DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
This action is in response to the arguments filed on 02/12/2026. Claims 1-20 are pending in the application and have been considered below.
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:
For Step 1, the claim is a computer-implemented method so it does recite a statutory category of invention.
For Step 2A, Prong 1:
The claim recites the limitation of “accessing, for an analytics session, analysis-goal information associated with an action to be achieved based on commands via an [analytics interface] of an application.” The accessing limitation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is nothing in the claim precludes the accessing step from practically being performed in the human mind. This limitation is a mental process.
The claim recites the limitation of “selecting, for the analysis-goal information, an analysis-goal from a plurality of analysis-goals, wherein the plurality of analysis-goals are identified based on corresponding analysis-goal models, wherein the analysis-goal
information is based on a user selection of the analysis-goal from the plurality of
analysis-goals.” The selecting limitation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is nothing in the claim precludes the selecting step from practically being performed in the human mind. This limitation is a mental process.
The claim recites the limitation of “identifying, for the analysis-goal, a probable command that corresponds to the analysis-goal associated with analyzing data via the analytics interface of the application, wherein the probable command is identified based on a plurality of goal-informed models that are machine-learning models.” The identifying generating limitation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is nothing in the claim precludes the identifying step from practically being performed in the human mind. This limitation is a mental process.
The claim recites the limitation of “generating a goal orientation score based on a goal-specific command probability distribution, wherein the goal orientation score quantifies a
degree to which the probable command aligns with the analysis-goal, wherein the
goal orientation score is separate from a probability of selecting the probable
command, wherein the goal orientation score is computed based on divergence
between a predicted command distribution and a goal-defined command
distribution.” The generating limitation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is nothing in the claim precludes the generating step from practically being performed in the human mind. This limitation is a mental process.
For Step 2A, Prong 2, the claim recites additional elements: analytics interface, analysis-goal models, “goal-informed models that are machine-learning models,”
“training the plurality of goal-informed models is based on offline training operations comprising:
training based on a plurality of previous sequences of commands observed
for corresponding analysis-goals in a first training stage, wherein training based
on the plurality of previous sequences of commands is based on using a custom
loss function, the custom loss function defines a plurality of probability
distributions that are different for the corresponding analysis-goals to generate a
plurality of fine-tuned models that are adjusted based on the custom loss function
to perform corresponding analysis-goals,
wherein the plurality of goal-informed models predict probable commands
for the corresponding analysis-goals,
wherein training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and
training based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal;” and “communicating the goal orientation score and the probable command as a command recommendation for the analysis.: goal, wherein the goal orientation score is provided as additional command recommendation data for guided assistance of a user via the analytics interface.”
The recited "analytics interface, analysis-goal models” and “goal-informed models that are machine-learning models “are generic computer components to apply an abstract idea under MPEP 2106.05 (f).
The “wherein training the plurality of goal-informed models is based on offline training operations comprising:
training based on a plurality of previous sequences of commands observed
for corresponding analysis-goals in a first training stage, wherein training based
on the plurality of previous sequences of commands is based on using a custom
loss function, the custom loss function defines a plurality of probability
distributions that are different for the corresponding analysis-goals to generate a
plurality of fine-tuned models that are adjusted based on the custom loss function
to perform corresponding analysis-goals,
wherein the plurality of goal-informed models predict probable commands
for the corresponding analysis-goals,
wherein training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and
training based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal” is a generic training recitation that may amount to a generic computer component to apply an abstract idea under 2106.05(f).
The “communicating” step is a form of insignificant extra-solution activity because it is a mere nominal or tangential addition to the claim. See MPEP 2106.05(g).
Step 2B
The additional elements of “analytics interface, analysis-goal models, “goal-informed models that are machine-learning models,” “wherein training the plurality of goal-informed models is based on offline training operations comprising:
training based on a plurality of previous sequences of commands observed
for corresponding analysis-goals in a first training stage, wherein training based
on the plurality of previous sequences of commands is based on using a custom
loss function, the custom loss function defines a plurality of probability
distributions that are different for the corresponding analysis-goals to generate a
plurality of fine-tuned models that are adjusted based on the custom loss function
to perform corresponding analysis-goals,
wherein the plurality of goal-informed models predict probable commands
for the corresponding analysis-goals,
wherein training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and
training based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal” do not amount to significantly more for the reasons set forth in step 2A above.
Additionally, under the Subject Matter Eligibility, a conclusion that an additional element is insignificant extra-solution activity in Step 2A should be reevaluated in Step 2B.
Here the “communicating” step was considered to be extra-solution activity in Step 2A, and thus it is reevaluated in Step 2B to determine if it is more than what is well-understood, routine, conventional activity in the field. The addition of insignificant extra-solution activity does not amount to an inventive concept, particularly when the activity is well-understood or conventional (MPEP 2106.05(d)). This appears to be well-understood, routine, conventional as evidenced by MPEP 2106.05(d)(II)(i).
i. Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using “analytics interface, analysis-goal models, “goal-informed models that are machine-learning models,” “wherein training the plurality of goal-informed models is based on offline training operations comprising:
training based on a plurality of previous sequences of commands observed
for corresponding analysis-goals in a first training stage, wherein training based
on the plurality of previous sequences of commands is based on using a custom
loss function, the custom loss function defines a plurality of probability
distributions that are different for the corresponding analysis-goals to generate a
plurality of fine-tuned models that are adjusted based on the custom loss function
to perform corresponding analysis-goals, wherein the plurality of goal-informed models predict probable commands for the corresponding analysis-goals,
wherein training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and
training based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal” to perform the claim steps amount to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 2:
Claim 2, which incorporates the rejection of claim 1, recites further limitations such as “
the analysis-goal information is based on a user selection of the analysis-goal from the plurality of analysis-goals, wherein the analysis-information is a phrase representing [[an]] the analysis-goal having a corresponding analysis-goal model” that are part of the abstract idea.
The claim recites an additional element: analysis-goal model.
The recited "analysis-goal model “is a generic computer component to apply an abstract idea under MPEP 2106.05 (f).The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using “analysis-goal model” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 3:
Claim 3, which incorporates the rejection of claim 1, recites further limitations such as
“the [analysis-goal models] are generated using log data, wherein the log data comprises the plurality of previous sequences of commands that are analyzed using a [bi-term topic model]” that are part of the abstract idea.
The claim recites an additional element: bi-term topic model.
The recited " bi-term topic model “is a generic computer component to apply an abstract idea under MPEP 2106.05 (f).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using “bi-term topic model” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 4:
Claim 4, which incorporates the rejection of claim 1, recites further limitations such as
“the plurality of goal-informed models are generated based on implicitly incorporating analysis-goal information and explicitly incorporating analysis-goal information into respective goal-informed models” that are part of the abstract idea.
The claim recites an additional element: goal-informed models.
The recited "goal-informed models “is a generic computer component to apply an abstract idea under MPEP 2106.05 (f).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using “goal-informed models” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 5:
Claim 5, which incorporates the rejection of claim 1, recites further limitations such as
“the plurality of goal-informed models are generated based on different machine-learning modeling techniques, wherein the machine-learning techniques are selected from the following: an ensemble technique, a goal concatenated representation technique, a goal concatenated command technique, and a goal appended inputs technique” that are part of the abstract idea.
There are no additional elements recited in this claim that amount to an integration of the judicial exception into a practical application or significantly more than the judicial exception. Therefore, the claim is not eligible.
Regarding Claim 6:
Claim 6, which incorporates the rejection of claim 1 recites further limitations such as
“wherein applying the custom loss function to the plurality of [goal-informed models] fine-tunes the plurality of [goal-informed models], wherein fine-tuning the plurality of goal-informed models is based on the probability distribution” that are part of the abstract idea.
The claim recites an additional element: goal-informed models.
The claim recites an additional element: goal-informed models.
The recited " goal-informed models “is a generic computer component to apply an abstract idea under MPEP 2106.05 (f).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using “goal-informed models” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 7:
Claim 7, which incorporates the rejection of claim 1 recites additional element: analytics interface.
The recited "analytics interface “is a generic computer component to apply an abstract idea under MPEP 2106.05 (f).
The “command recommendation panel” step is a form of insignificant extra-solution activity because it is a mere nominal or tangential addition to the claim. See MPEP 2106.05(g).
Under the Subject Matter Eligibility, a conclusion that an additional element is insignificant extra-solution activity in Step 2A should be reevaluated in Step 2B.
The “command recommendation panel” step is a form of insignificant extra-solution activity. The addition of insignificant extra-solution activity does not amount to an inventive concept, particularly when the activity is well-understood or conventional (MPEP 2106.05(d)). This appears to be well-understood, routine, conventional as evidenced by MPEP 2106.05(d)(II)(i).
i. Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using “"analytics interface “” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 8:
For Step 1, the claim is a computer-storage media having computer-executable instructions so it does recite a statutory category of invention.
For Step 2A, Prong 1:
The claim recites the limitation of “using the plurality of [goal-informed models], “generating a goal orientation score based on a goal-specific command probability distribution, wherein the goal orientation score quantifies a degree to which the probable command aligns with the analysis-goal, wherein the goal orientation score is separate from a probability of selecting the probable command, wherein the goal orientation score is computed based on divergence between a predicted command distribution and a goal-defined command distribution.” The generating limitation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is nothing in the claim precludes the generating step from practically being performed in the human mind. This limitation is a mental process.
For Step 2A, Prong 2, the claim recites additional elements: computing system, processor, memory, analytics interface, analysis-goal models, “goal-informed models that are machine-learning models,” “training a plurality of goal-informed models on a plurality of previous sequence of commands observed for corresponding analysis-goals in a first training stage, wherein training based on the plurality of previous sequence of commands is based on using a custom loss function, the custom loss function defines a plurality of probability distributions that are different for the corresponding analysis-goals to generate a plurality of fine-tuned models that are adjusted based on the custom loss function to perform corresponding analysis goals, wherein the plurality of goal-informed models predict probable commands for the corresponding analysis-goals, wherein training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and
training the plurality of goal-informed models based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal,” and
“communicating the goal orientation score and the probable command as a command recommendation for the analysis-goal, wherein the goal orientation score is provided as additional command recommendation data for guided assistance of a user via [[the]] an analytics interface.”
