DETAILED ACTION
This action is in response to the amendment filed 02/18/2026. Claims 1-20 are pending and have been examined.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Claim 1:
Subject Matter Eligibility Analysis Step 1:
Claim 1 recites a method and is thus a process, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 1 recites
during each of one or more training iterations, for each respective training sample in the set of training samples: selecting the respective task-specific prompt representative of the respective task of the respective training sample (This limitation is a mental process as it encompasses a human mentally selecting a prompt.)
Therefore, claim 1 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 1 further recites additional elements of
A computer-implemented method when executed by data processing hardware causes the data processing hardware to perform operations comprising: (This element does not integrate the abstract idea into a practical application because it amounts to mere “apply it on a computer” (see MPEP 2106.05(f)).)
obtaining a set of training samples, each training sample in the set of training samples associated with a respective task of a plurality of different tasks (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
obtaining a task-invariant prompt representative of learned knowledge common to each respective task of the plurality of different tasks (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
for each respective task of the plurality of different tasks, obtaining a respective task-specific prompt representative of learned knowledge specific to the respective task (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
training a model by attaching the task-invariant prompt at a first layer of the model and attaching the selected respective task-specific prompt at a second layer of the model (This element does not integrate the abstract idea into a practical application because it amounts to mere “apply it on a computer” (see MPEP 2106.05(f)).)
Therefore, claim 1 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 1 do not provide significantly more than the abstract idea itself, taken alone and in combination because
A computer-implemented method when executed by data processing hardware causes the data processing hardware to perform operations comprising uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
obtaining a set of training samples, each training sample in the set of training samples associated with a respective task of a plurality of different tasks is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
obtaining a task-invariant prompt representative of learned knowledge common to each respective task of the plurality of different tasks is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
for each respective task of the plurality of different tasks, obtaining a respective task-specific prompt representative of learned knowledge specific to the respective task is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
training a model by attaching the task-invariant prompt at a first layer of the model and attaching the selected respective task-specific prompt at a second layer of the model uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
Therefore, claim 1 is subject-matter ineligible.
Regarding Claim 2:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 2 recites the same abstract ideas as claim 1. Therefore, claim 2 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 2 further recites additional elements of
wherein each respective training sample comprises an image. (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
Therefore, claim 2 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 2 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein each respective training sample comprises an image is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
Therefore, claim 2 is subject-matter ineligible.
Regarding Claim 3:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 3 recites the same abstract ideas as claim 1. Therefore, claim 3 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 3 further recites additional elements of
wherein training the model comprises updating a pre-trained model with the task-invariant prompt and the selected respective task-specific prompt. (This element does not integrate the abstract idea into a practical application because it amounts to mere “apply it on a computer” (see MPEP 2106.05(f)).)
Therefore, claim 3 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 3 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein training the model comprises updating a pre-trained model with the task-invariant prompt and the selected respective task-specific prompt uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
Therefore, claim 3 is subject-matter ineligible.
Regarding Claim 4:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 4 recites the same abstract ideas as claim 3. Therefore, claim 4 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 4 further recites additional elements of
updating the pre-trained model with the task invariant prompt and the selected respective task-specific prompt (This element does not integrate the abstract idea into a practical application because it amounts to mere “apply it on a computer” (see MPEP 2106.05(f)).)
inserting the task-invariant prompt at the first layer of the pre-trained model (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
inserting the respective task-specific prompt at the second layer of the pre-trained model. (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
Therefore, claim 4 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 4 do not provide significantly more than the abstract idea itself, taken alone and in combination because
updating the pre-trained model with the task invariant prompt and the selected respective task-specific prompt uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
inserting the task-invariant prompt at a first layer of the pre-trained model is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
inserting the respective task-specific prompt at a second layer of the pre-trained model is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
Therefore, claim 4 is subject-matter ineligible.
Regarding Claim 5:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 5 recites the same abstract ideas as claim 4. Therefore, claim 5 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 5 further recites additional elements of
wherein the first layer and the second layer are each a self-attention layer. (This element does not integrate the abstract idea into a practical application because it amounts to mere “apply it on a computer” (see MPEP 2106.05(f)).)
