scores based on the third set of data patterns”
[Mathematical Calculations] – determining a set of error scores based on the set of data patterns requires to compare each data’s reconstruction error that requires mathematical computation which recites to an abstract idea.
“determine the error threshold based on the first set of error scores and the second set of error scores, wherein the error threshold comprises a threshold value range determined based on a mean square error value of the first set of error scores and the second set of error scores”
[Mathematical Calculations] – determining an error threshold based on the sets of error scores using the mean squared errors requires a series of mathematical computations which recites to an abstract idea.
Step 2A Prong 2 – whether the claim recites additional elements that integrate the exception into a practical application of the exception?
The claim 20 does not recite any additional elements other than abstract ideas, so it does not integrate into a practical application. Thus, this claim is directed to the abstract idea.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1, 4-9, 12-16, 18-20are rejected under 35 U.S.C. 103 as being unpatentable over Xia et al. (Xia), Non-Patent Literature, “Learning Discriminative Reconstructions for Unsupervised Outlier Removal”, published on 2015, Pages: 9, in view of Tang et al. (Tang), Non-Patent Literature, “Kaizen: Practical self-supervised continual learning with continual fine-tuning”, published on January 2024, Pages: 10.
As to independent Claim 1,
Xia teaches a processor; and a non-transitory computer readable media having stored thereon instructions executable by the processor to cause the system to perform operations (Xia, Pg1512, Left Column, Third Bullet Point, Lines4-6, "so that we can handle large scale data in a streaming fashion and leverage the off-the shelf parallel computation framework on CPU/GPU", wherein using CPU/GPU for the computation framework inherently indicates that processors and non-transitory computer-readable mediums exist) comprising:
applying a first neural network model to a second dataset, the first neural network model being trained using a first dataset (Xia, Pg1512, Section3.1, Lines1-7, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector). We apply an autoencoder to first compress x into a lowdimensional intermediate representation and then map it back to a reconstructed copy, f(x). The form of an autoencoder f(·) is a neural network, with hidden linear or non-linear neurons."
Pg1511, Abstract, Lines3-7, "We address this problem by utilizing the reconstruction errors of an autoencoder. We observe that when data are reconstructed from low dimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors."
Pg1518, Figure10, "Figure 10. Comparisons on MNIST dataset when (a) images are represented by raw pixels and (b) images are represented by out puts of the fourth hidden layer in a pre-trained autoencoder", wherein Xia discloses a set of image representations (the corresponding second dataset as it is being applied to the autoencoder or the first neural network model for the reconstruction procedure to find reconstruction error). This autoencoder is pre-trained using MNIST or the corresponding first dataset. Also, Xia mentions that the reconstruction from the low dimensional representations pre-requires the autoencoder to be trained beforehand, thus it is functionally equivalent to the claimed invention);
determining, by the first neural network model, a data distribution representative of the second dataset (Xia, Pg1512, Section3.1, Paragraph3, Lines6-8, "in order to minimize the overall reconstruction error, an autoencoder has to find the representations that can capture statistical regularities of training set"
Pg1512, Section3.1 Paragraph1, Lines7-8, "The reconstruction error of x_i is the squared loss:
PNG
media_image1.png
27
155
media_image1.png
Greyscale
", wherein the statistical regularities are the corresponding distribution by using e_i to find how much each data x_i is deviated from the distribution. Thus, this aggregation of reconstruction errors is the corresponding representative of the second dataset);
determining, by the first neural network model, a third dataset corresponding to a subset of data in the second dataset based on applying a threshold to the data distribution, wherein the subset of data corresponds to new data patterns in the second dataset indicative of including different characteristics than data patterns in the first dataset (Xia, Pg1511, Right Column, Paragraph1, Lines7-11, "Based on this, one can conveniently identify the images with large reconstruction errors as outliers. We were both surprised and excited in finding that simply thresholding the reconstruction errors can lead to competitive results with most existing methods"
Pg1514, Section4.2, Paragraph3, "Since e_i is scalar, labeling data can be translated into sorting the reconstruction errors and then finding a cut-off threshold. Therefore, the objective in Equation 4 can be trivially optimized by linearly scanning an optimal thresh old. Then we label each sample x_i by comparing its reconstruction error e_i against the obtained optimal threshold", wherein Xia discloses that the optimal threshold, which was scanned from the distribution of the reconstruction error yielded from the autoencoder, and it being used in selecting outliers or the corresponding third dataset. This set of outliers comprises different characteristics from the first dataset, which is functionally equivalent to the claimed invention of determining a third dataset from the second dataset that were outliers compared to the first dataset's distribution of reconstructed errors.)
Xia teaches about the data generated in a network of the system (Xia, Pg1511, Introduction, Lines3-4, "Retrieving images from search engines"
Xia, Pg1512, Left Column, Third Bullet Point, Lines4-6, "so that we can handle large scale data in a streaming fashion and leverage the off-the shelf parallel computation framework on CPU/GPU", wherein using the search engines and handling the streaming of largescale data indicates that the data is being received using a network, rendering it functionally equivalent to the claimed invention of receiving the data generated in the network of the system).
Xia, however, does not teach the following limitations but from the same field of endeavor, Tang teaches the following:
a system for performing continual learning of neural network models for performing one or more given tasks (Tang, Pg2831, Right Column, Lines15-18, "Kaizen puts forward a general architecture in which both the feature extraction and the fine-tuning are performed continually from a stream of both large unlabeled and small labelled data.”)
obtaining a second neural network model trained using the first dataset (Tang, Pg2832, Right Column, Subsection: Knowledge distillation, Lines4-6, "we make a frozen copy of the trained model (fO_t-1 and g_t-1) from the previous task for knowledge distillation before we start training"
Pg2832, Right Column, Subsection: Knowledge distillation, Paragraph2, Lines2-5, "The predictions from the Current Classifier g_t are made to be similar to the predictions (or soft labels) from the Previous Classifier (g_t-1) using the categorical cross-entropy loss", wherein Tang discloses a second neural network, g_t-1, that has been trained using the previous task or the corresponding first dataset, which the network will used as a baseline for incremental learning of the current classifier, g_t. In other words, g_t-1 is being fine-tuned to be the current classifier, g_t, which is functionally equivalent to the claimed invention of using a trained second neural network for the incremental learning or fine-tuning.)
Xia teaches about collecting the outliers from the second dataset and build a third dataset to be used by the second neural network. However, Xia does not teach the following limitation of training the second neural network model using the third dataset, wherein training the second neural network model finetunes the second neural network model in performing the one or more given tasks. But from the same field of endeavor, Tang teaches this limitation (Tang, Pg2832, Right Column, Paragraph1, Lines1-6, "If the label for the image is available, we train the classifier in a supervised manner. The embedding obtained from the Current Feature Extractor will be fed to the classifier network g_t to obtain class probabilities after softmax activation. Categorical cross-entropy loss LC^CT is used for training the classifier"
Pg2831, Right Column, Lines15-18, "Kaizen puts forward a general architecture in which both the feature extraction and the fine-tuning are performed continually from a stream of both large unlabeled and small labelled data."