The processor is recited at a high level of generality, i.e., as a generic processor performing a generic computer function of processing data. This generic processor limitation is no more than mere instructions to apply the exception using a generic computer component. (MPEP 2106.05(f)).
The recited " computing system, processor, memory, analytics interface, analytics interface, analysis-goal models and “goal-informed models that are machine-learning models “are generic computer components to apply an abstract idea under MPEP 2106.05 (f).
The “training a plurality of goal-informed models on a plurality of previous sequence of commands observed for corresponding analysis-goals in a first training stage, wherein training based on the plurality of previous sequence of commands is based on using a custom loss function, the custom loss function defines a plurality of probability distributions that are different for the corresponding analysis-goals to generate a plurality of fine-tuned models that are adjusted based on the custom loss function to perform corresponding analysis goals,
wherein the plurality of goal-informed models predict probable commands
for the corresponding analysis-goals,
wherein training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and
training the plurality of goal-informed models based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal” is a generic training recitation that may amount to a generic computer component to apply an abstract idea under 2106.05(f).
The “communicating the goal orientation score and the probable command as a command recommendation for the analysis-goal, wherein the goal orientation score is provided as additional command recommendation data for guided assistance of a user via an analytics interface” step is a form of insignificant extra-solution activity because it is a mere nominal or tangential addition to the claim. See MPEP 2106.05(g).
Step 2B
The additional elements of “computing system, processor, memory, analytics interface, analysis-goal models, “goal-informed models that are machine-learning models,” “training a plurality of goal-informed models on a plurality of previous sequence of commands observed for corresponding analysis-goals in a first training stage, wherein training based on the plurality of previous sequence of commands is based on using a custom loss function, the custom loss function defines a plurality of probability distributions that are different for the corresponding analysis-goals to generate a plurality of fine-tuned models that are adjusted based on the custom loss function to perform corresponding analysis goals,
wherein the plurality of goal-informed models predict probable commands
for the corresponding analysis-goals,
wherein training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and
training the plurality of goal-informed models based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal” do not amount to significantly more for the reasons set forth in step 2A above.
Additionally, under the Subject Matter Eligibility, a conclusion that an additional element is insignificant extra-solution activity in Step 2A should be reevaluated in Step 2B.
The “communicating” step is a form of insignificant extra-solution activity. The addition of insignificant extra-solution activity does not amount to an inventive concept, particularly when the activity is well-understood or conventional (MPEP2106.05(d)).
This appears to be well-understood, routine, conventional in accordance with MPEP 2106.05(d)(II)(i), which states: The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. This appears to be well-understood, routine, conventional as evidenced by MPEP 2106.05(d)(II)(i).
i. Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using “computing system, processor, memory, analytics interface, analysis-goal models, “goal-informed models that are machine-learning models,” “training a plurality of goal-informed models on a plurality of previous sequence of commands observed for corresponding analysis-goals in a first training stage, wherein training based on the plurality of previous sequence of commands is based on using a custom loss function, the custom loss function defines a plurality of probability distributions that are different for the corresponding analysis-goals to generate a plurality of fine-tuned models that are adjusted based on the custom loss function to perform corresponding analysis goals,
wherein the plurality of goal-informed models predict probable commands
for the corresponding analysis-goals,
wherein training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and
training the plurality of goal-informed models based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal” to perform the claim steps amount to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 9:
Claim 9, which incorporates the rejection of claim 8, recites further limitations such as
accessing, for an analytics session, analysis-goal information associated with an action to be achieved based on commands via the [analytics interface] of an application;
selecting, for the analysis-goal information, the analysis-goal from a plurality of analysis-goals, wherein the plurality of analysis-goals are identified based on corresponding analysis-goal models; and
identifying, for the analysis-goal, the probable command that corresponds to the analysis-goal associated with analyzing data via the [analytics interface] of the application, wherein the probable command is identified based on the plurality of goal-informed
models that are [machine-learning models]” that are part of the abstract idea and do not amount to an inventive concept.
The claim recites additional elements: analysis-goal model, machine-learning models, and analytics interface.
The recited " analysis-goal model, machine-learning models, and analytics interface” are generic computer components to apply an abstract idea under MPEP 2106.05 (f).
The “training the plurality of goal-informed models is based on offline training operations” is a generic training recitation that may amount to a generic computer component to apply an abstract idea under 2106.05(f).
The additional elements of " analysis-goal model, machine-learning models, analytics interface” and “training the plurality of goal-informed models is based on offline training operations” do not amount to significantly more for the reasons set forth in step 2A above.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of " analysis-goal model, machine-learning models, analytics interface” and “training the plurality of goal-informed models is based on offline training operations” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 10
Claim 10, which incorporates the rejection of claim 8, recites further limitations such as “the analysis-goal information is based on a user selection of the analysis-goal from the plurality of analysis-goals, wherein the analysis-goal information is a phrase representing the analysis-goal having a corresponding [analysis-goal model]” that are part of the abstract idea.
The claim recites an additional element: analysis-goal model.
The recited " analysis-goal model “is a generic computer component to apply an abstract idea under MPEP 2106.05 (f).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using “an analysis-goal model” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 11
Claim 11, which incorporates the rejection of claim 8, recites further limitations such as
“the plurality of goal-informed models are generated based on implicitly incorporating analysis-goal information and explicitly incorporating analysis-goal information into respective [goal-informed models]” that are part of the abstract idea.
The claim recites an additional element: goal-informed models.
The recited " goal-informed models “is a generic computer component to apply an abstract idea under MPEP 2106.05 (f).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using “goal-informed models” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 12:
Claim 12, which incorporates the rejection of claim 8, recites further limitations such as “
machine-learning techniques are selected from the following: an ensemble technique, a goal concatenated representation technique, a goal concatenated command technique, and a goal appended inputs technique” that are part of the abstract idea.
There are no additional elements recited in this claim that amount to an integration of the judicial exception into a practical application or significantly more than the judicial exception. Therefore, the claim is not eligible.
Regarding Claim 13:
Claim 13, which incorporates the rejection of claim 8 recites further limitations such as
“analysis-goal models are generated using log data, wherein the log data comprises the plurality of previous sequences of commands that are analyzed using a bi-term topic model” that are part of the abstract idea.
The claim recites additional elements: analysis-goal models and bi-term topic models.
The recited analysis-goal models and bi-term topic models are generic computer component to apply an abstract idea under MPEP 2106.05 (f).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using “analysis-goal models and bi-term topic models” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 14:
Claim 14, which incorporates the rejection of claim 8 recites additional element: analytics interface.
The recited "analytics interface “is a generic computer component to apply an abstract idea under MPEP 2106.05 (f).
The “command recommendation panel” step is a form of insignificant extra-solution activity because it is a mere nominal or tangential addition to the claim. See MPEP 2106.05(g).
Under the Subject Matter Eligibility, a conclusion that an additional element is insignificant extra-solution activity in Step 2A should be reevaluated in Step 2B.
The “command recommendation panel” step is a form of insignificant extra-solution activity. The addition of insignificant extra-solution activity does not amount to an inventive concept, particularly when the activity is well-understood or conventional (MPEP 2106.05(d)). This appears to be well-understood, routine, conventional as evidenced by MPEP 2106.05(d)(II)(i).
i. Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using “analytics interface” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 15:
For Step 1, the claim is a computerized system so it does recite a statutory category of invention.
For Step 2A, Prong 1:
The claim recites the limitation of “accessing, for an analytics session, analysis-goal information associated with an action to be achieved based on commands via an [analytics interface] of an application.” The accessing limitation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is nothing in the claim precludes the accessing step from practically being performed in the human mind. This limitation is a mental process.
The claim recites the limitation of “selecting, for the analysis-goal information, an analysis-goal from a plurality of analysis-goals, wherein the plurality of analysis-goals are identified based on corresponding analysis-goal models, wherein the analysis-goal
information is based on a user selection of the analysis-goal from the plurality of
analysis-goals.” The selecting limitation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is nothing in the claim precludes the selecting step from practically being performed in the human mind. This limitation is a mental process.
The claim recites the limitation of “identifying, for the analysis-goal, a probable command that corresponds to the analysis-goal associated with analyzing data via the analytics interface of the application, wherein the probable command is identified based on a plurality of goal-informed models that are machine-learning models.” The identifying generating limitation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is nothing in the claim precludes the identifying step from practically being performed in the human mind. This limitation is a mental process.
The claim recites the limitation of “using the plurality of [goal-informed models],
generating a goal orientation score based on a goal-specific command probability distribution, wherein the goal orientation score quantifies a degree to which the probable command aligns with the analysis-goal, wherein the goal orientation score is separate from a probability of selecting the probable command, wherein the goal orientation score is computed based on divergence between a predicted command distribution and a goal-defined command distribution.”
The generating limitation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is nothing in the claim precludes the generating step from practically being performed in the human mind. This limitation is a mental process.
For Step 2A, Prong 2, the claim recites additional elements: computerized system, processors, computer storage media, analytics interface, analysis-goal models, “goal-informed models that are machine-learning models, “wherein training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and training based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal; and “communicating the goal orientation score and the probable command as a command recommendation for the analysis.: goal, wherein the goal orientation score is provided as additional command recommendation data for guided assistance of a user via the analytics interface.”
The processor is recited at a high level of generality, i.e., as a generic processor performing a generic computer function of processing data. This generic processor limitation is no more than mere instructions to apply the exception using a generic computer component. (MPEP 2106.05(f)).
The recited " computerized system, processors, computer storage media, analytics interface, analytics interface, analysis-goal models and “goal-informed models that are machine-learning models “are generic computer components to apply an abstract idea under MPEP 2106.05 (f).
The “training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and training based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal” is a generic training recitation that may amount to a generic computer component to apply an abstract idea under 2106.05(f).
The “communicating” step is a form of insignificant extra-solution activity because it is a mere nominal or tangential addition to the claim. See MPEP 2106.05(g).
Step 2B
The additional elements of “ computerized system, processors, computer storage media, analytics interface, analysis-goal models, “goal-informed models that are machine-learning models,” training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and training based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal” do not amount to significantly more for the reasons set forth in step 2A above.