Therefore, claim 5 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 5 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein the first layer and the second layer are each a self-attention layer uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
Therefore, claim 5 is subject-matter ineligible.
Regarding Claim 6:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 6 recites the same abstract ideas as claim 4. Therefore, claim 6 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 6 further recites additional elements of
wherein inserting the task-invariant prompt at the first layer of the pre-trained model comprises prepending the task-invariant prompt to an input embedding feature of the first layer. (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
Therefore, claim 6 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 6 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein inserting the task-invariant prompt at the first layer of the pre-trained model comprises prepending the task-invariant prompt to an input embedding feature of the first layer is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
Therefore, claim 6 is subject-matter ineligible.
Regarding Claim 7:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 7 recites the same abstract ideas as claim 4. Therefore, claim 7 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 7 further recites additional elements of
wherein inserting the respective task-specific prompt at the second layer of the pre-trained model comprises prepending the respective task-specific prompt to an input embedding feature of the second layer. (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
Therefore, claim 7 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 7 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein inserting the respective task-specific prompt at the second layer of the pre-trained model comprises prepending the respective task-specific prompt to an input embedding feature of the second layer is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
Therefore, claim 7 is subject-matter ineligible.
Regarding Claim 8:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 8 recites the same abstract ideas as claim 1. Therefore, claim 8 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 8 further recites additional elements of
wherein each respective task-specific prompt is associated with task-specific key representative of one or more features of the respective task. (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
Therefore, claim 8 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 8 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein each respective task-specific prompt is associated with task-specific key representative of one or more features of the respective task is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
Therefore, claim 8 is subject-matter ineligible
Regarding Claim 9:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 9 recites
determining a cross-entropy loss (This limitation is a mental process as it encompasses a human mentally determining a cross-entropy loss.)
Therefore, claim 9 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 9 further recites additional elements of
wherein training the model using the task-invariant prompt and the selected respective task-specific prompt comprises determining a cross entropy
loss. (This element does not integrate the abstract idea into a practical application because it amounts to mere “apply it on a computer” (see MPEP 2106.05(f)).)
Therefore, claim 9 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 9 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein training the model using the task-invariant prompt and the selected respective task-specific prompt comprises determining a cross entropy loss uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
Therefore, claim 9 is subject-matter ineligible.
Regarding Claim 10:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 10 recites the same abstract ideas as claim 1. Therefore, claim 10 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 10 further recites additional elements of
wherein each respective task of the plurality of different tasks comprises image classification. (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
Therefore, claim 10 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 10 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein each respective task of the plurality of different tasks comprises image classification is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
Therefore, claim 10 is subject-matter ineligible.
Regarding Claim 11:
Subject Matter Eligibility Analysis Step 1:
Claim 11 recites a method and is thus a process, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 11 recites
during each of one or more training iterations, for each respective training sample in the set of training samples: selecting the respective task-specific prompt representative of the respective task of the respective training sample (This limitation is a mental process as it encompasses a human mentally selecting a prompt.)
Therefore, claim 11 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 11 further recites additional elements of
A system comprising: data processing hardware; and memory hardware in communication with the data processing hardware, the memory hardware storing instructions that when executed on the data processing hardware cause the data processing hardware to perform operations comprising: (This element does not integrate the abstract idea into a practical application because it amounts to mere “apply it on a computer” (see MPEP 2106.05(f)).)
obtaining a set of training samples, each training sample in the set of training samples associated with a respective task of a plurality of different tasks (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
obtaining a task-invariant prompt representative of learned knowledge common to each respective task of the plurality of different tasks (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
for each respective task of the plurality of different tasks, obtaining a respective task-specific prompt representative of learned knowledge specific to the respective task (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
training a model by attaching the task-invariant prompt at a first layer of the model and attaching the selected respective task-specific prompt at a second layer of the model (This element does not integrate the abstract idea into a practical application because it amounts to mere “apply it on a computer” (see MPEP 2106.05(f)).)