Pg2830, Figure1, "Kaizen (b) leverages distillation across both the feature extraction and fine-tuning steps for each task", wherein Tang discloses the current classifier network g_t (the corresponding second neural network) which was trained above from g_t-1, is being trained continually or fine-tuned using the stream of labelled data or the current task (the corresponding third dataset). This third dataset can be replaced by the set of outliers found by Xia, which is then functionally equivalent to the claimed invention of fine-tunning the second network for continual learning using the third dataset.)
Xia and Tang are analogous to the claimed invention as they are from the same field of endeavor of computer vision tasks utilizing deep neural network models to process and learn from visual datasets. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date, to combine unsupervised autoencoder reconstruction-error thresholding for automatic outlier removal and data pattern separation of Xia with a practical self-supervised continual learning framework with continual fine-tuning that sequentially updates a feature extractor and a classifier from a continuous data stream. The motivation to combine is as recited by Tang (Tang, Pg2829, Right Column, Paragraph1, Lines10-12, “Data availability comes also with the cost of labelling samples for supervised learning or fine-tuning, since only a limited amount of data can be annotated at a time”) such that in practical machine learning deployments where filtering and annotating data samples manually is exceptionally time-consuming, expensive, and heavily restricted by data availability limitations, thus combination of the two can yield a system that automatically filter and isolate distinctive or novel data subsets from a noisy, unannotated stream without human intervention, thereby directly providing clean, high-quality data patterns to optimize the downstream end-to-end continual fine-tuning phase.
As to dependent Claim 4,
The combination of Xia and Tang teaches, as mentioned in Claim 1, the overall architecture of using dual-pipeline of sifting incoming data with the trained autoencoder to collect outliers as a new pattern and input them into the second network to be continually fine-tuned.
Xia further teaches the system of claim 1, wherein determining the data distribution representative of the second dataset comprises:
determining, by the first neural network model, a data distribution representative of the first dataset (Xia, Pg1512, Section3.1, Paragraph3, Lines6-8, "in order to minimize the overall reconstruction error, an autoencoder has to find the representations that can capture statistical regularities of training set",
Pg1511, Abstract, Lines3-7, "We address this problem by utilizing the reconstruction errors of an autoencoder. We observe that when data are reconstructed from low dimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors",
Pg1518, Figure10, "Figure 10. Comparisons on MNIST dataset when (a) images are represented by raw pixels and (b) images are represented by out puts of the fourth hidden layer in a pre-trained autoencoder", wherein the autoencoder (the corresponding first neural network) to find representations of the training set has been trained using the MNIST dataset (the corresponding first dataset). Wherein the statistical regularities found within this trained baseline are compared to identify inliers and outliers, which inherently means a distribution being compared to that of the MNIST dataset, thus rendering it functionally equivalent to the claimed invention)
determining, by the first neural network model, a first set of data patterns in the second dataset based on a similarity with the data distribution of the first dataset (Xia, Pg1511, Abstract, Lines5-7, "We observe that when data are reconstructed from lowdimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors”,
Pg1512, Section3.1, Lines1-2, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector)”, wherein the autoencoder identifies the set of images (the corresponding second dataset) as inliers (the corresponding similarity dataset) and outliers according to their reconstruction errors which is inherently being compared to the distribution of the first dataset mentioned above), and
determining, by the first neural network model, a second set of data patterns in the second dataset based on a difference with the data distribution of the first dataset (Xia, Pg1511, Abstract, Lines5-7, "We observe that when data are reconstructed from lowdimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors”,
Pg1512, Section3.1, Lines1-2, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector)”, wherein the autoencoder identifies the set of images (the corresponding second dataset) as inliers and outliers (the corresponding difference dataset) according to their reconstruction errors which is inherently being compared to the distribution of the first dataset mentioned above);
wherein the second set of data patterns comprises one or more data points including different characteristics from one or more data points in the first dataset (Xia, Pg1511, Right Column, Paragraph1, Lines7-11, "Based on this, one can conveniently identify the images with large reconstruction errors as outliers. We were both surprised and excited in finding that simply thresholding the reconstruction errors can lead to competitive results with most existing methods", wherein the outliers with the large reconstruction errors inherently means that the difference dataset comprises data that is significantly and characteristically different from the first dataset, which is functionally equivalent to the claimed invention.)
As to dependent Claim 5,
The combination of Xia and Tang teaches, as mentioned in Claim 4, that the autoencoder that was trained using the first dataset finds the distribution of the reconstruction errors of the second dataset by comparing them to the distribution of the first dataset to identify the inliers and the outliers or the first and second sets of data patterns.
Xia further teaches the system of claim 4,
wherein the first set of data patterns and the second set of data patterns comprises reconstructed samples generated based on applying the first neural network model trained using the first dataset to the second dataset (Xia, Pg1512, Section3.1, Lines1-5, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector). We apply an autoencoder to first compress x into a low dimensional intermediate representation and then map it back to a reconstructed copy, f(x)",
Pg1512, Section3.1 Paragraph1, Lines7-8, "The reconstruction error of x_i is the squared loss:
PNG
media_image1.png
27
155
media_image1.png
Greyscale
",
Pg1518, Figure10, "Figure 10. Comparisons on MNIST dataset when (a) images are represented by raw pixels and (b) images are represented by out puts of the fourth hidden layer in a pre-trained autoencoder", wherein the set of images, x_i in x, (the corresponding second dataset) is inputted into the autoencoder to create a reconstructed copy f(x_i), which then finds the reconstruction error of f(x_i) compared to the original data x_i to identify whether x_i belongs to the inliers or the outliers (the corresponding first and second sets of data patterns), which is functionally equivalent to the claimed invention.)
As to dependent Claim 6,
The combination of Xia and Tang teaches, as mentioned in Claim 4, that the autoencoder that was trained using the first dataset finds the distribution of the reconstruction errors of the second dataset by comparing them to the distribution of the first dataset to identify the inliers and the outliers or the first and second sets of data patterns.
Xia further teaches the system of claim 4, wherein determine the third dataset corresponding to the new data patterns in the second dataset based on applying the threshold to the data distribution further comprises:
determining a set of error scores based on the second set of data patterns (Xia, Pg1512, Section3.1 Paragraph1, Lines7-8, "The reconstruction error of x_i is the squared loss:
PNG
media_image1.png
27
155
media_image1.png
Greyscale
", wherein this reconstruction error, e_i, computed by the autoencoder of the input x_i corresponds to the error score of the claimed invention), and
determining the new data patterns based on the set of error scores based on the second set of data patterns, wherein the new data patterns comprises each data point in the second set of data patterns comprising an error score exceeding the threshold (Xia, Pg1511, Right Column, Paragraph1, Lines7-11, "Based on this, one can conveniently identify the images with large reconstruction errors as outliers. We were both surprised and excited in finding that simply thresholding the reconstruction errors can lead to competitive results with most existing methods",
Pg1514, Section4.2, Paragraph3, "Since e_i is scalar, labeling data can be translated into sorting the reconstruction errors and then finding a cut-off threshold. Therefore, the objective in Equation 4 can be trivially optimized by linearly scanning an optimal threshold. Then we label each sample x_i by comparing its reconstruction error e_i against the obtained optimal threshold", wherein applying a ‘cut-off threshold’ to partition data where the outlier class inherently possesses larger error values than the inlier class means that an error score must exceed said threshold to be designated as an outlier, thus rendering it functionally equivalent to the claimed invention.)