Additionally, under the Subject Matter Eligibility, a conclusion that an additional element is insignificant extra-solution activity in Step 2A should be reevaluated in Step 2B.
The “communicating” step is a form of insignificant extra-solution activity. The addition of insignificant extra-solution activity does not amount to an inventive concept, particularly when the activity is well-understood or conventional (MPEP2106.05(d)).
This appears to be well-understood, routine, conventional in accordance with MPEP 2106.05(d)(II)(i), which states: The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. This appears to be well-understood, routine, conventional as evidenced by MPEP 2106.05(d)(II)(i).
i. Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of
“computerized system, processors, computer storage media, analytics interface, analysis-goal models, “goal-informed models that are machine-learning models,” training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and training based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal” to perform the claim steps amount to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 16:
Claim 16, which incorporates the rejection of claim 15, recites further limitations such as “…analysis-information is a phrase representing an analysis-goal having a corresponding analysis-goal model” that are part of the abstract idea.
The claim recites an additional element: analysis-goal model.
The recited "analysis-goal model “is a generic computer component to apply an abstract idea under MPEP 2106.05 (f).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using “analysis-goal model” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 17:
Claim 17, which incorporates the rejection of claim 15, recites further limitations such as “analysis-goal models are generated using log data, wherein the log data comprises the plurality of previous sequences of commands that are analyzed using a bi-term topic model” that are part of the abstract idea.
The claim recites additional elements: analysis-goal models and bi-term topic models.
The recited analysis-goal models and bi-term topic models are generic computer component to apply an abstract idea under MPEP 2106.05 (f).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using “analysis-goal models and bi-term topic models” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 18:
Claim 18, which incorporates the rejection of claim 15, recites further limitations such as
“[goal-informed models] are generated based on implicitly incorporating goal information and explicitly incorporating goal information into respective goal-informed models” that are part of the abstract idea.
The claim recites an additional element: goal-informed models.
The recited " goal-informed models “is a generic computer component to apply an abstract idea under MPEP 2106.05 (f).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using “goal-informed models” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 19:
Claim 19, which incorporates the rejection of claim 15, recites further limitations such as “[goal-informed models] are generated based on different machine-learning modeling techniques, wherein the machine-learning techniques are selected from the following: an ensemble technique, a goal concatenated representation technique, a goal concatenated command technique, and a goal appended inputs technique” that are part of the abstract idea.
There are no additional elements recited in this claim that amount to an integration of the judicial exception into a practical application or significantly more than the judicial exception. Therefore, the claim is not eligible.
Regarding Claim 20:
Claim 20, which incorporates the rejection of claim 15, recites further limitations such as “wherein applying the custom loss function to the plurality of [goal-informed models] fine-tunes the plurality of [goal-informed models], wherein fine-tuning the plurality of goal-informed models is based on the probability distribution” that are part of the abstract idea.
The claim recites an additional element: goal-informed models.
The recited " goal-informed models “is a generic computer component to apply an abstract idea under MPEP 2106.05 (f).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using “goal-informed models” to perform the claim steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-7 and 15-20 are rejected under 35 U.S.C. 103 as being unpatentable over Aadhavan M. Nambhi (“Stuck? No worries! Task-aware Command Recommendation and Proactive Help for Analysts,” hereinafter referred to as Nambhi), in view of Bao et al. (US 11,556,836 B1, hereinafter referred to as Bao), and further in view of Saidutta et al. (US 2023/0116456 A1, hereinafter referred to as Saidutta).
As to claim 1, Nambhi teaches a computer-implemented method, the method comprising:
accessing, for an analytics session, analysis-goal information (Abstract: log data of a web-based analytics software) associated with an action to be achieved based on commands via an analytics interface of an application (Introduction, interaction of a user with an analytics interface in terms of commands and tasks. Commands are the lowest level of interactions that a user can have with the UI (e.g., clicking on a button that sorts the data as per given column’s values, drag-and-drop actions, etc.). We assume that the commands are executed in a sequence to achieve intermediate goals, called tasks, achieve intermediate goals, called tasks);
selecting, for the analysis goal information (Abstract: log data of a web-based analytics software), an analysis-goal (Abstract: Fig. 1, Right: Proactive help is provided to the user) from a plurality of analysis-goals, wherein the plurality of analysis-goals are identified based on corresponding analysis-goal models (page 272, right column, wherein using the broadest reasonable interpretation, Examiner interprets “For our proactive help models we use the same usage log data, but additional pre-processing steps were carried out beyond the steps discussed above. Out of the 300 commands, 14 commands were manually identified to denote users’ requirement of help (for e.g., click on help icon) to teach the limitation) wherein the analysis-goal information is based on a user selection of the analysis-goal from the plurality of analysis-goals, (page 271, Introduction We model the interaction of a user with an analytics interface in terms of commands and tasks. Commands are the lowest level of interactions that a user can have with the UI (e.g., clicking on a button that sorts the data as per given column’s values, drag-and-drop actions, etc.). We assume that the commands are executed in a sequence to achieve intermediate goals, called tasks; (i) we propose a method to generate contextual command recommendations by incorporating ongoing task information, (ii) we propose an LSTM-based method to detect if a user is in need of help, and (iii) we comprehensively evaluate our proposed models to establish their superiority over competitive baselines. We believe our proposed methods will improve the quality of user interaction with data analytics software); wherein using the broadest reasonable interpretation, Examiner interprets “clicking on the commands buttons that are executed in a sequence to achieve intermediate goals, called tasks” to include a user selection of the analysis-goal);”
identifying, for the analysis-goal, a probable command that corresponds to the analysis-goal associated with analyzing data via the analytics interface of the application, wherein the probable command is identified based on a plurality of goal-informed models that are machined learning models (pages 272-273, 3.1 Command Recommendation Models: Task-aware Probabilistic Suffix Trees (TaskPST): In order to make the predictions of a PST dependent on the task, we introduce the notion of Task-aware PSTs. For each task identified by BTM, we train one dedicated PST using sequences that are most likely to belong to that particular task. If there are K tasks identified using BTM, this results in K PSTs. At test time, a sequence is passed to all of the PSTs, and output from individual PSTs, which are probability distributions over the entire command vocabulary, are first weighted according to the task distribution of the test sequence, and then added to get the final output. This ensures that the final output, which is again a probability distribution over the entire command vocabulary, is influenced proportionately by the output of individual PSTs based on the task distribution of the sequence. Our proposed models use multi-layers LSTM cells to encode the input sequence of commands into vectors of fixed dimensionality. These vectors are used by another LSTM (a decoder) to generate commands that align with the context of the sequence exposed so far”), wherein training the plurality of goal-informed models is based on [offline] training operations comprising:
training based on a plurality of previous sequences of commands observed for corresponding analysis-goals in a first training stage, wherein training based
on the plurality of previous sequences of commands is based on using a custom
loss function, the custom loss function defines a plurality of probability
distributions that are different for the corresponding analysis-goals to generate a
plurality of fine-tuned models that are adjusted based on the custom loss function to perform corresponding analysis-goals, wherein the plurality of goal-informed models predict probable commands for corresponding analysis-goals (page-273, 3.1 Command Recommendation Models: Task-aware Probabilistic Suffix Trees (TaskPST) During the training phase, the generated command is compared to the ground truth command and the loss is backpropagated to update model parameters. Mathematically, at each unfolding of the decoder LSTM, the following probabilities are computed to generate the next command; with page. 2/5, left column, Section 2 “command sequences” for “training”; page 3/5, train one dedicated PST using sequences that are most likely to belong to that particular task. If there are K tasks identified using BTM, this results in K PST; Task-aware RNN (TaskRNN): we have access to the entire sequence of commands at training time to determine the task distribution; Joint Task and Command RNN (JTC-RNN): During the training phase, the predicted task distribution is compared with the ground truth task distribution TS using Kullback-Leibler divergence [20]. Furthermore, at each timestep, the output of this module, i.e.,
PNG
media_image1.png
22
22
media_image1.png
Greyscale
, is concatenated with the trainable vector embeddings of current command in the sequence ct and is used by the second module to predict next command in the sequence ct+1; page 274, left column, last paragraph Also, we train our models with early stopping based on accuracy over validation set. While presenting the results in Table 1 and 2, we provide the average of quantified values over 5 different runs; pages 273-274, left column, second paragraph, during training we minimize cross-entropy loss L1 However, since JTC-RNN has an additional sub-module that estimates task-distribution at every timestep, it has an additional loss component – apart from L1 – which is given by KL divergence of estimated task distribution
PNG
media_image1.png
22
22
media_image1.png
Greyscale
with respect to the ground truth task distribution TS; 5 EVALUATION Table 1 summarizes the results of command recommendation models. Note that the superior performance of the TaskRNN model can be attributed to utilizing the task distribution of the entire sequence when recommending commands. Given that such information about the task distribution will not be available during model deployment, the results of JTC-RNN can be deemed
as realistic performance; wherein Examiner interprets the “command recommendation models based on performance” as fine-tuned models in view of paragraph [0035] of the specification);
wherein training using the custom loss function (page 274, left column, second paragraph, “During training, we minimize cross-entropy loss”) identifies additional
parameters associated with identifying a next command recommendation (pages 272-274, 3.1 Command Recommendation Models Variable length Markov Model (PST): In a Probabilistic Suffix Tree (PST), the root node is assigned a ‘null’ and every other node represents the sequence of commands that have to be executed in order to reach that node. The edge from a node to its children represents the probability of executing the next command in the sequence. Thus, given a sequence of commands which is represented as a node in a PST, we find the most probable future command by
traversing the edge with highest probability from the node to its children; wherein Examiner interprets “we find the most probable future command by traversing the edge with highest probability from the node to its children” to teach the limitation); and
generating a goal orientation score based on a goal-specific command probability distribution wherein the goal orientation score quantifies a degree to which the probable command aligns with the analysis-goal, wherein the goal orientation score is separate from a probability of selecting the probable command (3.1 Command Recommendation Models Variable length Markov Model (PST): In a Probabilistic Suffix Tree (PST), every other node represents the sequence of commands that have to be executed in order to reach that node. The edge from a node to its children represents the probability of executing the next command in the sequence. Thus, given a sequence of commands which is represented as a node in a PST, we find the most probable future command by traversing the edge with highest probability from the node to its children; wherein Examiner interprets “highest probability” as a score; Task-aware Probabilistic Suffix Trees (TaskPST): probabilities are computed to generate the next command Eq. (1); page 273 Joint Task and Command RNN (JTC-RNN): During the training phase, the predicted task distribution is compared with the ground truth task distribution TS using Kullback-Leibler divergence [20]. Furthermore, at each timestep, the output of this module, i.e.,
PNG
media_image1.png
22
22
media_image1.png
Greyscale
, is concatenated with the trainable vector embeddings of current command in the sequence ct and is used by the second module to predict next command in the sequence ct+1, 5 EVALUATION, left column, last paragraph To quantify the performance of command recommendation models, we use Top-1 and Top-5 accuracy. We rank order and compare the top 1 (and top 5) recommended command(s) by these models);
wherein the goal orientation score is computed based on divergence between a predicted command distribution and a goal-defined command distribution (page 273-274,
PNG
media_image2.png
336
748
media_image2.png
Greyscale
); and
communicating the goal orientation score and the probable command as a command recommendation for the analysis goal (3.1 Command Recommendation Models Task-aware Probabilistic Suffix Trees (TaskPST): probabilities are computed to generate the next command Eq. (1); “contextual information to recommend future commands”).
