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 .
Response to Amendment
Acknowledgement is made of Applicant’s claim amendments on 03/05/2026. The claim amendments are entered. Presently, claims 1-19 remain pending. Claims 1 and 12 have been amended.
Response to Arguments
Regarding the 35 USC 112(b) indefinite rejection, Applicant's arguments filed 03/05/2026 have been fully considered but they are not persuasive. Claims 6 and 15 remain unamended, therefore, the 112(b) rejection are maintained.
Regarding the 35 USC 101 rejection, Applicant's arguments filed 03/05/2026 have been fully considered but they are not persuasive.
Applicant argues: The claims are directed towards an improvement and amount to significantly more than the judicial exception (pages 7-8 of remarks).
Examiner response: Examiner respectfully disagrees. Independent claims 1 and 10 recite a number of judicial exceptions such as “determining…”, “generating…”, and “wherein the dissimilarities are determined…”. These steps are directed to mental processes as they encompass an observation, evaluation, judgement, or opinion. See MPEP § 2106.04(a)(2), subsection III. Applicant argues the claimed invention may solve a technical problem in task continual learning by using a task similarity function that does not require additional learning. However, determining similarities and dissimilarities is practically implementable in the human mind as a human may determine whether a new task is similar or not similar to a previous task. MPEP 2106.05(a)(II) states “[h]owever, it is important to keep in mind that an improvement in the abstract idea itself (e.g. a recited fundamental economic concept) is not an improvement in technology”, therefore, determining similarities and dissimilarities between a new and previous task cannot be an improvement in technology. In addition, the claims recite “applying the previously used task-specific encoder to the new task based on the generated test error value”. This limitation is directed to mere instructions to apply the judicial exception generic computer component as the encoder is being generally applied to the new task. See MPEP 2106.05(f). This does not integrate the claims into a practical application and does not amount to significantly more than the judicial exception. Arguments are not persuasive.
Regarding the 35 USC 103 rejections, Applicant’s arguments with respect to claims 1-19 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Regarding the limitation “determining, in response to receiving a new task, one or more of similarities and dissimilarities between at least one previously learned task and the new task”, Rostami teaches determining a similarity between the distribution of the new task and previously learned tasks (Rostami para [0068] “As such, the system operates based on training encoder ϕ.sub.v 304 such that all tasks share a similar distribution in the embedding, i.e., the new tasks are learned such that their distribution in the embedding match the past experience, captured in the shared distribution.”). Rostami also teaches determining discrepancies (i.e., dissimilarities) between the probability distributions of the new task and previously learned tasks (Rostami para [0071] “Let {circumflex over (p)}.sub.J.sup.(0)(z) denote this parametric distribution. This distribution is updated after learning each task to incorporate what has been learned from the new task. As a result, this distribution captures knowledge about the past.” Para [0072] “where D(.Math.,.Math.) is a discrepancy measure, i.e., a metric, between two probability distributions and λ is a trade-off parameter. The first four terms in Eq. 2 are empirical classification risk and autoencoder reconstruction loss terms for the current task and the generated pseudo-dataset. The third and the fourth terms enforce learning (updating) the current task 400 such that the past learned knowledge is not forgotten. The fifth term is added to enforce the learned embedding distribution for the current task 400 to be similar to what has been learned in the past, i.e., task-invariant.”). A new ground of rejection is made to address the limitation “wherein the dissimilarities are determined when at least one of data from the one or more previously learned tasks belonging to the same group of classes or predictors of the one or more previously learned tasks being drawn from a same distribution is not determined”.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-19 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claims 1 and 10 recite the limitation: “wherein the dissimilarities are determined when at least one of data from the one or more previously learned tasks belonging to the same group of classes or predictors of the one or more previously learned tasks being drawn from a same distribution is not determined”. It is unclear to one of ordinary skill in the art how dissimilarities between a previously learned task and a new task can be determined from only determining using data from the previously learned tasks belong to the same group of classes or predictors. According to pages 10-11 of Applicant’s specification, the dissimilarities are determined when data from the previously learned task and the new task do not belong to the same group of classes or predictors and not only the previously learned task. For Examination purposes, Examiner interprets the limitation “wherein the dissimilarities are determined when at least one of data from the one or more previously learned tasks and one of data from the new task belonging to the same group of classes or predictors of the one or more previously learned tasks being drawn from a same distribution is not determined”. Claims 2-9 and 11-19 are dependent claims that do not cure the deficiencies and are rejected for the same reasons.