As to dependent Claim 7,
The combination of Xia and Tang teaches, as mentioned in Claim 1, the overall architecture of using dual-pipeline of sifting incoming data with the trained autoencoder to collect outliers as a new pattern and input them into the second network to be continually fine-tuned.
Xia further teaches the system of claim 1, wherein the first neural network model comprises an auto encoder neural network model (Xia, Pg1512, Section3.1, Lines2-7, "We apply an autoencoder to first compress x into a lowdimensional intermediate representation and then map it back to a reconstructed copy, f(x). The form of an autoencoder f(·) is a neural network, with hidden linear or non-linear neurons.")
As to independent Claim 8,
it is a method claim that contains similar limitations of Claim 1 and thus rejected under the same rationale.
As to dependent Claim 9,
The combination of Xia and Tang teaches, as mentioned in Claim 8, about the overall method of using dual-pipeline of sifting incoming data with the trained autoencoder, that was trained with the first dataset, to collect outliers as a new pattern from the second dataset and input them into the second network as the third dataset to be continually fine-tuned.
Tang further teaches the computer-implemented method of claim 8,
wherein the second neural network model comprises a previous neural network model trained using the first dataset (Tang, Pg2832, Right Column, Subsection: Knowledge distillation, Lines4-6, "we make a frozen copy of the trained model (fO_t-1 and g_t-1) from the previous task for knowledge distillation before we start training"
Pg2832, Right Colomn, Subsection: Knowledge distillation, Paragraph2, Lines2-5, "The predictions from the Current Classifier g_t are made to be similar to the predictions (or soft labels) from the Previous Classifier (g_t-1) using the categorical cross-entropy loss", wherein Tang discloses a neural network of g_t-1 that is a frozen copy (the corresponding second neural network) which was trained using the same dataset, which is functionally equivalent to the claimed invention of utilizing the two identical neural networks before the fine-tuning phase.)
As to dependent Claim 12,
The combination of Xia and Tang teaches, as mentioned in Claim 8, about the overall method of using dual-pipeline of sifting incoming data with the trained autoencoder, that was trained with the first dataset, to collect outliers as a new pattern from the second dataset and input them into the second network as the third dataset to be continually fine-tuned.
Xia further teaches the computer-implemented method of claim 8, wherein determining the data distribution representative of the second dataset comprises:
determining, by the first neural network model, a data distribution representative of the first dataset (Xia, Pg1512, Section3.1, Paragraph3, Lines6-8, "in order to minimize the overall reconstruction error, an autoencoder has to find the representations that can capture statistical regularities of training set",
Pg1511, Abstract, Lines3-7, "We address this problem by utilizing the reconstruction errors of an autoencoder. We observe that when data are reconstructed from low dimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors",
Pg1518, Figure10, "Figure 10. Comparisons on MNIST dataset when (a) images are represented by raw pixels and (b) images are represented by out puts of the fourth hidden layer in a pre-trained autoencoder", wherein the autoencoder (the corresponding first neural network) to find representations of the training set has been trained using the MNIST dataset (the corresponding first dataset). Wherein the statistical regularities found within this trained baseline are compared to identify inliers and outliers, which inherently means a distribution being compared to that of the MNIST dataset, thus rendering it functionally equivalent to the claimed invention),
determining, by the first neural network model, a first set of data patterns in the second dataset based on a similarity with the data distribution of the first dataset (Xia, Pg1511, Abstract, Lines5-7, "We observe that when data are reconstructed from lowdimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors”,
Pg1512, Section3.1, Lines1-2, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector)”, wherein the autoencoder identifies the set of images (the corresponding second dataset) as inliers (the corresponding similarity dataset) and outliers according to their reconstruction errors which is inherently being compared to the distribution of the first dataset mentioned above), and
determining, by the first neural network model, a second set of data patterns in the second dataset based on a difference with the data distribution of the first dataset (Xia, Pg1511, Abstract, Lines5-7, "We observe that when data are reconstructed from lowdimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors”,
Pg1512, Section3.1, Lines1-2, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector)”, wherein the autoencoder identifies the set of images (the corresponding second dataset) as inliers and outliers (the corresponding difference dataset) according to their reconstruction errors which is inherently being compared to the distribution of the first dataset mentioned above),
wherein the first set of data patterns and the second set of data patterns comprises reconstructed samples generated based on applying the first neural network model trained using the first dataset to the second dataset (Xia, Pg1512, Section3.1, Lines1-5, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector). We apply an autoencoder to first compress x into a low dimensional intermediate representation and then map it back to a reconstructed copy, f(x)",
Pg1512, Section3.1 Paragraph1, Lines7-8, "The reconstruction error of x_i is the squared loss:
PNG
media_image1.png
27
155
media_image1.png
Greyscale
",
Pg1518, Figure10, "Figure 10. Comparisons on MNIST dataset when (a) images are represented by raw pixels and (b) images are represented by out puts of the fourth hidden layer in a pre-trained autoencoder", wherein the set of images, x_i in x, (the corresponding second dataset) is inputted into the autoencoder to create a reconstructed copy f(x_i), which then finds the reconstruction error of f(x_i) compared to the original data x_i to identify whether x_i belongs to the inliers or the outliers (the corresponding first and second sets of data patterns), which is functionally equivalent to the claimed invention.)
As to dependent Claim 13,
it is a method claim that contains similar limitations of Claim 6 and thus rejected under the same rationale.
As to dependent Claim 14,
The combination of Xia and Tang teaches, as mentioned in Claim 8, about the overall method of using dual-pipeline of sifting incoming data with the trained autoencoder, that was trained with the first dataset, to collect outliers as a new pattern from the second dataset and input them into the second network as the third dataset to be continually fine-tuned.
Xia further teaches the computer-implemented method of claim 8, further comprising:
training the first neural network model using the third dataset, wherein training the first neural network model enables determining data patterns in new data generated in the computing network (Xia, Pg1514, Subsection: Discriminative Labeling,
PNG
media_image2.png
359
494
media_image2.png
Greyscale
Pg1514, Section4.2, Paragraph3, "Since e_i is scalar, labeling data can be translated into sorting the reconstruction errors and then finding a cut-off threshold. Therefore, the objective in Equation 4 can be trivially optimized by linearly scanning an optimal thresh old. Then we label each sample x_i by comparing its reconstruction error e_i against the obtained optimal threshold."