Therefore, claim 11 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 11 do not provide significantly more than the abstract idea itself, taken alone and in combination because
A system comprising: data processing hardware; and memory hardware in communication with the data processing hardware, the memory hardware storing instructions that when executed on the data processing hardware cause the data processing hardware to perform operations comprising uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
obtaining a set of training samples, each training sample in the set of training samples associated with a respective task of a plurality of different tasks is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
obtaining a task-invariant prompt representative of learned knowledge common to each respective task of the plurality of different tasks is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
for each respective task of the plurality of different tasks, obtaining a respective task-specific prompt representative of learned knowledge specific to the respective task is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
training a model using the task-invariant prompt and the selected respective task-specific prompt uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
Therefore, claim 11 is subject-matter ineligible.
Regarding claim 12, claim 12 recites substantially similar limitations to claim 2, and is therefore rejected under the same analysis.
Regarding claim 13, claim 13 recites substantially similar limitations to claim 3, and is therefore rejected under the same analysis.
Regarding claim 14, claim 14 recites substantially similar limitations to claim 4, and is therefore rejected under the same analysis.
Regarding claim 15, claim 15 recites substantially similar limitations to claim 5, and is therefore rejected under the same analysis.
Regarding claim 16, claim 16 recites substantially similar limitations to claim 6, and is therefore rejected under the same analysis.
Regarding claim 17, claim 17 recites substantially similar limitations to claim 7, and is therefore rejected under the same analysis.
Regarding claim 18, claim 18 recites substantially similar limitations to claim 8, and is therefore rejected under the same analysis.
Regarding claim 19, claim 19 recites substantially similar limitations to claim 9, and is therefore rejected under the same analysis.
Regarding claim 20, claim 20 recites substantially similar limitations to claim 20, and is therefore rejected under the same analysis.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claim(s) 1-4, 8-14, 18-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Pham et al. (“DualNet: Continual Learning, Fast and Slow”) (hereafter referred to as Pham) in view of Butvinik et al. (US 2022/0261633 A1) (hereafter referred to as Butvinik).
Regarding claim 1, Pham teaches
obtaining a set of training samples, each training sample in the set of training samples associated with a respective task of a plurality of different tasks (Pham, page 3, 1st paragraph, “We consider the online continual learning setting [Lopez-Paz and Ranzato, 2017, Chaudhry et al., 2019a] over a continuum of data D = {xi, ti, yi}I, where each instance is a labeled sample {xi, yi} with an optional task identifier ti. Each labeled data sample is drawn from an underlying distribution Pt(X, Y) that represents a task and can suddenly change to Pt+1, indicating a task switch.” Examiner notes that the data samples are the training samples.);
obtaining a task-invariant prompt representative of learned knowledge common to each respective task of the plurality of different tasks (Pham, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample” where “(i) the slow learner is responsible for learning a general, task agnostic representation; and (ii) the fast learner learns with labeled data from the continuum to quickly capture the new information and then consolidate the knowledge to the slow learner” (Pham, page 3, 2nd paragraph). Examiner notes that the task-invariant prompt is the query from the fast learner to the slow learner);
for each respective task of the plurality of different tasks, obtaining a respective task-specific prompt representative of learned knowledge specific to the respective task (Pham, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample. The incurred loss will be backpropagated into both learners for supervised knowledge consolidation. Additionally, the fast learner’s adaptation is per-sample based and do not require additional information such as the task identifiers” where “(i) the slow learner is responsible for learning a general, task agnostic representation; and (ii) the fast learner learns with labeled data from the continuum to quickly capture the new information and then consolidate the knowledge to the slow learner” (Pham, page 3, 2nd paragraph). Examiner notes that the task-specific prompt is the arrival of the labeled sample to the fast learner to learn the sample.);