Claim 6 recites the limitation "the re-used task specific encoder" in “wherein the encoder portion of the task specific encoder is used to train only the classifier head for the new task based on the re-used task specific encoder or the trained new task specific encoder”. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, Examiner interprets the limitation as “wherein the encoder portion of the task specific encoder is used to train only the classifier head for the new task based on the previously used task specific encoder or the trained new task specific encoder”.
Claim 15 recites the limitation "the server" in “transmitting the classification head for the new task to the server to be stored in the memory”. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, Examiner interprets the limitation as “transmitting the classification head for the new task to a server to be stored in the memory”.
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 1-19 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1
According to the first part of the analysis, in the instant case, claims 1-9 are directed to a method and claims 10-19 are directed to a user equipment comprising at least a processor. Thus, each of the claims falls within one of the four statutory categories (i.e. process, machine, manufacture, or composition of matter).
Claim 1 recites:
Step 2A, prong 1
“determining, in response to receiving a new task, one or more of similarities and dissimilarities between at least one previously learned task and the new task” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can determine similarities and dissimilarities between a new task and previous task (i.e. observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
“generating, based on the one or more similarities and dissimilarities determined and a previously used task-specific encoder corresponding to the at least one previously learned task, a test error value for classifying the new task” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can generate an error value based on similarities and dissimilarities between tasks and an output from a task-specific encoder (i.e. observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
“wherein the dissimilarities are determined when at least one of data from the one or more previously learned tasks belonging to the same group of classes or predictors of the one or more previously learned tasks being drawn from a same distribution is not determined.” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can determine dissimilarities when a previous task does not belong to a same group of classifications or prediction models (i.e. observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
Step 2A, prong 2
“maintaining a memory comprising one or more previously learned tasks” (storing data in memory to perform the abstract idea is considered “mere data gathering” and is an insignificant extra-solution activity, which does not integrate the abstract idea into a practical application. See MPEP 2106.05 (g).).)
“applying the previously used task-specific encoder to the new task based on the generated test error value” (mere instructions to apply the judicial exception using a generic computer component. See MPEP 2106.05(f).)
The additional elements do not integrate into a practical application.
Step 2B
“maintaining a memory comprising one or more previously learned tasks” (This step is directed to storing data in memory, which is well-understood, routine, and conventional. iv. Storing and retrieving information in memory, Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015); (" See MPEP 2106.05 (d) (II)).
“applying the previously used task-specific encoder to the new task based on the generated test error value” (mere instructions to apply the judicial exception using a generic computer component. See MPEP 2106.05(f).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 2 recites:
Step 2A, prong 1
“determining whether a similar task to the new task is stored in the memory” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can determine a new task is similar to a previous task (i.e. observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
Step 2A, prong 2
“wherein the previously used task specific encoder is applied to the new task in response to determining that the similar task is stored in the memory” (mere instructions to apply the judicial exception using a generic computer component. See 2106.05(f).)
The additional elements do not integrate into a practical application.
Step 2B
“wherein the previously used task specific encoder is applied to the new task in response to determining that the similar task is stored in the memory” (mere instructions to apply the judicial exception using a generic computer component. See 2106.05(f).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 3 recites:
Step 2A, prong 1
Claim 3 recites at least the abstract idea identified above in claim 2.
Step 2A, prong 2
“training a new task specific encoder when determining that no similar task to the new task is stored in the memory” (mere instructions to apply the judicial exception using a generic computer component. See MPEP 2106.05(f).)
“storing the new task specific encoder in the memory.” (storing data in memory to perform the abstract idea is considered “mere data gathering” and is an insignificant extra-solution activity, which does not integrate the abstract idea into a practical application. See MPEP 2106.05 (g).).)
The additional elements do not integrate into a practical application.
Step 2B
“training a new task specific encoder when determining that no similar task to the new task is stored in the memory” (mere instructions to apply the judicial exception using a generic computer component. See MPEP 2106.05(f).)