Pg1514, Subsection: Reconstruction Learning, Equation 5,
PNG
media_image3.png
64
350
media_image3.png
Greyscale
Pg1516, Left Column, Second Bullet Point, "The better reconstructions, the smaller errors for positives. Meanwhile, most outliers are not well reconstructed and even be pushed away by the discriminative term in our loss function. Then the reconstruction error h becomes more discriminative"
Pg1515, Equation 6,
PNG
media_image4.png
141
428
media_image4.png
Greyscale
Pg1512, Section3.1, Paragraph3, Lines6-8, "in order to minimize the overall reconstruction error, an autoencoder has to find the representations that can capture statistical regularities of training set", wherein Xia details an iterative 'reconstruction learning' procedure where the network parameters of the autoencoder (corresponding to the first neural network model) are updated via back-propagation using a specialized loss function of
PNG
media_image3.png
64
350
media_image3.png
Greyscale
. The discriminative term h = sigma_w / sigma_t mathematically depends on the within-class variance sigma_w, which explicitly incorporates the scalar error profiles of the outlier cluster:
PNG
media_image5.png
24
155
media_image5.png
Greyscale
where y_i = 0 targets the outliers. Because the computation gradients of equation6 used to update and train the autoencoder inherently utilize the structural regularities and variance profiles of the isolated outlier data points (the third dataset) to deliberately push away outliers and maximize the separability of data distributions, which is functionally equivalent to the claimed invention.)
As to independent Claim 15,
Xia teaches apply a first neural network model to a second dataset, wherein the first neural network model is trained using a first dataset (Xia, Pg1512, Section3.1, Lines1-7, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector). We apply an autoencoder to first compress x into a lowdimensional intermediate representation and then map it back to a reconstructed copy, f(x). The form of an autoencoder f(·) is a neural network, with hidden linear or non-linear neurons."
Pg1511, Abstract, Lines3-7, "We address this problem by utilizing the reconstruction errors of an autoencoder. We observe that when data are reconstructed from low dimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors."
Pg1518, Figure10, "Figure 10. Comparisons on MNIST dataset when (a) images are represented by raw pixels and (b) images are represented by out puts of the fourth hidden layer in a pre-trained autoencoder", wherein Xia discloses a set of image representations (the corresponding second dataset as it is being applied to the autoencoder or the first neural network model for the reconstruction procedure to find reconstruction error). This autoencoder is pre-trained using MNIST or the corresponding first dataset. Also, Xia mentions that the reconstruction from the low dimensional representations pre-requires the autoencoder to be trained beforehand, thus it is functionally equivalent to the claimed invention);
determine, by the first neural network model, a first set of data patterns in the second dataset based on a similarity with a data distribution of the first dataset (Xia, Pg1511, Abstract, Lines5-7, "We observe that when data are reconstructed from lowdimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors”,
Pg1512, Section3.1, Lines1-2, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector)”, wherein the autoencoder identifies the set of images (the corresponding second dataset) as inliers (the corresponding similarity dataset) and outliers according to their reconstruction errors which is inherently being compared to the distribution of the first dataset mentioned above);
determine, by the first neural network model, a second set of data patterns in the second dataset based on differences with the data distribution of the first dataset (Xia, Pg1511, Abstract, Lines5-7, "We observe that when data are reconstructed from lowdimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors”,
Pg1512, Section3.1, Lines1-2, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector)”, wherein the autoencoder identifies the set of images (the corresponding second dataset) as inliers and outliers (the corresponding difference dataset) according to their reconstruction errors which is inherently being compared to the distribution of the first dataset mentioned above);
determine, by the first neural network model, an error threshold to identify new data patterns in the second dataset, wherein the new data patterns in the second dataset are indicative of including different characteristics that data patterns in the first dataset (Xia, Pg1511, Right Column, Paragraph1, Lines7-11, "Based on this, one can conveniently identify the images with large reconstruction errors as outliers. We were both surprised and excited in finding that simply thresholding the reconstruction errors can lead to competitive results with most existing methods"
Pg1514, Section4.2, Paragraph3, "Since e_i is scalar, labeling data can be translated into sorting the reconstruction errors and then finding a cut-off threshold. Therefore, the objective in Equation 4 can be trivially optimized by linearly scanning an optimal threshold. Then we label each sample x_i by comparing its reconstruction error e_i against the obtained optimal threshold", wherein Xia discloses that the optimal threshold (the corresponding error threshold), which was scanned from the distribution of the reconstruction error yielded from the autoencoder, and it being used in selecting outliers from the second dataset that exceeds the threshold (the corresponding new data patterns). This set of outliers comprises different characteristics from the first dataset, which is functionally equivalent to the claimed invention of determining a third dataset from the second dataset that were outliers compared to the first dataset's distribution of reconstructed errors.);
determine, by the first neural network model, a third dataset corresponding to the new data patterns in the second dataset based on applying the error threshold to the first set of data patterns and the second set of data patterns (Xia, Pg1511, Right Column, Paragraph1, Lines7-11, "Based on this, one can conveniently identify the images with large reconstruction errors as outliers. We were both surprised and excited in finding that simply thresholding the reconstruction errors can lead to competitive results with most existing methods"
Pg1514, Section4.2, Paragraph3, "Since e_i is scalar, labeling data can be translated into sorting the reconstruction errors and then finding a cut-off threshold. Therefore, the objective in Equation 4 can be trivially optimized by linearly scanning an optimal threshold. Then we label each sample x_i by comparing its reconstruction error e_i against the obtained optimal threshold",
Pg1516, Left Column, Second Bullet Point, "The better reconstructions, the smaller errors for positives. Meanwhile, most outliers are not well reconstructed and even be pushed away by the discriminative term in our loss function. Then the reconstruction error h becomes more discriminative", wherein the second set of data patterns includes outliers such that among these outliers that exceeds the threshold will be pushed away meaning these outliers will be selected as the outliers of the outliers or the corresponding third dataset, which is functionally equivalent to the claimed invention);
While Xia teaches that a non-transitory computer readable media having stored therein instructions executable by a processor to perform operations (Xia, Pg1512, Left Column, Third Bullet Point, Lines4-6, "so that we can handle large scale data in a streaming fashion and leverage the off-the shelf parallel computation framework on CPU/GPU), Xia does not teach that it is done for performing continual learning of neural network models. From the same field of endeavor, Tang teaches this (Pg2831, Right Column, Lines15-18, "Kaizen puts forward a general architecture in which both the feature extraction and the fine-tuning are performed continually from a stream of both large unlabeled and small labelled data".)
Xia also does not teach the following limitations but from the same field of endeavor, Tang teaches the following:
obtaining a second neural network model trained using the first dataset (Tang, Pg2832, Right Column, Subsection: Knowledge distillation, Lines4-6, "we make a frozen copy of the trained model (fO_t-1 and g_t-1) from the previous task for knowledge distillation before we start training"
Pg2832, Right Column, Subsection: Knowledge distillation, Paragraph2, Lines2-5, "The predictions from the Current Classifier g_t are made to be similar to the predictions (or soft labels) from the Previous Classifier (g_t-1) using the categorical cross-entropy loss", wherein Tang discloses a second neural network, g_t-1, that has been trained using the previous task or the corresponding first dataset, which the network will used as a baseline for incremental learning of the current classifier, g_t. In other words, g_t-1 is being fine-tuned to be the current classifier, g_t, which is functionally equivalent to the claimed invention of using a trained second neural network for the incremental learning or fine-tuning.); and
train the second neural network model using the third dataset (Tang, Pg2832, Right Column, Paragraph1, Lines1-6, "If the label for the image is available, we train the classifier in a supervised manner. The embedding obtained from the Current Feature Extractor will be fed to the classifier network g_t to obtain class probabilities after softmax activation. Categorical cross-entropy loss LC^CT is used for training the classifier"
Pg2831, Right Column, Lines15-18, "Kaizen puts forward a general architecture in which both the feature extraction and the fine-tuning are performed continually from a stream of both large unlabeled and small labelled data."