and during each of one or more training iterations, for each respective training sample in the set of training samples: selecting the respective task-specific prompt representative of the respective task of the respective training sample (Pham, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample. The incurred loss will be backpropagated into both learners for supervised knowledge consolidation. Additionally, the fast learner’s adaptation is per-sample based and do not require additional information such as the task identifiers” where “(i) the slow learner is responsible for learning a general, task agnostic representation; and (ii) the fast learner learns with labeled data from the continuum to quickly capture the new information and then consolidate the knowledge to the slow learner” (Pham, page 3, 2nd paragraph) and “We simulate the synchronous training property in DualNet by training the slow learner with n iterations using the episodic memory data before observing a mini-batch of labeled data”(Pham, page 7, 3rd paragraph). Examiner notes that the task-specific prompt is the arrival of the labeled sample to the fast learner to learn the sample and the labeled sample is selected by learning on the sample.);
training a model by attaching the task-invariant prompt at a first layer of the model (Pham, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample” where “(i) the slow learner is responsible for learning a general, task agnostic representation; and (ii) the fast learner learns with labeled data from the continuum to quickly capture the new information and then consolidate the knowledge to the slow learner” (Pham, page 3, 2nd paragraph) and Pham, page 2, Figure 1
PNG
media_image1.png
433
945
media_image1.png
Greyscale
Examiner notes that the task-invariant prompt is the query from the fast learner to the slow learner. Examiner further notes that the query is attached or joined to the slow net which is the first layer.
and attaching the selected respective task-specific prompt at a second layer of the model (Pham, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample. The incurred loss will be backpropagated into both learners for supervised knowledge consolidation. Additionally, the fast learner’s adaptation is per-sample based and do not require additional information such as the task identifiers” and Pham, page 4, Figure 2,
PNG
media_image2.png
395
926
media_image2.png
Greyscale
Examiner notes that the task-specific prompt is the arrival of the labeled sample to the fast learner to learn the sample, and the labeled sample is selected by learning on the sample. Examiner further notes that the image sample is input into a convolutional layer as shown in Figure 2, and thus the task-specific prompt is attached at a second layer.)
Pham does not explicitly disclose, but Butvinik does disclose
A computer-implemented method when executed by data processing hardware causes the data processing hardware to perform operations comprising (Butvinik, page 23, paragraph 0153, “Embodiments of the invention may include one or more article(s) (e.g. memory 120 or storage 130) such as a computer or processor non-transitory readable medium, or a computer or processor non-transitory storage medium, such as for example a memory, a disk drive, or a USB flash memory, encoding, including or storing instructions, e.g., computer-executable instructions, which, when executed by a processor or controller, carry out methods disclosed therein.”)
Pham and Butvinik are analogous to the claimed invention because they both use continual learning to train a model. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date to have implemented Pham on data processing hardware. Thus, this would be applying a known technique (continual learning) to a known device (data processing hardware) ready for improvement to yield predictable results (train a model) (MPEP 2143 I. (C) Use of known technique to improve similar devices (methods, or products) in the same way).
Regarding claim 2, Pham in view of Butvinik teaches
The method of claim 1, wherein each respective training sample comprises an image (Pham, page 3, last paragraph, “Therefore, we consider Barlow Twins [Zbontar et al., 2021], a recent state-of-the-art SSL method that achieved promising results with minimal computational overheads. Formally, Barlow Twins requires two views MA and MB by applying two different data transformation to a batch of images M sampled from the memory.”).
Regarding claim 3, Pham in view of Butvinik teaches
The method of claim 1, wherein training the model comprises updating a pre- trained model with the task-invariant prompt and the selected respective task-specific prompt (Pham, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample. The incurred loss will be backpropagated into both learners for supervised knowledge consolidation. Additionally, the fast learner’s adaptation is per-sample based and do not require additional information such as the task identifiers” where “(i) the slow learner is responsible for learning a general, task agnostic representation; and (ii) the fast learner learns with labeled data from the continuum to quickly capture the new information and then consolidate the knowledge to the slow learner” (Pham, page 3, 2nd paragraph) and “We simulate the synchronous training property in DualNet by training the slow learner with n iterations using the episodic memory data before observing a mini-batch of labeled data”(Pham, page 7, 3rd paragraph). Examiner notes that the task-specific prompt is the arrival of the labeled sample to the fast learner to learn the sample and the labeled sample is selected by learning on the sample. Examiner further notes that since the DualNet model is trained in iterations, the model comprises updating a pre-trained model.).