However, Nambhi fails to explicitly teach:
communicating the goal orientation score, wherein the goal orientation score is provided as additional command recommendation data for guided assistance of a user via the analytics interface.
Bao, in combination with Nambhi, teaches in recommending actions to users to achieve a goal (col. 18, lines 65-67, recommending specialist doctors to contact) communicating the goal orientation score and the probable recommendation, wherein the goal orientation score is provided as additional, recommendation data for guided assistance of a user via the analytics interface (Fig. 4 & col. 12, lines 42-54, “the specialist recommendation data is provided to the user, the specialist recommendation data includes a listing of multiple recommended specialists, a probability score or display indicating the calculated probability the specialists listed are a good match”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to provide to a user, along with the command/action recommendation of Nambhi, the probability score of Nambhi indicating the degree of recommendation, in the manner that Bao provides their recommendation probability score along with their recommendation. The motivation to do so is to provide additional information to a user to guide their choice, i.e. “why the specialist is being recommended” (Bao, column 12, lines 50-51).
However, Nambhi and Bao fail to explicitly teach wherein training the plurality of goal-informed models is based on offline training operations.
Saidutta, in combination with Nambhi and Bao, teaches wherein training the plurality of goal-informed models is based on offline training operations (paragraphs [0050]-[0051] In one or more embodiments, the learning model/component may be trained before the drilling operation using an "offline" training process with offline training data, as will be described in further detail below. The offline training may include one or more of historical data, expert recommendations, and simulation results. The offline trained model/component may then be fine-tuned for the particular well using online training data, e.g., real-time sensor measurements acquired during the drilling operation).
It would have been obvious to one of ordinary skill in the art before the effective filing of the claimed invention to modify the combination system of Nambhi and Bao to add offline training to the combination system of Nambhi and Bao, as taught by Saidutta above. The modification would have been obvious because one of ordinary skill would be motivated to fine tune the offline trained model to avoid online learning from scratch, as suggested by Saidutta ([0050]).
As to claim 2, which incorporates the rejection of claim 1, Nambhi teaches wherein the analysis-goal information is based on a user selection of the analysis-goal from the plurality of analysis-goals, wherein the analysis-goal information is a phrase representing an analysis-goal having a corresponding analysis-goal model (ABSTRACT task-aware command recommendation system, to guide the user on what commands could be executed next; 1 INTRODUCTION Fig. 1 "Click to expand”; 3.1 Command Recommendation Models - Task-aware Probabilistic Suffix Trees (TaskPST): “Our proposed models use multi-layers LSTM cells to encode the input sequence of commands into vectors of fixed dimensionality. These vectors are used by another LSTM (a decoder) to generate commands that align with the context of the sequence exposed so far” with page. 2, left column, Section 2 “command sequences” for “training”).
As to claim 3, which incorporates the rejection of claim 1, Nambhi teaches wherein the analysis-goal models are generated using log data, wherein the log data comprises the plurality of previous sequences of commands that are analyzed using a bi-term topic model (3 MODELS, treat each of the command sequences, which are obtained from the usage log data, as a document and train a topic model with Biterm Topic Modeling (BTM)).
As to claim 4, which incorporates the rejection of claim 1, Nambhi teaches wherein the plurality of goal- informed models are generated based on implicitly incorporating goal information and explicitly incorporating analysis-goal information into respective goal-informed models (Abstract- lntroduction, building machine learning models, commands are executed in a sequence to achieve intermediate goals, called tasks).
As to claim 5, which incorporates the rejection of claim 1, Nambhi teaches wherein the plurality of goal-informed models are generated based on different machine-learning modeling techniques, wherein the machine-learning techniques are selected from the following: an ensemble technique, a goal concatenated representation technique, a goal concatenated command technique, and a goal appended inputs technique (lntroduction We model the interaction of a user with an analytics interface in terms of commands and tasks. Commands are the lowest level of interactions that a user can have with the UI (e.g., clicking on a button that sorts the data as per given column’s values, drag-and-drop actions, etc.). We assume that the commands are executed in a sequence to achieve intermediate goals, called tasks; 3.1 Command Recommendation Models; 3.2 Proactive Help Models, proactive help recommendation as a supervised binary classification problem; Random Forest classifier and an LSTM classifier for this classification problem).
As to claim 6, which incorporates the rejection of claim 1, Nambhi teaches applying a loss function to the plurality of goal-informed models to fine-tune the plurality of goal-informed model, wherein fine-tuning the plurality of goal-informed models is based on probability distribution (pages 272-273, 3.1 Command Recommendation Models: Task-aware Probabilistic Suffix Trees (TaskPST): In order to make the predictions of a PST dependent on the task, we introduce the notion of Task-aware PSTs. For each task identified by BTM, we train one dedicated PST using sequences that are most likely to belong to that particular task. If there are K tasks identified using BTM, this results in K PSTs. At test time, a sequence is passed to all of the PSTs, and output from individual PSTs, which are probability distributions over the entire command vocabulary, are first weighted according to the task distribution of the test sequence, and then added to get the final output. This ensures that the final output, which is again a probability distribution over the entire command vocabulary, is influenced proportionately by the output of individual PSTs based on the task distribution of the sequence. Our proposed models use multi-layers LSTM cells to encode the input sequence of commands into vectors of fixed dimensionality. These vectors are used by another LSTM (a decoder) to generate commands that align with the context of the sequence exposed so far.” During the training phase, the generated command is compared to the ground truth command and the loss is backpropagated to update model parameters. Mathematically, at each unfolding of the decoder LSTM, the following probabilities are computed to generate the next command; pages 273-274, 5 EVALUATION Table 1 summarizes the results of command recommendation models. Note that the superior performance of the TaskRNN model can be attributed to utilizing the task distribution of the entire sequence when recommending commands. Given that such information about the task distribution will not be available during model deployment, the results of JTC-RNN can be deemed as realistic performance; wherein Examiner interprets the “command recommendation models based on performance” as fine-tuned models).
As to claim 7, which incorporates the rejection of claim 1, Nambhi teaches wherein the goal orientation score is associated with one or more interface elements visually presented using the analytics interface (1 INTRODUCTION., right column, Fig. 1 .... model the interaction of a user with an analytics interface in terms of commands and tasks; 5 EVALUATION, quantify the performance of command recommendation models, rank order and compare), wherein the goal orientation score and the probable command are caused to be presented via the analytics interface comprising a command recommendation panel that supports presenting probable commands (Figure 1: Left: The task-distribution of the ongoing sequence of commands is used to recommend future commands. Right: Proactive help is provided to the user; 3.1 Command Recommendation Models, Task-aware Probabilistic Suffix Trees (TaskPST): wherein using the broadest reasonable interpretation, Examiner interprets the probability of Eq. (1) as a score and “the task-distribution of the ongoing sequence of commands” to be included in the command recommendation panel).