Pg2830, Figure1, "Kaizen (b) leverages distillation across both the feature extraction and fine-tuning steps for each task", wherein Tang discloses the current classifier network g_t (the corresponding second neural network) which was trained above from g_t-1, is being trained continually or fine-tuned using the stream of labelled data or the current task (the corresponding third dataset). This third dataset can be replaced by the set of outliers found by Xia, which is then functionally equivalent to the claimed invention of fine-tunning the second network for continual learning using the third dataset.)
Xia and Tang are analogous to the claimed invention as they are from the same field of endeavor of computer vision tasks utilizing deep neural network models to process and learn from visual datasets. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date, to combine unsupervised autoencoder reconstruction-error thresholding for automatic outlier removal and data pattern separation of Xia with a practical self-supervised continual learning framework with continual fine-tuning that sequentially updates a feature extractor and a classifier from a continuous data stream. The motivation to combine is as recited by Tang (Tang, Pg2829, Right Column, Paragraph1, Lines10-12, “Data availability comes also with the cost of labelling samples for supervised learning or fine-tuning, since only a limited amount of data can be annotated at a time”) such that in practical machine learning deployments where filtering and annotating data samples manually is exceptionally time-consuming, expensive, and heavily restricted by data availability limitations, thus combination of the two can yield a system that automatically filter and isolate distinctive or novel data subsets from a noisy, unannotated stream without human intervention, thereby directly providing clean, high-quality data patterns to optimize the downstream end-to-end continual fine-tuning phase.
As to dependent Claim 16,
it is a non-transitory computer-readable medium claim that contains similar limitations of Claim 2 and thus rejected under the same rationale.
As to dependent Claim 18,
The combination of Xia and Tang teaches, as mentioned in Claim 15, about the overall instructions of using dual-pipeline of sifting incoming data with the trained autoencoder, that was trained with the first dataset, to collect outliers from the two sets of data patterns derived from the second dataset that exceeded the error threshold and input them into the second network as the third dataset to be continually fine-tuned.
Xia further teaches the non-transitory computer readable media of claim 15, wherein applying the first neural network model to the second dataset comprises:
translate, by the first neural network model based on a first function, the second dataset into encoded representations of the second dataset (Xia, Pg1512, Section3.1, Lines1-7, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector). We apply an autoencoder to first compress x into a lowdimensional intermediate representation and then map it back to a reconstructed copy, f(x). The form of an autoencoder f(·) is a neural network, with hidden linear or non-linear neurons", wherein the first neural network model doing the encoding using the first function is functionally identical to Xia’s compression of x into a lowdimensional intermediate representation of the set of images (the corresponding second dataset) that is just an encoding part of the autoencoder, rendering it functionally equivalent to the claimed invention),
translate, by the first neural network model based on a second function, the encoded representations of the second dataset into a reconstructed dataset of the second dataset (Xia, Pg1512, Section3.1, Lines1-7, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector). We apply an autoencoder to first compress x into a lowdimensional intermediate representation and then map it back to a reconstructed copy, f(x). The form of an autoencoder f(·) is a neural network, with hidden linear or non-linear neurons", wherein the first neural network doing the encoding using the second function is identical to Xia’s mapping back into the reconstructed copy, which is the decoding function of the autoencoder, rendering it functionally equivalent to the claimed invention), and
determine one or more data patterns in the reconstructed dataset of the second dataset based on the first dataset (Xia, Pg1512, Section3.1, Paragraph3, Lines6-8, "in order to minimize the overall reconstruction error, an autoencoder has to find the representations that can capture statistical regularities of training set"
Pg1512, Section3.1 Paragraph1, Lines7-8, "The reconstruction error of x_i is the squared loss:
PNG
media_image1.png
27
155
media_image1.png
Greyscale
”,
Pg1511, Abstract, Lines5-7, "We observe that when data are reconstructed from lowdimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors”, wherein the reconstructed copy (the corresponding reconstructed dataset of the second dataset) will be categorized into either an inlier or an outlier according to its reconstruction error such that the outliers will be selected once their reconstruction error exceeds the optimized threshold as mentioned above as the autoencoder has been trained using the first dataset which inherently compares the first and second datasets when computing the reconstruction errors, thus it is functionally equivalent to the claimed invention.)
As to dependent Claim 19,
The combination of Xia and Tang teaches, as mentioned in Claim 18, about the autoencoder (the corresponding first neural network) translates or encodes the data to obtain the representations and decodes those representations to reconstructed dataset of the second dataset where the reconstructed dataset comprises the data patterns.