“storing the new task specific encoder in the memory.” (This step is directed to storing data in memory, which is well-understood, routine, and conventional. iv. Storing and retrieving information in memory, Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015); See MPEP 2106.05 (d) (II)).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 4 recites:
Step 2A, prong 1
Claim 4 recites at least the abstract idea identified above in claim 2.
Step 2A, prong 2
“using an encoder portion of the task specific encoder as a feature extraction backbone to train only a classifier head for the new task” (Linking the judicial exception to a particular technological environment or field of use. See MPEP 2106.05(h).)
The additional elements do not integrate into a practical application.
Step 2B
“using an encoder portion of the task specific encoder as a feature extraction backbone to train only a classifier head for the new task” (Linking the judicial exception to a particular technological environment or field of use. See MPEP 2106.05(h).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 5 recites:
Step 2A, prong 1
Claim 5 recites at least the abstract idea identified above in claim 2.
Step 2A, prong 2
“storing the classification head for the new task in the memory” (Storing data in memory to perform the abstract idea is considered “mere data gathering” and is an insignificant extra-solution activity, which does not integrate the abstract idea into a practical application. See MPEP 2106.05 (g).).)
The additional elements do not integrate into a practical application.
Step 2B
“storing the classification head for the new task in the memory” (This step is directed to storing data in memory, which is well-understood, routine, and conventional. iv. Storing and retrieving information in memory, Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015); (" See MPEP 2106.05 (d) (II)).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 6 recites:
Step 2A, prong 1
Claim 6 recites at least the abstract idea identified above in claim 2.
Step 2A, prong 2
“wherein the encoder portion of the task specific encoder is used to train only the classifier head for the new task based on the re-used task specific encoder or the trained new task specific encoder” (Linking the judicial exception to a particular technological environment or field of use. See MPEP 2106.05(h).)
The additional elements do not integrate into a practical application.
Step 2B
“wherein the encoder portion of the task specific encoder is used to train only the classifier head for the new task based on the re-used task specific encoder or the trained new task specific encoder” (Linking the judicial exception to a particular technological environment or field of use. See MPEP 2106.05(h).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 7 recites:
Step 2A, prong 1
Claim 7 recites at least the abstract idea identified above in claim 2.
Step 2A, prong 2
“wherein a style modulation technique is used to train the new task specific encoder when determining that no similar task to the new task is stored in the memory” (Linking the judicial exception to a particular technological environment or field of use. See MPEP 2106.05(h).)
The additional elements do not integrate into a practical application.
Step 2B
“wherein a style modulation technique is used to train the new task specific encoder when determining that no similar task to the new task is stored in the memory” (Linking the judicial exception to a particular technological environment or field of use. See MPEP 2106.05(h).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 8 recites:
Step 2A, prong 1
“wherein the similarity is determined based on a score of a training memory for a task, the score being calculated by a distribution consistency estimator” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can determine a similarity between tasks based on a score (i.e. observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
Step 2A, Prong 2 & 2B
This claim does not recite any additional elements.
Claim 9 recites:
Step 2A, prong 1
“wherein the similarity is determined based on a score calculated in a predictor-label association analysis” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can determine a similarity between tasks based on a score in a label prediction analysis (i.e. observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
Step 2A, Prong 2 & 2B
This claim does not recite any additional elements.
Claim 10 recites: See the rejection of claim 1 above. Same rationale applies.
2A Prong 2 & 2B: The claim recites another additional element “A user equipment (UE), comprising: at least one processor; and at least one memory operatively connected with the at least one processor, the at least one memory storing instructions, which when executed, instruct the at least one processor to perform a method of continual learning” (mere instructions to apply the exception using a generic computer component)
Claim 11 recites: See the rejection of claim 2 above. Same rationale applies.
Claim 12 recites:
Step 2A, prong 1
Claim 12 recites at least the abstract idea identified above in claim 11.
Step 2A, prong 2
“re-using the previously used task specific encoder of the similar task when determining that the similar task is stored in the memory” (mere instructions to apply the judicial exception using a generic computer component. See MPEP 2106.05(f).)
The additional elements do not integrate into a practical application.