As to claim 15, Nambhi teaches a computerized system comprising:
one or more processors; and
one or more computer storage media storing computer-useable
instructions that, when used by the one or more processors (page 272, “interactions with computer interfaces almost always involve users executing a complex series of commands,” Examiner interprets “computer interfaces” to be included in a computer), cause the one or more processors to execute:
accessing, for an analytics session, analysis-goal information associated with an action to be achieved based on commands via an analytics interface of an application (Introduction, interaction of a user with an analytics interface in terms of commands and tasks. Commands are the lowest level of interactions that a user can have with the UI (e.g., clicking on a button that sorts the data as per given column’s values, drag-and-drop actions, etc.). We assume that the commands are executed in a sequence to achieve intermediate goals, called tasks, achieve intermediate goals, called tasks);
selecting, for the analysis goal information, an analysis-goal from a plurality of analysis-goals, wherein the plurality of analysis-goals are identified based on corresponding analysis-goal models, wherein the analysis-goal information is based on a user selection of the analysis-goal from the plurality of analysis-goals, wherein the analysis-goal information is a phrase representing on analysis-goal having a corresponding analysis-goal model (Introduction, We model the
interaction of a user with an analytics interface in terms of commands and tasks. Commands are the lowest level of interactions that a user can have with the UI (e.g., clicking on a button that sorts the data as per given column’s values, drag-and-drop actions, etc.). We assume that the commands are executed in a sequence to
achieve intermediate goals, called tasks; ((i) we propose a method to generate contextual command recommendations by incorporating ongoing task information, (ii) we propose an LSTM-based method to detect if a user is in need of help, and (iii) we comprehensively evaluate our proposed models to establish their superiority over competitive baselines. We believe our proposed methods will improve the quality of user interaction with data analytics software);
identifying, for the analysis-goal, a probable command that corresponds to the analysis-goal associated with analyzing data via the analytics interface of the application, wherein the probable command is identified based on a plurality of goal-informed models that are machined learning models (pages 272-273, 3.1 Command Recommendation Models: Task-aware Probabilistic Suffix Trees (TaskPST): In order to make the predictions of a PST dependent on the task, we introduce the notion of Task-aware PSTs. For each task identified by BTM, we train one dedicated PST using sequences that are most likely to belong to that particular task. If there are K tasks identified using BTM, this results in K PSTs. At test time, a sequence is passed to all of the PSTs, and output from individual PSTs, which are probability distributions over the entire command vocabulary, are first weighted according to the task distribution of the test sequence, and then added to get the final output. This ensures that the final output, which is again a probability distribution over the entire command vocabulary, is influenced proportionately by the output of individual PSTs based on the task distribution of the sequence. Our proposed models use multi-layers LSTM cells to encode the input sequence of commands into vectors of fixed dimensionality. These vectors are used by another LSTM (a decoder) to generate commands that align with the context of the sequence exposed so far.”, wherein training the plurality of goal-informed models is based on offline training operations comprising:
training based on a plurality of previous sequences of commands observed for corresponding analysis-goals in a first training stage, wherein training based
on the plurality of previous sequences of commands is based on using a custom
loss function, the custom loss function defines a plurality of probability
distributions that are different for the corresponding analysis-goals to generate a
plurality of fine-tuned models that are adjusted based on the custom loss function to perform corresponding analysis-goals, wherein the plurality of goal-informed models predict probable commands for corresponding analysis-goals (pages 272-273, 3.1 Command Recommendation Models: Task-aware Probabilistic Suffix Trees (TaskPST): During the training phase, the generated command is compared to the ground truth command and the loss is backpropagated to update model parameters. Mathematically, at each unfolding of the decoder LSTM, the following probabilities are computed to generate the next command (page. 2, left column, Section 2 “command sequences” for “training”; page 3/5, train one dedicated PST using sequences that are most likely to belong to that particular task. If there are K tasks identified using BTM, this results in K PST; Task-aware RNN (TaskRNN): we have access to the entire sequence of commands at training time to determine the task distribution; Joint Task and Command RNN (JTC-RNN): During the training phase, the predicted task distribution is compared with the ground truth task distribution TS using Kullback-Leibler divergence [20]. Furthermore, at each timestep, the output of this module, i.e.,
PNG
media_image1.png
22
22
media_image1.png
Greyscale
, is concatenated with the trainable vector embeddings of current command in the sequence ct and is used by the second module to predict next command in the sequence ct+1; page 274, left column, last paragraph Also, we train our models with early stopping based on accuracy over validation set. While presenting the results in Table 1 and 2, we provide the average of quantified values over 5 different runs; pages 273-274, left column, second paragraph, we minimize cross-entropy loss L1 However, since JTC-RNN has an additional sub-module that estimates task-distribution at every timestep, it has an additional loss component – apart from L1 – which is given by KL divergence of estimated task distribution
PNG
media_image1.png
22
22
media_image1.png
Greyscale
with respect to the ground truth task distribution TS; 5 EVALUATION Table 1 summarizes the results of command recommendation models. Note that the superior performance of the TaskRNN model can be attributed to utilizing the task distribution of the entire sequence when recommending commands. Given that such information about the task distribution will not be available during model deployment, the results of JTC-RNN can be deemed
as realistic performance; wherein Examiner interprets the “command recommendation models based on performance” as fine-tuned models);
wherein training using the custom loss function (page 274, left column, second paragraph, “During training, we minimize cross-entropy loss”) identifies additional
parameters associated with identifying a next command recommendation (pages 272-274, 3.1 Command Recommendation Models Variable length Markov Model (PST): In a Probabilistic Suffix Tree (PST), the root node is assigned a ‘null’ and every other node represents the sequence of commands that have to be executed in order to reach that node. The edge from a node to its children represents the probability of executing the next command in the sequence. Thus, given a sequence of commands which is represented as a node in a PST, we find the most probable future command by
traversing the edge with highest probability from the node to its children; wherein Examiner interprets “we find the most probable future command by traversing the edge with highest probability from the node to its children” to teach the limitation); and
using the plurality of goal-informed models, generating a goal orientation score based on a goal-specific command probability distribution wherein the goal orientation score quantifies a degree to which the probable command aligns with the analysis-goal, wherein the goal orientation score is separate from a probability of selecting the probable command (3.1 Command Recommendation Models Variable length Markov Model (PST): In a Probabilistic Suffix Tree (PST), every other node represents the sequence of commands that have to be executed in order to reach that node. The edge from a node to its children represents the probability of executing the next command in the sequence. Thus, given a sequence of commands which is represented as a node in a PST, we find the most probable future command by traversing the edge with highest probability from the node to its children; wherein Examiner interprets “highest probability” as a score; Task-aware Probabilistic Suffix Trees (TaskPST): probabilities are computed to generate the next command Eq. (1); page 274, 5 EVALUATION, left column, last paragraph To quantify the performance of command recommendation models, we use Top-1 and Top-5 accuracy. We rank order and compare the top 1 (and top 5) recommended command(s) by these models);
wherein the goal orientation score is computed based on divergence between a predicted command distribution and a goal-defined command distribution (page 273-274,
PNG
media_image2.png
336
748
media_image2.png
Greyscale
wherein Examiner interprets the KL divergence used to compute to determine the divergence the command distribution); and
communicating the goal orientation score and the probable command as a command recommendation for the analysis goal (3.1 Command Recommendation Models Task-aware Probabilistic Suffix Trees (TaskPST): probabilities are computed to generate the next command Eq. (1); “contextual information to recommend future commands”).
However, Nambhi fails to explicitly teach:
communicating the goal orientation score, wherein the goal orientation score is provided as additional command recommendation data for guided assistance of a user via the analytics interface.
Bao, in combination with Nambhi, teaches in recommending actions to users to achieve a goal (col. 18, lines 65-67, recommending specialist doctors to contact) communicating the goal orientation score and the probable [recommendation], wherein the goal orientation score is provided as additional, recommendation data for guided assistance of a user via the analytics interface (Fig. 4 & col. 12, lines 42-54, “the specialist recommendation data is provided to the user, the specialist recommendation data includes a listing of multiple recommended specialists [and] a probability score or display indicating the calculated probability the specialists listed are a good match”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to provide to a user, along with the command/action recommendation of Nambhi, the probability score of Nambhi indicating the degree of recommendation, in the manner that Bao provides their recommendation probability score along with their recommendation. The motivation to do so is to provide additional information to a user to guide their choice, i.e. “why the specialist is being recommended” (Bao, column 12, lines 50-51).
However, Nambhi and Bao fail to explicitly teach wherein training the plurality of goal-informed models is based on offline training operations.
Saidutta, in combination with Nambhi and Bao, teaches wherein training the plurality of goal-informed models is based on offline training operations (paragraphs [0050]-[0051] In one or more embodiments, the learning model/component may be trained before the drilling operation using an "offline" training process with offline training data, as will be described in further detail below. The offline training may include one or more of historical data, expert recommendations, and simulation results. The offline trained model/component may then be fine-tuned for the particular well using online training data, e.g., real-time sensor measurements acquired during the drilling operation).
It would have been obvious to one of ordinary skill in the art before the effective filing of the claimed invention to modify the combination system of Nambhi and Bao to add offline training to the combination system of Nambhi and Bao, as taught by Saidutta above. The modification would have been obvious because one of ordinary skill would be motivated to fine tune the offline trained model to avoid online learning from scratch, as suggested by Saidutta ([0051).
As to claim 16, which incorporates the rejection of claim 15, Nambhi teaches wherein the analysis-goal information is based on a user selection of the analysis-goal from the plurality of analysis-goals, wherein the analysis-goal information is a phrase representing an analysis-goal having a corresponding analysis-goal model (ABSTRACT (task-aware command recommendation system, to guide the user on what commands could be executed next; 1 INTRODUCTION Fig. 1 "Click to expand”; 3.1 Command Recommendation Models - Task-aware Probabilistic Suffix Trees (TaskPST): “Our proposed models use multi-layers LSTM cells to encode the input sequence of commands into vectors of fixed dimensionality. These vectors are used by another LSTM (a decoder) to generate commands that align with the context of the sequence exposed so far” with page. 2, left column, Section 2 “command sequences” for “training”).
As to claim 17, which incorporates the rejection of claim 15, Nambhi teaches wherein the analysis-goal models are generated using log data, wherein the log data comprises the plurality of previous sequences of commands that are analyzed using a bi-term topic model (3 MODELS, treat each of the command sequences, which are obtained from the usage log data, as a document and train a topic model with Biterm Topic Modeling (BTM)).
As to claim 18, which incorporates the rejection of claim 15, Nambhi teaches wherein the plurality of goal- informed models are generated based on implicitly incorporating goal information and explicitly incorporating analysis-goal information into respective goal-informed models (Abstract- lntroduction, building machine learning models, commands are executed in a sequence to achieve intermediate goals, called tasks).
As to claim 19, which incorporates the rejection of claim 15, Nambhi teaches wherein the plurality of goal-informed models are generated based on different machine-learning modeling techniques, wherein the machine-learning techniques are selected from the following: an ensemble technique, a goal concatenated representation technique, a goal concatenated command technique, and a goal appended inputs technique (lntroduction We model the interaction of a user with an analytics interface in terms of commands and tasks. Commands are the lowest level of interactions that a user can have with the UI (e.g., clicking on a button that sorts the data as per given column’s values, drag-and-drop actions, etc.). We assume that the commands are executed in a sequence to achieve intermediate goals, called tasks; 3.1 Command Recommendation Models; 3.2 Proactive Help Models, proactive help recommendation as a supervised binary classification problem; Random Forest classifier and an LSTM classifier for this classification problem).
As to claim 20, which incorporates the rejection of claim 15, Nambhi fails to explicitly teach: applying a loss function to the plurality of goal-informed models to fine-tune the plurality of goal-informed model, wherein fine-tuning the plurality of goal-informed models is based on probability distribution.