Regarding claim 4, Pham in view of Butvinik teaches
The method of claim 3, wherein updating the pre-trained model with the task- invariant prompt and the selected respective task-specific prompt comprises: inserting the task-invariant prompt at a first layer of the pre-trained model (Pham, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample” where “(i) the slow learner is responsible for learning a general, task agnostic representation; and (ii) the fast learner learns with labeled data from the continuum to quickly capture the new information and then consolidate the knowledge to the slow learner” (Pham, page 3, 2nd paragraph) and Pham, page 2, Figure 1
PNG
media_image1.png
433
945
media_image1.png
Greyscale
Examiner notes that the task-invariant prompt is the query from the fast learner to the slow learner. Examiner further notes that the query is inserted into the slow net which is the first layer.);
and inserting the respective task-specific prompt at a second layer of the pre-trained model (Pham, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample. The incurred loss will be backpropagated into both learners for supervised knowledge consolidation. Additionally, the fast learner’s adaptation is per-sample based and do not require additional information such as the task identifiers” and Pham, page 4, Figure 2,
PNG
media_image2.png
395
926
media_image2.png
Greyscale
Examiner notes that the task-specific prompt is the arrival of the labeled sample to the fast learner to learn the sample and the labeled sample is selected by learning on the sample. Examiner further notes that the image sample is input into a convolutional layer as shown in Figure 2.).
Regarding claim 8, Pham in view of Butvinik teaches
The method of claim 1, wherein each respective task-specific prompt is associated with task-specific key representative of one or more features of the respective task (Pham, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample. The incurred loss will be backpropagated into both learners for supervised knowledge consolidation. Additionally, the fast learner’s adaptation is per-sample based and do not require additional information such as the task identifiers” where “We consider the online continual learning setting [Lopez-Paz and Ranzato, 2017, Chaudhry et al., 2019a] over a continuum of data D = {xi, ti, yi}I, where each instance is a labeled sample {xi, yi} with an optional task identifier ti” and“(i) the slow learner is responsible for learning a general, task agnostic representation; and (ii) the fast learner learns with labeled data from the continuum to quickly capture the new information and then consolidate the knowledge to the slow learner” (Pham, page 3, 2nd paragraph). Examiner notes that the task-specific prompt is the arrival of the labeled sample to the fast learner to learn the sample. Examiner further notes that the key is the identifier.).
Regarding claim 9, Pham in view of Butvinik teaches
The method of claim 1, wherein training the model using the task-invariant prompt and the selected respective task-specific prompt comprises determining a cross- entropy loss (Pham, page 5, 2nd paragraph, “To further facilitate the fast learner’s knowledge acquisition during supervised learning, we also mix the current sample with previous data in the episodic memory, which is a form of experience replay (ER). Particularly, given the incoming labeled sample {x,y} and a mini batch of memory data M, we consider the ER with a soft label loss [van de Ven and Tolias, 2018] for the supervised learning phase as:
PNG
media_image3.png
83
844
media_image3.png
Greyscale
where CE is the cross-entropy loss.”).
Regarding claim 10, Pham in view of Butvinik teaches
The method of claim 1, wherein each respective task of the plurality of different tasks comprises image classification (Pham, page 3, last paragraph, “Therefore, we consider Barlow Twins [Zbontar et al., 2021], a recent state-of-the-art SSL method that achieved promising results with minimal computational overheads. Formally, Barlow Twins requires two views MA and MB by applying two different data transformation to a batch of images M sampled from the memory” where “We consider the online continual learning setting [Lopez-Paz and Ranzato, 2017, Chaudhry et al., 2019a] over a continuum of data D = {xi, ti, yi}I, where each instance is a labeled sample {xi, yi} with an optional task identifier ti. Each labeled data sample is drawn from an underlying distribution Pt(X, Y) that represents a task and can suddenly change to Pt+1, indicating a task switch When the task identifier t is given as an input, the setting follows the multi-head evaluation where only the corresponding classifier is selected to make a prediction” (Pham, page 3, 1st paragraph). Examiner notes that the data are images and the images are fed into a classifier for image classification. ).