Step 2B
“re-using the previously used task specific encoder of the similar task when determining that the similar task is stored in the memory” (mere instructions to apply the judicial exception using a generic computer component. See MPEP 2106.05(f).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 13 recites: See the rejection of claim 3 above. Same rationale applies.
Claim 14 recites: See the rejection of claim 4 above. Same rationale applies.
Claim 15 recites:
Step 2A, prong 1
Claim 15 recites at least the abstract idea identified above in claim 14.
Step 2A, prong 2
“transmitting the classification head for the new task to the server to be stored in the memory” (Transmitting data to perform the abstract idea is considered “mere data gathering” and is an insignificant extra-solution activity, which does not integrate the abstract idea into a practical application. See MPEP 2106.05 (g).).)
The additional elements do not integrate into a practical application.
Step 2B
“transmitting the classification head for the new task to the server to be stored in the memory” (This step is directed to transmitting and receiving data, which is well-understood, routine, and conventional. i. Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); See MPEP 2106.05 (d) (II)).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 16 recites: See the rejection of claim 6 above. Same rationale applies.
Claim 17 recites: See the rejection of claim 7 above. Same rationale applies.
Claim 18 recites: See the rejection of claim 8 above. Same rationale applies.
Claim 19 recites:
Step 2A, prong 1
“wherein the test error is determined based on a score calculated in a predictor-label association analysis” (This step is a recitation of a mental process that is practical to perform in the human mind. A human can determine an error value based on a score in a label prediction analysis (i.e. observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
Step 2A, Prong 2 & 2B
This claim does not recite any additional elements.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-2 and 10-12 are rejected under 35 U.S.C. 103 as being unpatentable over Rostami et al. (US-20210019632-A1) in view of Ke et al. (“Continual Learning of a Mixed Sequence of Similar and Dissimilar Tasks”).
Regarding Claim 1,
Rostami (US 20210019632 A1) teaches a method of continual learning, comprising:
maintaining a memory comprising one or more previously learned tasks (para [0004] “Learning systems are networks or machines that attempt to learn based on memory of previous experiences.” para [0061] “The method enables an agent to remember previously learned tasks and easily adapt to learn new tasks without corrupting the knowledge of previously learned task.”);
determining, in response to receiving a new task, one or more of similarities and dissimilarities between at least one previously learned task and the new task (para [0061] “The method enables an agent to remember previously learned tasks and easily adapt to learn new tasks without corrupting the knowledge of previously learned task.” para [0068] “As such, the system operates based on training encoder ϕ.sub.v 304 such that all tasks share a similar distribution in the embedding, i.e., the new tasks are learned such that their distribution in the embedding match the past experience, captured in the shared distribution.” Similar distribution (i.e., similarities). para [0071] “Let {circumflex over (p)}.sub.J.sup.(0)(z) denote this parametric distribution. This distribution is updated after learning each task to incorporate what has been learned from the new task. As a result, this distribution captures knowledge about the past.” Para [0072] “where D(.Math.,.Math.) is a discrepancy measure, i.e., a metric, between two probability distributions and λ is a trade-off parameter. The first four terms in Eq. 2 are empirical classification risk and autoencoder reconstruction loss terms for the current task and the generated pseudo-dataset. The third and the fourth terms enforce learning (updating) the current task 400 such that the past learned knowledge is not forgotten. The fifth term is added to enforce the learned embedding distribution for the current task 400 to be similar to what has been learned in the past, i.e., task-invariant.” Discrepancy (i.e., dissimilarities) between distribution of prior tasks and current tasks is measured.);
generating, based on the one or more similarities and dissimilarities determined and a previously used task-specific encoder corresponding to the at least one previously learned task (para [0068] “As such, the system operates based on training encoder ϕ.sub.v 304 such that all tasks share a similar distribution in the embedding, i.e., the new tasks are learned such that their distribution in the embedding match the past experience, captured in the shared distribution.” Similarities between distributions between previous and new tasks. Para [0072] “where D(.Math.,.Math.) is a discrepancy measure, i.e., a metric, between two probability distributions and λ is a trade-off parameter. The first four terms in Eq. 2 are empirical classification risk and autoencoder reconstruction loss terms for the current task and the generated pseudo-dataset.” Discrepancy used to calculate loss (i.e., test error).), a test error value for classifying the new task (para [0070]-[0072] reconstruction loss (i.e., test error).); and
applying the previously used task-specific encoder to the new task based on the generated test error value (para [0072] “The first four terms in Eq. 2 are empirical classification risk and autoencoder reconstruction loss terms for the current task and the generated pseudo-dataset. The third and the fourth terms enforce learning (updating) the current task 400 such that the past learned knowledge is not forgotten. The fifth term is added to enforce the learned embedding distribution for the current task 400 to be similar to what has been learned in the past, i.e., task-invariant. Note that the distance between two distribution on classes was conditioned to avoid class matching challenge, i.e., when wrong classes across two tasks are matched in the embedding, as well as to prevent mode collapse from happening.” The encoder is updated based on the loss.),
…the one or more previously learned tasks being drawn from a same distribution… (para [0068] “As such, the system operates based on training encoder ϕ.sub.v 304 such that all tasks share a similar distribution in the embedding, i.e., the new tasks are learned such that their distribution in the embedding match the past experience, captured in the shared distribution.”).