However, Shen, in combination with Nambhi teaches:
applying a loss function to the plurality of goal-informed models to fine-tune the plurality of goal-informed model, wherein fine-tuning the plurality of goal-informed models is based on probability distribution ([0064], probability distribution; [0097]-[0098], fine-tune the model by minimizing L2 loss using a loss function; [0102]).
It would have been obvious to one of ordinary skill in the art before the effective filing of
the claimed invention to modify the system of Nambhi to add a loss function to the system of Nambhi, as taught by Shen, above. The modification would have been obvious because one of ordinary skill would be motivated to provide efficient recommendations and accurate predictions using the trained operation model, as suggested by Shen ([0057]).
Claims 8 and 10-14 are rejected under 35 U.S.C. 103 as being unpatentable over Aadhavan M. Nambhi (“Stuck? No worries! Task-aware Command Recommendation and Proactive Help for Analysts,” hereinafter referred to as Nambhi), in view of Bao et al. (US 11,556,836 B1, hereinafter referred to as Bao).
As to claim 8, Nambhi teaches one or more computer-storage media having computer-executable instructions embodied thereon that, when executed by a computing system having a processor and memory (page 272, “interactions with computer interfaces almost always involve users executing a complex series of commands,” Examiner interprets “computer interfaces” to be included in a computer to teach the limitations), cause the processor to perform operations comprising:
training a plurality of goal-informed models on a plurality of previous sequences of commands observed for corresponding analysis-goals in a first training stage, wherein training based on the plurality of previous sequences of commands is based on using a custom loss function, the custom loss function defines a plurality of probability distributions that are different for the corresponding analysis-goals to generate a plurality of fine-tuned models that are adjusted based on the custom loss function to perform corresponding analysis-goals, wherein the plurality of goal-informed models predict probable commands for corresponding analysis-goals (pages 272-273, 3.1 Command Recommendation Models: Task-aware Probabilistic Suffix Trees (TaskPST): In order to make the predictions of a PST dependent on the task, we introduce the notion of Task-aware PSTs. For each task identified by BTM, we train one dedicated PST using sequences that are most likely to belong to that particular task. If there are K tasks identified using BTM, this results in K PSTs. At test time, a sequence is passed to all of the PSTs, and output from individual PSTs, which are probability distributions over the entire command vocabulary, are first weighted according to the task distribution of the test sequence, and then added to get the final output. This ensures that the final output, which is again a probability distribution over the entire command vocabulary, is influenced proportionately by the output of individual PSTs based on the task distribution of the sequence. Our proposed models use multi-layers LSTM cells to encode the input sequence of commands into vectors of fixed dimensionality. These vectors are used by another LSTM (a decoder) to generate commands that align with the context of the sequence exposed so far.”
During the training phase, the generated command is compared to the ground truth command and the loss is backpropagated to update model parameters. Mathematically, at each unfolding of the decoder LSTM, the following probabilities are computed to generate the next command; with page. 2, left column, Section 2 “command sequences” for “training”; page 3/5, train one dedicated PST using sequences that are most likely to belong to that particular task. If there are K tasks identified using BTM, this results in K PST; Task-aware RNN (TaskRNN): we have access to the entire sequence of commands at training time to determine the task distribution; Joint Task and Command RNN (JTC-RNN): During the training phase, the predicted task distribution is compared with the ground truth task distribution TS using Kullback-Leibler divergence [20]. Furthermore, at each timestep, the output of this module, i.e.,
PNG
media_image1.png
22
22
media_image1.png
Greyscale
, is concatenated with the trainable vector embeddings of current command in the sequence ct and is used by the second module to predict next command in the sequence ct+1; page 274, left column, last paragraph Also, we train our models with early stopping based on accuracy over validation set. While presenting the results in Table 1 and 2, we provide the average of quantified values over 5 different runs; pages 273-274, left column, second paragraph, we minimize cross-entropy loss L1 However, since JTC-RNN has an additional sub-module that estimates task-distribution at every timestep, it has an additional loss component – apart from L1 – which is given by KL divergence of estimated task distribution
PNG
media_image1.png
22
22
media_image1.png
Greyscale
with respect to the ground truth task distribution TS; 5 EVALUATION Table 1 summarizes the results of command recommendation models. Note that the superior performance of the TaskRNN model can be attributed to utilizing the task distribution of the entire sequence when recommending commands. Given that such information about the task distribution will not be available during model deployment, the results of JTC-RNN can be deemed
as realistic performance; wherein Examiner interprets the “command recommendation models based on performance” as fine-tuned models);
wherein training using the custom loss function (page 274, left column, second paragraph, “During training, we minimize cross-entropy loss”) identifies additional
parameters associated with identifying a next command recommendation (pages 272-274, 3.1 Command Recommendation Models Variable length Markov Model (PST): In a Probabilistic Suffix Tree (PST), the root node is assigned a ‘null’ and every other node represents the sequence of commands that have to be executed in order to reach that node. The edge from a node to its children represents the probability of executing the next command in the sequence. Thus, given a sequence of commands which is represented as a node in a PST, we find the most probable future command by
traversing the edge with highest probability from the node to its children; wherein Examiner interprets “we find the most probable future command by traversing the edge with highest probability from the node to its children” to teach the limitation); and
communicating the goal orientation score and the probable command as a command recommendation for the analysis goal (3.1 Command Recommendation Models Task-aware Probabilistic Suffix Trees (TaskPST): probabilities are computed to generate the next command Eq. (1); “recommend future commands”);
training the plurality of goal-informed models based on the plurality of
probability distributions for the corresponding analysis-goals in the first training
stage, wherein training generates the plurality of fine-tuned models in a second
training stage, each for a corresponding analysis-goal (page 274, left column, last paragraph Also, we train our models with early stopping based on accuracy over validation set. While presenting the results in Table 1 and 2, we provide the average of quantified values over 5 different runs; pages 273-274, left column, second paragraph, we minimize cross-entropy loss L1 However, since JTC-RNN has an additional sub-module that estimates task-distribution at every timestep, it has an additional loss component – apart from L1 – which is given by KL divergence of estimated task distribution
PNG
media_image1.png
22
22
media_image1.png
Greyscale
with respect to the ground truth task distribution TS; 5 EVALUATION Table 1 summarizes the results of command recommendation models. Note that the superior performance of the TaskRNN model can be attributed to utilizing the task distribution of the entire sequence when recommending commands. Given that such information about the task distribution will not be available during model deployment, the results of JTC-RNN can be deemed as realistic performance; wherein Examiner interprets the “command recommendation models based on performance” as fine-tuned models);
using the plurality of goal-informed models, generating a goal orientation score based on a goal-specific command probability distribution wherein the goal orientation score quantifies a degree to which the probable command aligns with the analysis-goal, wherein the goal orientation score is separate from a probability of selecting the probable command (3.1 Command Recommendation Models Variable length Markov Model (PST): In a Probabilistic Suffix Tree (PST), every other node represents the sequence of commands that have to be executed in order to reach that node. The edge from a node to its children represents the probability of executing the next command in the sequence. Thus, given a sequence of commands which is represented as a node in a PST, we find the most probable future command by traversing the edge with highest probability from the node to its children; wherein Examiner interprets “highest probability” as a score; Task-aware Probabilistic Suffix Trees (TaskPST): probabilities are computed to generate the next command Eq. (1); page 274, 5 EVALUATION, left column, last paragraph To quantify the performance of command recommendation models, we use Top-1 and Top-5 accuracy. We rank order and compare the top 1 (and top 5) recommended command(s) by these models);
wherein the goal orientation score is computed based on divergence between a predicted command distribution and a goal-defined command distribution (page 273-274,
PNG
media_image2.png
336
748
media_image2.png
Greyscale
); and
However, Nambhi fails to explicitly teach:
communicating the goal orientation score, wherein the goal orientation score is provided as additional command recommendation data for guided assistance of a user via an analytics interface.
Bao, in combination with Nambhi, teaches in recommending actions to users to achieve a goal (col. 18, lines 65-67, recommending specialist doctors to contact) communicating the goal orientation score and the probable [recommendation], wherein the goal orientation score is provided as additional, recommendation data for guided assistance of a user via the analytics interface (Fig. 4 & col. 12, lines 42-54, “the specialist recommendation data is provided to the user, the specialist recommendation data includes a listing of multiple recommended specialists [and] a probability score or display indicating the calculated probability the specialists listed are a good match”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to provide to a user, along with the command/action recommendation of Nambhi, the probability score of Nambhi indicating the degree of recommendation, in the manner that Bao provides their recommendation probability score along with their recommendation. The motivation to do so is to provide additional information to a user to guide their choice, i.e. “why the specialist is being recommended” (Bao, column 12, lines 50-51).
As to claim 10, which incorporates the rejection of claim 9, Nambhi teaches
wherein the analysis-goal information is based on a user selection of the analysis-goal from the plurality of analysis-goals, wherein the analysis-goal information is a phrase representing the analysis-goal having a corresponding analysis-goal model (Introduction, right column, wherein Examiner interprets “Fig.1, left, “Task Description: Task description (i.e.” phrase”) is presented here” and first paragraph, “We model the interaction of a user with an analytics interface in terms of commands and tasks. Commands are the lowest level of interactions that a user can have with the UI (e.g., clicking on a button that sorts the data as per given column’s values, drag-and-drop actions, etc.). We assume that the commands are executed in a sequence to achieve intermediate goals, called tasks” to teach the limitation).
As to claim 11, which incorporates the rejection of claim 8, Nambhi teaches wherein the plurality of goal- informed models are generated based on implicitly incorporating goal information and explicitly incorporating analysis-goal information into respective goal-informed models (Abstract- lntroduction, building machine learning models, commands are executed in a sequence to achieve intermediate goals, called tasks).
As to claim 12, which incorporates the rejection of claim 8, Nambhi teaches wherein the plurality of goal-informed models are generated based on different machine-learning modeling techniques, wherein the machine-learning techniques are selected from the following: an ensemble technique, a goal concatenated representation technique, a goal concatenated command technique, and a goal appended inputs technique (lntroduction We model the interaction of a user with an analytics interface in terms of commands and tasks. Commands are the lowest level of interactions that a user can have with the UI (e.g., clicking on a button that sorts the data as per given column’s values, drag-and-drop actions, etc.). We assume that the commands are executed in a sequence to achieve intermediate goals, called tasks; 3.1 Command Recommendation Models; 3.2 Proactive Help Models, proactive help recommendation as a supervised binary classification problem; Random Forest classifier and an LSTM classifier for this classification problem).