Xia further teaches the non-transitory computer readable media of claim 18, wherein the operations further comprising:
apply the first neural network model to the first dataset (Xia, Pg1518, Figure10, "Figure 10. Comparisons on MNIST dataset when (a) images are represented by raw pixels and (b) images are represented by out puts of the fourth hidden layer in a pre-trained autoencoder", wherein the autoencoder (the corresponding first neural network) has been trained using the MNIST dataset, the corresponding first dataset),
translate, by the first neural network model based on the first function, the first dataset into encoded representation of the first dataset (Xia, Pg1517, Left Column, First Bullet Point, "MNIST [13] contains 60,000 handwritten digits from 0 to 9. For each category of digit, we simulate outliers as randomly sampled images from other categories",
Pg1518, Section5.4, Lines1-4, "For the raw pixel inputs of MNIST, we use four hidden layers (784-1000-500-250-30) and sigmoid neurons [9] in the autoencoder. The performance is shown in Fig. 10 (a), where DRAE(Discriminative Reconstruction Autoencoder) still achieves the best performance",
Pg1512, Section3.1, Lines1-7, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector). We apply an autoencoder to first compress x into a lowdimensional intermediate representation and then map it back to a reconstructed copy, f(x). The form of an autoencoder f(·) is a neural network, with hidden linear or non-linear neurons", wherein Xia discloses that autoencoder has been trained using the first dataset using DRAE, which uses the identical mechanism of identifying inliers and outliers by comparing reconstruction errors and a threshold, which inherently includes the step of translating the dataset into the encoded representation of the first dataset, rendering it functionally equivalent to the claimed invention),
translate, by the first neural network model based on the second function, the encoded representations of the first dataset into a reconstructed dataset of the first dataset (Xia, Pg1517, Left Column, First Bullet Point, "MNIST [13] contains 60,000 handwritten digits from 0 to 9. For each category of digit, we simulate outliers as randomly sampled images from other categories",
Pg1518, Section5.4, Lines1-4, "For the raw pixel inputs of MNIST, we use four hidden layers (784-1000-500-250-30) and sigmoid neurons [9] in the autoencoder. The performance is shown in Fig. 10 (a), where DRAE(Discriminative Reconstruction Autoencoder) still achieves the best performance",
Pg1512, Section3.1, Lines1-7, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector). We apply an autoencoder to first compress x into a lowdimensional intermediate representation and then map it back to a reconstructed copy, f(x). The form of an autoencoder f(·) is a neural network, with hidden linear or non-linear neurons", wherein Xia discloses that autoencoder has been trained using the first dataset using DRAE, which uses the identical mechanism of identifying inliers and outliers by comparing reconstruction errors and a threshold, which inherently includes the step of translating the encoded representations of the first dataset into a reconstructed dataset, rendering it functionally equivalent to the claimed invention),
determine one or more data patterns in the reconstructed dataset of the first dataset based on the first dataset (Xia, Pg1517, Left Column, First Bullet Point, "MNIST [13] contains 60,000 handwritten digits from 0 to 9. For each category of digit, we simulate outliers as randomly sampled images from other categories",
Pg1518, Section5.4, Lines1-4, "For the raw pixel inputs of MNIST, we use four hidden layers (784-1000-500-250-30) and sigmoid neurons [9] in the autoencoder. The performance is shown in Fig. 10 (a), where DRAE(Discriminative Reconstruction Autoencoder) still achieves the best performance",
Pg1512, Section3.1, Lines1-7, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector). We apply an autoencoder to first compress x into a lowdimensional intermediate representation and then map it back to a reconstructed copy, f(x). The form of an autoencoder f(·) is a neural network, with hidden linear or non-linear neurons",
Pg1511, Abstract, Lines5-7, "We observe that when data are reconstructed from lowdimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors”, wherein the autoencoder (the corresponding first neural network) that uses DRAE which functionally follows what it is doing to the second dataset, which separates the inliers and outliers (the corresponding data patterns), rendering it functionally equivalent to the claimed invention),
determine, by the first neural network model, the first set of data patterns and the second set of data patterns based on the second reconstructed dataset (Xia, Pg1512, Section3.1, Paragraph3, Lines6-8, "in order to minimize the overall reconstruction error, an autoencoder has to find the representations that can capture statistical regularities of training set"
Pg1512, Section3.1 Paragraph1, Lines7-8, "The reconstruction error of x_i is the squared loss:
PNG
media_image1.png
27
155
media_image1.png
Greyscale
”,
Pg1511, Abstract, Lines5-7, "We observe that when data are reconstructed from lowdimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors”,
Xia, Pg1511, Right Column, Paragraph1, Lines7-11, "Based on this, one can conveniently identify the images with large reconstruction errors as outliers. We were both surprised and excited in finding that simply thresholding the reconstruction errors can lead to competitive results with most existing methods", wherein the reconstructed dataset of the second dataset will be categorized under two groups, the inliers and the outliers, according to the reconstruction errors such that ones with higher reconstruction errors will be identified as the outliers, the second set of data patterns, and vice versa for the inliers, rendering it functionally equivalent to the claimed invention), and
determine, by the first neural network model, a third set of data patterns based on the first reconstructed dataset (Xia, Pg1512, Section3.1, Paragraph3, Lines6-8, "in order to minimize the overall reconstruction error, an autoencoder has to find the representations that can capture statistical regularities of training set"
Pg1512, Section3.1 Paragraph1, Lines7-8, "The reconstruction error of x_i is the squared loss:
PNG
media_image1.png
27
155
media_image1.png
Greyscale
”,
Pg1511, Abstract, Lines5-7, "We observe that when data are reconstructed from lowdimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors”,
Pg1512, Section 3.1, Lines8-9, Equation, "The autoencoder can be learned by minimizing the average reconstruction error:
PNG
media_image6.png
70
441
media_image6.png
Greyscale
", wherein the third set of data patterns corresponds to a normative profile or statistical regularity of the baseline population which was analyzed the reconstruction output of the first dataset. Xia discloses that during the baseline training reconstruction process, the autoencoder functions to capture the statistical regularities of the training set by minimizing the average reconstruction error meaning that it is functionally identical to determining a baseline profile, which is functionally equivalent to the claimed invention of finding a third set of data patterns)
wherein the reconstructed dataset of the second dataset is determined based on the reconstructed dataset of the first dataset (Xia, Pg1512, Section3.1, Paragraph3, Lines6-8, "in order to minimize the overall reconstruction error, an autoencoder has to find the representations that can capture statistical regularities of training set"
Pg1512, Section3.1 Paragraph1, Lines7-8, "The reconstruction error of x_i is the squared loss:
PNG
media_image1.png
27
155
media_image1.png
Greyscale
”,
Pg1511, Abstract, Lines5-7, "We observe that when data are reconstructed from lowdimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors”,
Xia, Pg1511, Right Column, Paragraph1, Lines7-11, "Based on this, one can conveniently identify the images with large reconstruction errors as outliers. We were both surprised and excited in finding that simply thresholding the reconstruction errors can lead to competitive results with most existing methods",
Pg1512, Section3.1, Lines1-7, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector). We apply an autoencoder to first compress x into a lowdimensional intermediate representation and then map it back to a reconstructed copy, f(x). The form of an autoencoder f(·) is a neural network, with hidden linear or non-linear neurons", wherein as mentioned above the trained autoencoder that outputs the reconstructed dataset of the second dataset, which functionally compares the reconstruction errors such that the autoencoder inherently compares them to the baseline data (the corresponding reconstructed dataset of the first dataset) to compute the reconstruction errors in order to identify reconstructed copies as inliers or outliers, thus rendering it functionally equivalent to the claimed invention.)
As to dependent Claim 20,
The combination of Xia and Tang teaches, as mentioned in Claim 19, about the autoencoder (the corresponding first neural network) translates or encodes the data to obtain the representations and decodes those representations to reconstructed dataset of the first dataset where the reconstructed dataset comprises the data patterns such that the dataset is used to determine the first and second sets of the data patterns based on the reconstructed dataset of the second dataset and the third dataset based on the reconstructed dataset of the first dataset to provide the baseline profile.