Rostami does not explicitly disclose
wherein the dissimilarities are determined when at least one of data from the one or more previously learned tasks belonging to the same group of classes or predictors of the one or more previously learned tasks… is not determined.
However, Ke (“Continual Learning of a Mixed Sequence of Similar and Dissimilar Tasks”) teaches
determining, in response to receiving a new task, one or more of similarities and dissimilarities between at least one previously learned task and the new task (pg. 3, section 3.1; “Let the set of tasks learned so far be T (before learning a new task t). Let Tsim ⊆ T be a set of similar tasks to t and Tdis = T − Tsim be the set of dissimilar tasks to t. We will discuss how to compute Tsim in Section 3.3.”);
wherein the dissimilarities are determined when at least one of data from the one or more previously learned tasks belonging to the same group of classes or predictors of the one or more previously learned tasks… is not determined (pg. 6, section 3.3; “We define task similarity by determining whether there is a positive knowledge transfer from a previous task k to the current task t. A transfer model fk→t is used to transfer knowledge from task k to task t. A single task model f∅, called the reference model, is used to learn t independently. If the following statistical risk holds, which indicates a positive knowledge transfer, we say that task k is similar to task t; otherwise task k is dissimilar to task t.
PNG
media_image1.png
45
730
media_image1.png
Greyscale
We use a validation set to check whether Eq. 9 holds. Specifically, if the transfer model fk→t classifies the validation data of task t better than the reference model f∅, then we say k contains shareable prior knowledge that can help t learn a better model than without the knowledge, f∅, indicating positive knowledge transfer. We set TSV (t)[k] = 1 indicating that k is similar to t; otherwise TSV (t)[k] = 0 indicating that k is dissimilar to t.” It is not determined that the task belongs to the same group of classes or predictors when TSV (t)[k] = 0.).
Rostami and Ke are analogous because they are directed to continual learning of a sequence of tasks focused on dealing with catastrophic forgetting.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the CL model of Rostami with the method of determining dissimilar tasks of Ke.
Doing so would allow for learning a mixed sequence of similar and dissimilar tasks learned from the previous tasks to protect the knowledge learned for those dissimilar tasks so that their important parameters are not affected (Ke pg. 2;)
Regarding Claim 2,
Rostami and Ke teach the method of claim 1. Rostami further teaches further comprising:
determining whether a similar task to the new task is stored in the memory (para [0064] “The agent learns a new task at each time step and proceeds to learn the next task. Each task is learned based upon the experiences, gained from learning past tasks.”),
wherein the previously used task specific encoder is applied to the new task in response to determining that the similar task is stored in the memory (para [0068] “As such, the system operates based on training encoder ϕ.sub.v 304 such that all tasks share a similar distribution in the embedding, i.e., the new tasks are learned such that their distribution in the embedding match the past experience, captured in the shared distribution.”).
Regarding Claim 10,
Claim 10 is the user equipment corresponding to the method of claim 1. Claim 10 is substantially similar to claim 1 and is rejected on the same grounds.