As to claim 13, which incorporates the rejection of claim 8, Nambhi teaches wherein the analysis-goal models are generated using log data, wherein the log data comprises the plurality of previous sequences of commands that are analyzed using a bi-term topic model (page 272, right column, 3 MODELS, We treat each of the command sequences, which are obtained from the usage log data, as a document
and train a topic model with Biterm Topic Modeling (BTM). We use BTM, instead of more popular approaches for topic modeling such as Latent Dirichlet Allocation (LDA), to alleviate the data sparsity problem).
As to claim 14, which incorporates the rejection of claim 8, Nambhi teaches wherein the goal orientation score is associated with one or more interface elements visually presented using the analytics interface (1 INTRODUCTION., right column, Fig. 1, model the interaction of a user with an analytics interface in terms of commands and tasks; 5 EVALUATION, quantify the performance of command recommendation models, rank order and compare), wherein the goal orientation score and the probable command are caused to be presented via the analytics interface comprising a command recommendation panel that supports presenting probable commands (Figure 1: Left: The task-distribution of the ongoing sequence of commands is used to recommend future commands. Right: Proactive help is provided to the user; 3.1 Command Recommendation Models, Task-aware Probabilistic Suffix Trees (TaskPST): wherein using the broadest reasonable interpretation, Examiner interprets the probability of Eq. (1) as a score and “the task-distribution of the ongoing sequence of commands” to be included in the command recommendation panel).
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Aadhavan M. Nambhi (“Stuck? No worries! Task-aware Command Recommendation and Proactive Help for Analysts,” hereinafter referred to as Nambhi), in view of Bao et al. (US 11,556,836 B1, hereinafter referred to as Bao).and further in view of Saidutta et al. (US 2023/0116456 A1, hereinafter referred to as Saidutta).
As to claim 9, which incorporates the rejection of claim 8, Nambhi teaches:
accessing, for an analytics session, analysis-goal information associated
with an action to be achieved based on commands via the analytics interface of an application (Introduction, interaction of a user with an analytics interface in terms of commands and tasks. Commands are the lowest level of interactions that a user can have with the UI (e.g., clicking on a button that sorts the data as per given column’s values, drag-and-drop actions, etc.). We assume that the commands are executed in a sequence to achieve intermediate goals, called tasks, achieve intermediate goals, called tasks);
selecting, for the analysis-goal information, the analysis-goal from a
plurality of analysis-goals, wherein the plurality of analysis-goals are identified
based on corresponding analysis-goal models (Introduction, We model the
interaction of a user with an analytics interface in terms of commands and tasks. Commands are the lowest level of interactions that a user can have with the UI (e.g., clicking on a button that sorts the data as per given column’s values, drag-and-drop actions, etc.). We assume that the commands are executed in a sequence to
achieve intermediate goals, called tasks; (i) we propose a method to generate contextual command recommendations by incorporating ongoing task information, (ii) we propose an LSTM-based method to detect if a user is in need of help, and (iii) we comprehensively evaluate our proposed models to establish their superiority over competitive baselines. We believe our proposed methods will improve the quality of user interaction with data analytics software);
identifying, for the analysis-goal, the probable command that corresponds
to the analysis-goal associated with analyzing data via the analytics interface of
the application, wherein the probable command is identified based on the plurality of goal-informed models that are machine-learning models (3.1 Command Recommendation Models Task-aware Probabilistic Suffix Trees (TaskPST): "Our proposed models use multilayers LSTM cells to encode the input sequence of commands into vectors of fixed dimensionality. These vectors are used by another LSTM (a decoder) to generate commands that align with the context of the sequence exposed so far" with page. 2, left column, Section 2 "command sequences" for "training").
However, Nambhi and Bao fail to explicitly teach wherein training the plurality of goal-informed models is based on offline training operations.
Saidutta, in combination with Nambhi and Shen, teaches wherein training the plurality of goal-informed models is based on offline training operations.
Saidutta, in combination with Nambhi and Bao, teaches wherein training the plurality of goal-informed models is based on offline training operations (paragraphs [0050]- [0051] In one or more embodiments, the learning model/component may be trained before the drilling operation using an "offline" training process with offline training data, as will be described in further detail below. The offline training may include one or more of historical data, expert recommendations, and simulation results. The offline trained model/component may then be fine-tuned for the particular well using online training data, e.g., real-time sensor measurements acquired during the drilling operation).
It would have been obvious to one of ordinary skill in the art before the effective filing of the claimed invention to modify the combination system of Nambhi and Bao to add offline training to the combination system of Nambhi and Bao, as taught by Saidutta above. The modification would have been obvious because one of ordinary skill would be motivated to fine tune the offline trained model, as suggested by Saidutta ([0050]).
Response to Applicant’s arguments
Applicant's arguments on file on 02/12/2026 with respect to prior art rejection of claims 1-20 have been considered and are not persuasive.
Rejections based on 35 U.S.C. § 101
The claims are directed to patent eligible subject matter under the two-part analysis
set forth in Alice Corp. v. CLS Bank International, 134 S. Ct. 2347 (2014).
Under Part 1 of the Alice Analysis, the Claims Are Not Directed to an Abstract Idea.
Beginning with step one, Applicant respectfully submits that the amended claims
are not directed to a judicial exception. Rather, the claims are directed to a specific improvement in machine-learning-based command recommendation systems through a divergence-based goal alignment computation architecture.
Argument:
The incorporation of a divergence-based alignment computation into the command
recommendation pipeline constitutes an improvement to machine learning model architecture itself As in McRO, it is the incorporation of the claimed rules-not the mere use of a computer that improves the technological process. See McRO Inc. v. Bandai Namco Games Am. Inc., 837 F.3d 1299 (Fed. Cir. 2016).
The present claims do not recite a fundamental economic practice, a method of
Organizing human activity, or mental processes. The claims recite a specific computational technique for modifying model behavior using distributional divergence to quantify goal adherence. Such mathematical modeling integrated into a machine-learning system constitutes a technological improvement to computer functionality. Accordingly, under step one of Alice, the claims are not directed to an abstract idea.
Examiner response
Examiner respectfully disagrees. Applicant appears to assert that the claims are not directed to an abstract idea.
MPEP 2106.04(a) “Mental processes – concepts performed in the human mind (including an observation, evaluation, judgment, opinion).” See additionally see MPEP 2106.04(a)(2).
MPEP 2106.04(a)(2)(III)(C) “A Claim That Requires a Computer May Still Recite a Mental Process”
Examiner is interpreting the limitations of Claim 1as abstract ideas implemented on a generic computer. (Step 2A Prong 1).
The claim recites the limitation of “accessing, for an analytics session, analysis-goal information associated with an action to be achieved based on commands via an [analytics interface] of an application” is an observation or evaluation. This type of observation or evaluation is an act that can be practically performed in the human mind, similar to the mental thought processes that occur when a person has access to an information to analyze to achieve a goal or an objective. In other words, a person can analyze the information related to an objective and generate (i.e. provide) an observation or evaluation. Such mental observations or evaluations fall within the “mental processes” grouping of abstract idea set forth in the 2019 PEG. 2019 PEG Section I, 84 Fed. Reg. at 52.
MPEP 2106.04(a)(2)(III)(C) “A Claim That Requires a Computer May Still Recite a Mental Process.”
The claim recites the limitation of “selecting, for the analysis-goal information, an analysis-goal from a plurality of analysis-goals, wherein the plurality of analysis-goals are identified based on corresponding analysis-goal models, wherein the analysis-goal
information is based on a user selection of the analysis-goal from the plurality of
analysis-goals” is an observation or evaluation. This type of observation or evaluation is an act that can be practically performed in the human mind, similar to the mental thought processes that occur when a person selects an analysis related to a goal based on a model. In other words, a person can analyze the information related to an objective and generate (i.e. provide) an observation or evaluation. Such mental observations or evaluations fall within the “mental processes” grouping of abstract idea set forth in the 2019 PEG. 2019 PEG Section I, 84 Fed. Reg. at 52.
MPEP 2106.04(a)(2)(III)(C) “A Claim That Requires a Computer May Still Recite a Mental Process.”
The claim recites the limitation of “identifying, for the analysis-goal, a probable command that corresponds to the analysis-goal associated with analyzing data via the analytics interface of the application, wherein the probable command is identified based on a plurality of goal-informed models that are machine-learning models” is an observation or evaluation. This type of observation or evaluation is an act that can be practically performed in the human mind, similar to the mental thought processes that occur when a person identifies a command for analyzing goals associated with models. In other words, a person can analyze goals associated with models and provide an observation or evaluation. Such mental observations or evaluations fall within the “mental processes” grouping of abstract idea set forth in the 2019 PEG. 2019 PEG Section I, 84 Fed. Reg. at 52.
MPEP 2106.04(a)(2)(III)(C) “A Claim That Requires a Computer May Still Recite a Mental Process.”
The claim recites the limitation of “generating a goal orientation score based on a goal-specific command probability distribution, wherein the goal orientation score quantifies a
degree to which the probable command aligns with the analysis-goal, wherein the
goal orientation score is separate from a probability of selecting the probable
command, wherein the goal orientation score is computed based on divergence
between a predicted command distribution and a goal-defined command
distribution” is an observation or evaluation. This type of observation or evaluation is an act that can be practically performed in the human mind, similar to the mental thought processes that occur when a person generates a score using a probability distribution. In other words, a person can use a probability distribution and generate (i.e. provide) an observation or evaluation. Such mental observations or evaluations fall within the “mental processes” grouping of abstract idea set forth in the 2019 PEG. 2019 PEG Section I, 84 Fed. Reg. at 52.
MPEP 2106.04(a)(2)(III)(C) “A Claim That Requires a Computer May Still Recite a Mental Process.”