Xia further teaches the non-transitory computer readable media of claim 19, wherein the operations further comprising:
determine a first set of error scores based on the second set of data patterns (Xia, Pg1512, Section3.1 Paragraph1, Lines7-8, "The reconstruction error of x_i is the squared loss:
PNG
media_image1.png
27
155
media_image1.png
Greyscale
”,
Pg1512, Section3.1, Lines1-7, "Suppose we have a set {x1, ..., xn} where xi is an image representation (e.g., raw pixels or a feature vector). We apply an autoencoder to first compress x into a lowdimensional intermediate representation and then map it back to a reconstructed copy, f(x). The form of an autoencoder f(·) is a neural network, with hidden linear or non-linear neurons", wherein the set of images (the corresponding second dataset) will have a set of error scores as each data in the second dataset will have its own unique reconstruction error, e_i, where i represents i-th data in the set, which is functionally equivalent to the claimed invention),
determine a second set of error scores based on the third set of data patterns (Xia, Pg1512, Section3.1, Paragraph3, Lines6-8, "in order to minimize the overall reconstruction error, an autoencoder has to find the representations that can capture statistical regularities of training set"
Pg1512, Section3.1 Paragraph1, Lines7-8, "The reconstruction error of x_i is the squared loss:
PNG
media_image1.png
27
155
media_image1.png
Greyscale
”,
Pg1511, Abstract, Lines5-7, "We observe that when data are reconstructed from lowdimensional representations, the inliers and the outliers can be well separated according to their reconstruction errors”,
Pg1512, Section 3.1, Lines8-9, Equation, "The autoencoder can be learned by minimizing the average reconstruction error:
PNG
media_image6.png
70
441
media_image6.png
Greyscale
",
Pg1518, Section5.4, Lines1-4, "For the raw pixel inputs of MNIST, we use four hidden layers (784-1000-500-250-30) and sigmoid neurons [9] in the autoencoder. The performance is shown in Fig. 10 (a), where DRAE(Discriminative Reconstruction Autoencoder) still achieves the best performance", wherein as mentioned above the third set of data patterns represents the normative statistical regularities/distribution captured from the first dataset requires calculating a compilation of reconstruction error scores inherent to said baseline population. Xia discloses that individual reconstruction error scores, e_i, are calculated for every sing sample to evaluate the total average loss function of Equation1. This mathematical aggregation of baseline variations constitutes the literal determination of the claimed ‘second set of error scores’, rendering it functionally equivalent to the claimed invention), and
determine the error threshold based on the first set of error scores and the second set of error scores, (Xia, Pg1514, Subsection: Discriminative Labeling,
PNG
media_image2.png
359
494
media_image2.png
Greyscale
,
Pg1514, Section4.2, Paragraph3, "Since e_i is scalar, labeling data can be translated into sorting the reconstruction errors and then finding a cut-off threshold. Therefore, the objective in Equation 4 can be trivially optimized by linearly scanning an optimal thresh old. Then we label each sample x_i by comparing its reconstruction error e_i against the obtained optimal threshold",
Pg1512, Section3.1 Paragraph1, Lines7-8, "The reconstruction error of x_i is the squared loss:
PNG
media_image1.png
27
155
media_image1.png
Greyscale
”,
Pg1512, Section 3.1, Lines8-9, Equation, "The autoencoder can be learned by minimizing the average reconstruction error:
PNG
media_image6.png
70
441
media_image6.png
Greyscale
", wherein Xia establishes the optimal cut-off threshold boundary by solving an optimization objective designed to minimized the normalized within-class variance ratio of Equation 4. To mathematically evaluate this this objective function h during the linear threshold scanning process, the algorithm is explicitly required to concurrently calculate the summation of within-class variances, sigma_w, across two distinct sub-populations of the variance profile of the sorted inlier partition,
PNG
media_image7.png
28
151
media_image7.png
Greyscale
which represents evaluating and utilizing the second set of error scores derived from the baseline distribution, and the variance profile of the sorted outlier partition,
PNG
media_image8.png
30
154
media_image8.png
Greyscale
which represents evaluating and utilizing the first set of error scores derived from the incoming stream anomalies. Also, Xia discloses that the reconstruction error score for each datum is defined as
PNG
media_image1.png
27
155
media_image1.png
Greyscale
, which represents the nominal squared distance between the source matrix and its reconstructed counterpart. The average reconstruction error function is further minimized via the Equation 1. Consequently, the underlying ‘error scores’ themselves fundamentally represent and embody mean square error values. Also, Xia computes the within-class variance, sigma_w, which mathematically sums the squared deviations of individual error scores from their respective cluster means of
PNG
media_image9.png
29
391
media_image9.png
Greyscale
. Because compiling squared error magnitudes and scanning the resultant mean-squared variance profiles of both the normative inlier cluster (y_i = 1) and the variant outlier (y_i = 0) is the exact mathematical baseline to isolate the optimal threshold value range, such that both ways are functionally equivalent to the claimed invention.)
Claims 2, 3, 10, 11, 17 are rejected under 35 U.S.C. 103 as being unpatentable over Xia and Tang as mentioned in Claim 1 in further view of Julian et al. (Julian), Non-Patent Literature, “Never Stop Learning: The Effectiveness of Fine-Tuning in Robotic Reinforcement Learning”, published on July 2020, Pages: 13.
As to dependent Claim 2,
The combination of Xia and Tang teaches, as mentioned in Claim 1, the overall architecture of using dual-pipeline of sifting incoming data with the trained autoencoder to collect outliers as a new pattern and input them into the second network to be continually fine-tuned.
However, both Xia and Tang do not teach the following limitations, but from the same field of endeavor Julian teaches them:
the system of claim 1,
wherein the first dataset corresponds to data generated in the network during a first time period, wherein the second dataset corresponds to data generated in the network during a second time period; and wherein the first time period occurs before the second time period (Julian, Pg4, Right Column, Subsection A, Lines3-6, "First, we (1) pre-train a general grasping policy, as describe in Section III-A and [29]. To fine-tune a policy onto a new target task, we (2) use the pre-trained policy to collect an exploration dataset of attempts on the target task;"
Pg4, Figure3, "We pre-train a policy using the old data from the pre-training task, which is then adapted using the new data from the fine-tuning task", wherein the pre-training uses the old data (the corresponding first dataset) which then the trained model uses thew new data for fine-tuning procedure, which inherently indicates that the old data precedes the new data, leaving it functionally equivalent to the claimed invention.)
Xi, Tang and Julian are analogous to the claimed invention as they are from the same field of endeavor of deep neural network architectures and continuous machine learning frame works designed to process, classify, and adapt to non-stationary data streams or evolving environmental patterns. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date, to combine the unsupervised autoencoder reconstruction-error thresholding and outlier data pattern isolation technique of Xia, and the dual-network continual fine-tuning framework of Tang, with the multi-period sequential data processing, offline training, and continuous evaluation/development pipeline of Julian. The motivation is as recited by Julian (Julian, Pg1, Abstract, Lines1-3, “One of the great promises of robot learning systems is that they will be able to learn from their mistakes and continuously adapt to ever-changing environments”) such that when deployed in real-world systems characterized by persistently shifting and non-stationary data environments where static models suffer from severe degradation, integration of these components will have a system that can systematically manage data transitions over sequential time periods from incoming streams without human annotation, and seamlessly feed those isolated patterns into the dual-network architecture to fine-tune the downstream classifier.
As to dependent Claim 3,
The combination of Xia, Tang and Julian teaches, as mentioned in Claim 2, that the datasets being used for the training of the autoencoder and the second neural networks have different time periods where the first dataset precedes before the second dataset. Also, Claim 1 teaches about the autoencoder creating a third dataset that will be inputted to the second neural network for fine-tuning to continue the learning, which now the second neural network has the first dataset statistical regularities as well as the third dataset of outliers.