Regarding Claim 11,
Claim 11 is the user equipment corresponding to the method of claim 2. Claim 11 is substantially similar to claim 2 and is rejected on the same grounds.
Regarding Claim 12,
Rostami and Ke teach the UE of claim 11. Rostami further teaches wherein the processor further performs the method by: re-using the previously used task specific encoder of the similar task when determining that the similar task is stored in the memory (para [0068] “As such, the system operates based on training encoder ϕ.sub.v 304 such that all tasks share a similar distribution in the embedding, i.e., the new tasks are learned such that their distribution in the embedding match the past experience, captured in the shared distribution.”).
Claims 3-7, 13-14, and 16-17 are rejected under 35 U.S.C. 103 as being unpatentable over Rostami et al. (US-20210019632-A1) in view of Ke et al. (“Continual Learning of a Mixed Sequence of Similar and Dissimilar Tasks”), Lee et al. (“A NEURAL DIRICHLET PROCESS MIXTURE MODEL FOR TASK-FREE CONTINUAL LEARNING”) and Ghesu et al. (US-20230154164-A1).
Regarding Claim 3,
Rostami and Ke teach the method of claim 2.
Rostami and Ke do not explicitly disclose
further comprising: training a new task specific encoder when determining that no similar task to the new task is stored in the memory, and storing the new task specific encoder in the memory.
However, Lee (“A NEURAL DIRICHLET PROCESS MIXTURE MODEL FOR TASK-FREE CONTINUAL LEARNING”) teaches
further comprising: training a new task specific encoder when determining that no similar task to the new task is stored in the memory (pg. 2, section 1; “We are one of the first to propose an expansion-based approach for task-free CL. Hence, our model not only prevents catastrophic forgetting but also applies to the setting where no task definition and boundaries are given at both training and test time.” No previous tasks. Pg. 5, section 3.2; “We assume that samples sequentially arrive one at a time during training. For a new sample, we first decide whether the sample should be assigned to an existing expert or a new expert should be created for it.”), and
Rostami, Ke, and Lee are analogous because they are both directed to the field of Continual Learning.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the machine learning model of Rostami and Ke with the task-free continual learning of Lee.
Doing so would allow for training a model without needing previous tasks (Lee pg. 2, section 2; “All the works mentioned above heavily rely on explicit task definition. However, in real-world scenarios, task definition is rarely given at training time.”)
Ghesu (US 20230154164 A1) teaches
storing the new task specific encoder in the memory (para [0037] “For example, the optimized encoder network can be output by storing the optimized encoder network on a memory or storage of a computer system, or by transmitting the optimized encoder network to a remote computer system.”).
Rostami, Ke, and Ghesu are analogous because they are both directed to the field of Continual Learning.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the continual learning model of Rostami and Ke with the method of storing encoders of Ghesu.
Doing so would allow for transmitting the encoder to requesting client devices (Ghesu para [0037]).
Regarding Claim 4,
Rostami, Ke, Lee, and Ghesu teach the method of claim 3. Rostami further teaches further comprising: using an encoder portion of the task specific encoder as a feature extraction backbone (para [0060] “The encoder 304 maps the data points associated with the task into an embedding space 306, which describes the input in terms of abstract discriminative features (or classes).”) to train only a classifier head for the new task (para [0060] “Thus, the classifier network receives inputs from the embedding space 306 and then, at the output of the embedding space 306, predicts the label for the given input which has passed through the encoder 304 to the embedding space 306.”).
Regarding Claim 5,
Rostami, Ke, Lee, and Ghesu teach the method of claim 4. Rostami further teaches further comprising: storing the classification head for the new task in the memory (para [0057] “In addition, an aspect provides that one or more aspects of the present technology are implemented by utilizing one or more distributed computing environments, such as where tasks are performed by remote processing devices that are linked through a communications network, or such as where various program modules are located in both local and remote computer-storage media including memory-storage devices.” para [0060] “The encoder 304 then learns (updates) a new task conditioned on matching the embedded distribution. The learned new knowledge can then be used to update the embedding distribution in the embedding space 306. Thus, the classifier network receives inputs from the embedding space 306 and then, at the output of the embedding space 306, predicts the label for the given input which has passed through the encoder 304 to the embedding space 306.”).