The “training the plurality of goal-informed models is based on offline training operations comprising:
training based on a plurality of previous sequences of commands observed
for corresponding analysis-goals in a first training stage, wherein training based
on the plurality of previous sequences of commands is based on using a custom
loss function, the custom loss function defines a plurality of probability
distributions that are different for the corresponding analysis-goals to generate a
plurality of fine-tuned models that are adjusted based on the custom loss function
to perform corresponding analysis-goals,
wherein the plurality of goal-informed models predict probable commands
for the corresponding analysis-goals,
wherein training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and
training based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal” is a generic training recitation that may amount to a generic computer component to apply an abstract idea under 2106.05(f).
The “communicating the goal orientation score and the probable command as a command recommendation for the analysis-goal, wherein the goal orientation score is provided as additional command recommendation data for guided assistance of a user via an analytics interface” step is a form of insignificant extra-solution activity because it is a mere nominal or tangential addition to the claim. See MPEP 2106.05(g).
The “communicating” step is a form of insignificant extra-solution activity. The addition of insignificant extra-solution activity does not amount to an inventive concept, particularly when the activity is well-understood or conventional (MPEP2106.05(d)).
This appears to be well-understood, routine, conventional in accordance with MPEP 2106.05(d)(II)(i), which states: The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. This appears to be well-understood, routine, conventional as evidenced by MPEP 2106.05(d)(II)(i).
i. Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
The newly added claim features do not improve the functionality of a computer or any technology (see rejection above).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using “computing system, processor, memory, analytics interface, analysis-goal models, “goal-informed models that are machine-learning models,” “training a plurality of goal-informed models on a plurality of previous sequence of commands observed for corresponding analysis-goals in a first training stage, wherein training based on the plurality of previous sequence of commands is based on using a custom loss function, the custom loss function defines a plurality of probability distributions that are different for the corresponding analysis-goals to generate a plurality of fine-tuned models that are adjusted based on the custom loss function to perform corresponding analysis goals, wherein the plurality of goal-informed models predict probable commands for the corresponding analysis-goals, wherein training using the custom loss function identifies additional parameters associated with identifying a next command recommendation; and
training the plurality of goal-informed models based on the plurality of probability distributions for the corresponding analysis-goals in the first training stage, wherein training generates the plurality of fine-tuned models in a second training stage, each for a corresponding analysis-goal” to perform the claim steps amount to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Therefore, Examiner respectfully maintains the 35 USC 101 rejection upon claim 1 as well as independent claims 8 and 15, as they recite substantially same limitations.
For the dependent claims, no further arguments were presented, Examiner respectfully maintains the 35 USC 101 rejections upon them, due to their nature of dependence upon their respective independent claims.
Under Part 2 of the Alice Analysis, the Claims Recite Significantly More.
Even assuming arguendo that the claims were deemed to involve an abstract idea,
the claims recite additional elements amounting to significantly more than the alleged abstract idea.
This is not conventional use of a generic machine learning model. Conventional next-command prediction systems produce a single probability distribution and select the highest probability command. The present claims require computing divergence between two distributions and generating a separate alignment metric distinct from prediction likelihood.
The amended claims meaningfully limit any alleged abstract concept by requiring
a particular computational architecture and divergence-based scoring mechanism. The ordered combination of these elements amounts to significantly more than any alleged abstract idea.
For at least the reasons discussed above, Applicant respectfully submits that the
amended claims are directed to patent-eligible subject matter under 35 U.S.C. § 101. The claims recite a specific technological improvement in machine learning systems through divergence based goal alignment scoring. Applicant respectfully requests withdrawal of the§ 101 rejection.
Examiner response
Examiner respectfully disagrees. Applicant appears to assert that the amended claims are directed to patent-eligible subject matter under 35 U.S.C. § 101.
To train a neural network, you use a loss function to measure the difference between
the model's predictions and the actual target values. The goal is to minimize this loss,
which guides the network to learn better mappings from inputs to outputs.
A loss function is a mathematical expression that quantifies how well a neural network
is performing on a given task. It essentially measures the error between the model's
predictions and the true values.
The primary purpose of a loss function is to provide a quantifiable measure of the
model's performance and guide the training process. By minimizing the loss function,
the model learns to adjust its parameters (weights and biases) to make more accurate
predictions.
The use of technology, particularly machine learning algorithms, to solve a technological
problem related to data analysis as recited in the instant claims does not make the
claims eligible under the Subject Matter Eligibility as shown in the rejection above.
Therefore, the recited training is a generic training recitation that may amount to a generic computer component to apply an abstract idea under 2106.05(f).
In the instant claims, Examiner respectfully interprets the recited "the training operations for machine learning models, the use of custom loss functions" as generic training recitation that may amount to a generic computer component to apply an abstract idea under 2106.0S(f) and "the generation a plurality of fine-tuned models based on probability distribution" as a generic computer component that amount to mere instructions to apply the abstract idea. See MPEP 2106.0S(f). The "custom loss function provides additional parameters to more effectively identify a next command recommendation that corresponds to a user's goal" is used in the training operations Therefore, it is included in the generic com put er component to apply an abstract idea under 2106.05(f).
For similar reasons, independent claims 8 and 15 are not patent eligible.
Rejections based on 35 U.S.C. § 103
Applicant submits that the combination of references fails to teach or suggest:
"generating a goal orientation score based on a goal-specific command
probability distribution, wherein the goal orientation score quantifies a degree to which the probable command aligns with the analysis-goal, wherein the goal orientation score is separate from a probability of selecting the probable command, wherein the goal orientation score is computed based on divergence between a predicted command distribution and a goal-defined command distribution" as recited in amended independent claim 1.
The Office relies primarily on Nambhi for allegedly teaching a probability
associated with a predicted command and interprets the highest probability command as satisfying the claimed "goal orientation score." However, Nambhi merely predicts a next command based on historical command sequences and outputs a probability reflecting the likelihood of that command occurring in sequence. Nambhi does not disclose generating a goal orientation score based on a goal-specific command probability distribution, nor does Nambhi disclose computing a divergence between a predicted command distribution and a goal-defined command distribution.
Nambhi conditions prediction on task context but does not define a goal as a separate command probability distribution. Nor does Nambhi compute any divergence or distance metric between two distributions. The cited reference produces a single distribution over next commands and selects the highest probability command from that distribution. There is no second goal-defined distribution and no comparison between distributions. Accordingly, Nambhi cannot satisfy the requirement that the goal orientation score be computed based on divergence between a predicted command distribution and a goal-defined command distribution.
Bao discloses presenting a probability score associated with recommended specialists. However, Bao' s probability reflects match confidence and does not involve divergence between two distributions, nor does it involve a goal-specific command probability distribution.
Bao's disclosure is unrelated to modeling goal-defined distributions or computing alignment metrics between predicted and goal-defined distributions. Saidutta concerns offline training techniques and similarly fails to disclose computing a divergence-based goal alignment score.
The Examiner's interpretation improperly conflates predictive likelihood with distributional alignment. Conditioning prediction on a task does not equate to computing a divergence-based alignment metric between two probability distributions. The amended claim language now expressly requires this structural and mathematical distinction, which is absent from the cited art. As such, the cited references fail to teach or suggest the features of amended claim 1.
Independent claims 8 and 15 recite novel features of the claimed invention and, for at least the reasons set forth above with respect to independent claim 1, the cited references similarly fail to teach or suggest the features of independent claims 8 and 15. Each of claims 2-7, 9-14, and 16-20 depends, either directly or indirectly, from one of independent claims 1, 8, and 15.
The additional references cited to reject the dependent claims likewise fail to cure
the above-noted deficiencies. Accordingly, by virtue of their dependency, it is respectfully submitted that the cited references fail to teach or suggest the features of these claims. Applicant respectfully requests withdrawal of the rejection of claims 1-20. The claims are believed to be in condition for allowance and such favorable action is respectfully requested.
Examiner response:
Examiner respect fully disagrees. As for Nambhi does not disclose “generating a goal orientation score based on a goal-specific command probability distribution,” nor does Nambhi disclose “computing a divergence between a predicted command distribution and a goal-defined command distribution.”
Nambhi teaches” given a sequence of commands which is represented as a node in a PST, we find the most probable future command by traversing the edge with highest probability from the node to its children; wherein Examiner interprets “highest probability” as a score (page 273, left column, first paragraph). Therefore, Nambhi does disclose “generating a goal orientation score based on a goal-specific command probability distribution.”
Nambhi further teaches “During the training phase, the predicted task distribution is compared with the ground truth task distribution TS using Kullback-Leibler divergence [20]. Furthermore, at each timestep, the output of this module, i.e.,
PNG
media_image1.png
22
22
media_image1.png
Greyscale
, is concatenated with the trainable vector embeddings of current command in the sequence ct and is used by the second module to predict next command in the sequence ct+1 (page 273, right column, second paragraph). Nambhi does teaches the” Kullback-Leibler divergence” which is a statistical measure from information theory that quantifies how much one probability distribution differs from a reference distribution. Therefore, Nambhi does disclose “computing a divergence between a predicted command distribution and a goal-defined command distribution.”
As for “The Examiner's interpretation improperly conflates predictive likelihood with distributional alignment.” The Examiner's interpretation does not improperly conflate predictive likelihood with distributional alignment. Not only the claim does not recite “distributional alignment” but Nambhi does teach “During the training phase, the predicted task distribution is compared with the ground truth task distribution TS using Kullback-Leibler divergence” (see explanation above).
The cited references do teach or suggest the features of independent claims 1, 8 and 15.
No further arguments were presented for the dependent claims. By virtue of their dependency, it is respectfully submitted that the cited references do teach or suggest the features of these claims. Examiner respectfully maintains the rejection of claims 1-20. Therefore, the claims are not in condition for allowance.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ABABACAR SECK whose telephone number is (571)270-7146. The examiner can normally be reached Monday-Friday 8:00 A.M.-6:00 P.M..
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, Lamardo Viker can be reached on 571-270-5871. 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.
/ABABACAR SECK/Examiner, Art Unit 2122
/VIKER A LAMARDO/Supervisory Patent Examiner, Art Unit 2147
Prosecution Insights