Julian further teaches the system of claim 2, wherein the operations further comprising:
apply the second neural network model to new data generated in the network to classify data patterns in the new data based on the first dataset and the third dataset, wherein the new data corresponds to data generated in the network at a third time period, and wherein the third time period occurs after the second time period (Julian, Pg4, Right Column, Subsection A, Lines14-19, "Finally, we (5) evaluate the fine-tuned policy on the target task. Our method is offline, i.e. it uses a single dataset of target task attempts, and requires no robot interaction after initial dataset collection to compute a fine-tuned policy, which may then be deployed onto a robot."
Pg3, Figure2, using the robot camera collect the online data
PNG
media_image10.png
275
396
media_image10.png
Greyscale
Pg4, Figure3
PNG
media_image11.png
434
375
media_image11.png
Greyscale
Pg6, Figure4, "Every transition to a new scenario happens after 800 grasp"
Pg6, Section V, Paragraph2, Lines5-8, "we use this adapted policy—not the base policy—as the initialization for another iteration of our fine-tuning algorithm, this time targeting “Transparent Bottles.” We repeat this process until we have run out of new tasks..."
Pg5, Right Column, Subsection d), Lines1-4, "d) Evaluate performance: Finally, we evaluate all 48 policies on their target task by deploying them to the robot and executing 50 or more grasp attempts to calculate the policy’s final performance", wherein Julian details a five-step conceptual framework for continual fine-tuning. Following the dataset collection and model adaptation phases conducted during the second time period, Julian's framework moves to step (5), which dictates evaluating and deploying the finalized, fine-tuned policy onto the operational robot. During this evaluation and deployment phase, which structurally constitutes a third time period occurring sequentially after the fine-tuning phase has concluded, the updated neural network is actively applied to entirely new real-time environmental data generated in the network. Under the Broadest Reasonable Interpretation, a 'network' encompasses any interconnected computing infrastructure where data is transmitted between nodes or devices. Julian explicitly discloses a distributed system architecture comprising real robotic systems, camera sensors, and centralized processing units where data is continuously streamed, collected, and deployed. In other words, once the robots start the evaluation phase, it will create the new data to be processed. Furthermore, Julian's continual learning sequence demonstrates successive iterations where the system repeatedly transitions to new operational scenarios after discrete data milestones, thus it is functionally equivalent to the claimed invention.)
As to dependent Claim 10,
The combination of Xia and Tang teaches, as mentioned in Claims 8 and 9, about the second neural network has been trained with the same dataset as the first neural network such that the second neural network is subject to the fine-tuning procedure using the third dataset.
However, both Xia and Tang do not teach the following limitations, but from the same field of endeavor Julian teaches them:
the computer-implemented method of claim 9, further comprising:
applying the second neural network model trained using the third dataset to new data generated in the computing network to classify data patterns based on the first dataset and the third dataset (Julian, Pg4, Right Column, Subsection A, Lines14-19, "Finally, we (5) evaluate the fine-tuned policy on the target task. Our method is offline, i.e. it uses a single dataset of target task attempts, and requires no robot interaction after initial dataset collection to compute a fine-tuned policy, which may then be deployed onto a robot."
Pg3, Figure2, using the robot camera collect the online data
PNG
media_image10.png
275
396
media_image10.png
Greyscale
Pg4, Figure3
PNG
media_image11.png
434
375
media_image11.png
Greyscale
Pg6, Figure4, "Every transition to a new scenario happens after 800 grasp"
Pg6, Section V, Paragraph2, Lines5-8, "we use this adapted policy—not the base policy—as the initialization for another iteration of our fine-tuning algorithm, this time targeting “Transparent Bottles.” We repeat this process until we have run out of new tasks..."
Pg5, Right Column, Subsection d), Lines1-4, "d) Evaluate performance: Finally, we evaluate all 48 policies on their target task by deploying them to the robot and executing 50 or more grasp attempts to calculate the policy’s final performance", wherein Julian details a five-step conceptual framework for continual fine-tuning. Following the dataset collection and model adaptation phases conducted during the second time period, Julian's framework moves to step (5), which dictates evaluating and deploying the finalized, fine-tuned policy onto the operational robot. During this evaluation and deployment phase, which structurally constitutes a third time period occurring sequentially after the fine-tuning phase has concluded, the updated neural network is actively applied to entirely new real-time environmental data generated in the network. Under the Broadest Reasonable Interpretation, a 'network' encompasses any interconnected computing infrastructure where data is transmitted between nodes or devices. Julian explicitly discloses a distributed system architecture comprising real robotic systems, camera sensors, and centralized processing units where data is continuously streamed, collected, and deployed. In other words, once the robots start the evaluation phase, it will create the new data to be processed, thus it is functionally equivalent to the claimed invention);
wherein training the second neural network model comprises fine tuning the second neural network model trained using the first dataset with the third dataset so as to prevent catastrophic forgetting by the second neural network model in classifying the data patterns (Tang, Pg2829, Abstract, Lines15-18, "we introduce a training architecture that is able to mitigate catastrophic forgetting
for both the feature extractor and classifier with a carefully designed loss function"
Pg2831, Right Column, Lines15-18, "Kaizen puts forward a general architecture in which both the feature extraction and the fine-tuning are performed continually from a stream of both large unlabeled and small labelled data", wherein Kaizen itself is a continual fine-tuning system which is primarily designed to mitigate the catastrophic forgetting during the additional training of the neural networks, rendering it functionally equivalent to the claimed invention.)
Xi, Tang and Julian are analogous to the claimed invention as they are from the same field of endeavor of deep neural network architectures and continuous machine learning frame works designed to process, classify, and adapt to non-stationary data streams or evolving environmental patterns. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date, to combine the unsupervised autoencoder reconstruction-error thresholding and outlier data pattern isolation technique of Xia, and the dual-network continual fine-tuning framework of Tang, with the multi-period sequential data processing, offline training, and continuous evaluation/development pipeline of Julian. The motivation is as recited by Julian (Julian, Pg1, Abstract, Lines1-3, “One of the great promises of robot learning systems is that they will be able to learn from their mistakes and continuously adapt to ever-changing environments”) such that when deployed in real-world systems characterized by persistently shifting and non-stationary data environments where static models suffer from severe degradation, integration of these components will have a system that can systematically manage data transitions over sequential time periods from incoming streams without human annotation, and seamlessly feed those isolated patterns into the dual-network architecture to fine-tune the downstream classifier.
As to dependent Claim 11,
it is a method claim that contains similar limitations of Claim 2 and thus rejected under the same rationale.
As to dependent Claim 17,
it is a non-transitory computer-readable medium claim that contains similar limitations of Claim 3 and thus rejected under the same rationale.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DONG YOON JUNG whose telephone number is (571)270-0198. The examiner can normally be reached 8am-5pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Cesar Paula can be reached at (571) 272-4128. 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.
/DONG YOON JUNG/Examiner, Art Unit 2145
/CESAR B PAULA/Supervisory Patent Examiner, Art Unit 2145
Prosecution Insights