Regarding Claim 6,
Rostami, Ke, Lee, and Ghesu teach the method of claim 4. Rostami further teaches wherein the encoder portion of the task specific encoder is used to train only the classifier head for the new task based on the re-used task specific encoder or the trained new task specific encoder (para [0060] “The encoder 304 then learns (updates) a new task conditioned on matching the embedded distribution. The learned new knowledge can then be used to update the embedding distribution in the embedding space 306. Thus, the classifier network receives inputs from the embedding space 306 and then, at the output of the embedding space 306, predicts the label for the given input which has passed through the encoder 304 to the embedding space 306.”).
Regarding Claim 7,
Rostami, Ke, Lee, and Ghesu teach the method of claim 3.
Lee further teaches
wherein a style modulation technique is used to train the new task specific encoder when determining that no similar task to the new task is stored in the memory (pg. 5, section 3.2; “Training. We assume that samples sequentially arrive one at a time during training. For a new sample, we first decide whether the sample should be assigned to an existing expert or a new expert should be created for it. Suppose that samples up to (xn, yn) are sequentially processed and K experts are already created when a new sample (xn+1, yn+1) arrives.” The style of training (e.g., creating new expert or using old expert) is changed (i.e., modulated) based on the new training sample.).
Rostami, Ke, Ghesu, and Lee are analogous because they are both directed to the field of Continual Learning.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the machine learning model of Rostami, Ke, and Ghesu with the task-free continual learning of Lee.
Doing so would allow for training a model without needing previous tasks (Lee pg. 2, section 2; “All the works mentioned above heavily rely on explicit task definition. However, in real-world scenarios, task definition is rarely given at training time.”)
Regarding Claim 13,
Claim 13 is the user equipment corresponding to the method of claim 3. Claim 13 is substantially similar to claim 3 and is rejected on the same grounds.
Regarding Claim 14,
Claim 14 is the user equipment corresponding to the method of claim 4. Claim 14 is substantially similar to claim 4 and is rejected on the same grounds.
Regarding Claim 16,
Claim 16 is the user equipment corresponding to the method of claim 6. Claim 16 is substantially similar to claim 6 and is rejected on the same grounds.
Regarding Claim 17,
Claim 17 is the user equipment corresponding to the method of claim 7. Claim 17 is substantially similar to claim 7 and is rejected on the same grounds.
Claims 8-9 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Rostami et al. (US-20210019632-A1) in view of Ke et al. (“Continual Learning of a Mixed Sequence of Similar and Dissimilar Tasks”), Lee et al. (“A NEURAL DIRICHLET PROCESS MIXTURE MODEL FOR TASK-FREE CONTINUAL LEARNING”), Ghesu et al. (US-20230154164-A1), and Nadamuni et al. (US-20200302339-A1).
Regarding Claim 8,
Rostami, Ke, Lee, and Ghesu teach the method of claim 3.
Rostami, Ke, Lee, and Ghesu do not explicitly disclose
wherein the similarity is determined based on a score of a training memory for a task, the score being calculated by a distribution consistency estimator.
However, Nadamuni (US 20200302339 A1) teaches
wherein the similarity is determined based on a score of a training memory for a task, the score being calculated by a distribution consistency estimator (para [0032] trained memory. para [0046] “Thus, as a function of an embedding of a VAE corresponding to the archetype task 106 and a skill 116 assigned to the archetype task 106, generative memory 104 may use the VAE to determine a similarity score between a task for input data 120 and the archetype task 106.” Para [0077] “Thus, generative memory 104 may use the VAE to determine a similarity between a task for input data 120 and the cluster of tasks 108 represented by the archetype task 106. For example, generative memory 104 may sample a τ distribution defined by).
Rostami, Ke, Lee, Ghesu, and Nadamuni are analogous because they are directed towards variational autoencoders.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the machine learning model of Rostami with the similarity task scoring of Nadamuni.
Doing so would allow for only recalling relevant data to improve the performance and learning of the machine learning model (Nadamuni para [0088]).
Regarding Claim 9,
Rostami, Ke, Lee, Ghesu, and Nadamuni teach the method of claim 8. Nadamuni further teaches wherein the similarity is determined based on a score calculated in a predictor-label association analysis (para [0048] “Machine learning model 112 applies the one or more skills 116 to the plurality of inputs to obtain one or more output labels 122 for the plurality of inputs of input data 120. In some examples, output labels 122 comprise one or more action sequences for solving a task defined by input data 120. In some examples, the task defined by input data 120 is a new task not previously learned by machine learning model 112. In this fashion, machine learning model 112 is capable of scalable learning to obtain labels for new tasks for which machine learning model 112 has not previously been trained.”).
Rostami, Ke, Lee, Ghesu, and Nadamuni are analogous because they are directed towards variational autoencoders.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the machine learning model of Rostami with the similarity task scoring of Nadamuni.
Doing so would allow for only recalling relevant data to improve the performance and learning of the machine learning model (Nadamuni para [0088]).
Regarding Claim 18,
Claim 18 is the user equipment corresponding to the method of claim 8. Claim 18 is substantially similar to claim 8 and is rejected on the same grounds.
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Rostami et al. (US-20210019632-A1) in view of Ke et al. (“Continual Learning of a Mixed Sequence of Similar and Dissimilar Tasks”), Lee et al. (“A NEURAL DIRICHLET PROCESS MIXTURE MODEL FOR TASK-FREE CONTINUAL LEARNING”), Ghesu et al. (US-20230154164-A1), and Xu et al. (US-20200182618-A1).
Regarding Claim 15,
Rostami, Ke, Lee, and Ghesu teach the UE of claim 14.
Rostami, Ke, Lee, and Ghesu do not explicitly disclose
wherein the processor further performs the method by: transmitting the classification head for the new task to the server to be stored in the memory.
However, Xu (US 20200182618 A1) teaches
wherein the processor further performs the method by: transmitting the classification head for the new task to the server to be stored in the memory (para [0258] “In some embodiments, the training of the heading classifier may be performed by the server 1410 (or the processing engine 1412) and the provider terminal 1440 may retrieve the trained classifier from the server 1410 (or the storage device 1450 if the server 1410 transmits the trained classifier to the storage device 1450 for storage).”).
Rostami, Ke, Lee, Ghesu, and Xu are analogous because they are directed towards machine learning.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the machine learning model of Rostami with the method of transmitting a classifier of Xu.
Doing so would allow for retrieving a classifier from a server based on identity information (Xu para [0258]).
Claim 19 is rejected under 35 U.S.C. 103 as being unpatentable over Rostami et al. (US-20210019632-A1) in view of Ke et al. (“Continual Learning of a Mixed Sequence of Similar and Dissimilar Tasks”), Lee et al. (“A NEURAL DIRICHLET PROCESS MIXTURE MODEL FOR TASK-FREE CONTINUAL LEARNING”), Ghesu et al. (US-20230154164-A1), Nadamuni et al. (US-20200302339-A1), and Manchanda et al. (US-20230186361-A1).
Regarding Claim 19,
Rostami, Ke, Lee, Ghesu, and Nadamuni teach the UE of claim 18.
Rostami, Ke, Lee, and Ghesu do not explicitly disclose
wherein the test error is determined based on a score calculated in a predictor-label association analysis.
However, Manchanda (US 20230186361 A1) teaches
wherein the test error is determined based on a score calculated in a predictor-label association analysis (para [0060] “In some embodiments, to train the suggestion models based on a suggestion example, the suggestion scoring module 420 applies a suggestion relevance loss function that compares suggestion score generated by the suggestion scoring module 420 with a label assigned to the suggestion example.” Loss (i.e., test error).).
Rostami, Ke, Lee, Ghesu, Nadamuni, and Manchanda are analogous because they are directed towards machine learning.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the machine learning model of Rostami, Ke, Lee, Ghesu, and Nadamuni with the loss function of Manchanda.
Doing so would allow for assigning weight to the loss functions to balance how impactful each loss function is in training the model (Manchanda para [0070]).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 HENRY K NGUYEN whose telephone number is (571)272-0217. The examiner can normally be reached Mon - Fri 7:00am-4:30pm.
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, Li B Zhen can be reached at 5712723768. 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.
/HENRY NGUYEN/Examiner, Art Unit 2121
Prosecution Insights