Regarding claim 11, Pham teaches
obtaining a set of training samples, each training sample in the set of training samples associated with a respective task of a plurality of different tasks (Pham, page 3, 1st paragraph, “We consider the online continual learning setting [Lopez-Paz and Ranzato, 2017, Chaudhry et al., 2019a] over a continuum of data D = {xi, ti, yi}I, where each instance is a labeled sample {xi, yi} with an optional task identifier ti. Each labeled data sample is drawn from an underlying distribution Pt(X, Y) that represents a task and can suddenly change to Pt+1, indicating a task switch.” Examiner notes that the data samples are the training samples.);
obtaining a task-invariant prompt representative of learned knowledge common to each respective task of the plurality of different tasks (Pham, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample” where “(i) the slow learner is responsible for learning a general, task agnostic representation; and (ii) the fast learner learns with labeled data from the continuum to quickly capture the new information and then consolidate the knowledge to the slow learner” (Pham, page 3, 2nd paragraph). Examiner notes that the task-invariant prompt is the query from the fast learner to the slow learner);
for each respective task of the plurality of different tasks, obtaining a respective task-specific prompt representative of learned knowledge specific to the respective task (Pham, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample. The incurred loss will be backpropagated into both learners for supervised knowledge consolidation. Additionally, the fast learner’s adaptation is per-sample based and do not require additional information such as the task identifiers” where “(i) the slow learner is responsible for learning a general, task agnostic representation; and (ii) the fast learner learns with labeled data from the continuum to quickly capture the new information and then consolidate the knowledge to the slow learner” (Pham, page 3, 2nd paragraph). Examiner notes that the task-specific prompt is the arrival of the labeled sample to the fast learner to learn the sample.);
and during each of one or more training iterations, for each respective training sample in the set of training samples: selecting the respective task-specific prompt representative of the respective task of the respective training sample (Pham, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample. The incurred loss will be backpropagated into both learners for supervised knowledge consolidation. Additionally, the fast learner’s adaptation is per-sample based and do not require additional information such as the task identifiers” where “(i) the slow learner is responsible for learning a general, task agnostic representation; and (ii) the fast learner learns with labeled data from the continuum to quickly capture the new information and then consolidate the knowledge to the slow learner” (Pham, page 3, 2nd paragraph) and “We simulate the synchronous training property in DualNet by training the slow learner with n iterations using the episodic memory data before observing a mini-batch of labeled data”(Pham, page 7, 3rd paragraph). Examiner notes that the task-specific prompt is the arrival of the labeled sample to the fast learner to learn the sample and the labeled sample is selected by learning on the sample.);
and training a model using the task-invariant prompt and the selected respective task-specific prompt (Pham, page 3, 2nd paragraph “DualNet learns the data representation that is independent of the task’s label which allows for a better generalization capabilities across the task in the continual learning scenario. The model consists two main learning modules (Figure 1): (i) the slow learner is responsible for learning a general, task agnostic representation; and (ii) the fast learner learns with labeled data from the continuum to quickly capture the new information and then consolidate the knowledge to the slow learner.” Examiner notes that the task invariant prompt is given to the slow learner and the task specific prompt is given to the fast learner. ).
Pham does not explicitly disclose, but Butvinik does disclose
A system comprising: data processing hardware; and memory hardware in communication with the data processing hardware, the memory hardware storing instructions that when executed on the data processing hardware cause the data processing hardware to perform operations comprising (Butvinik, page 23, paragraph 0153, “Embodiments of the invention may include one or more article(s) (e.g. memory 120 or storage 130) such as a computer or processor non-transitory readable medium, or a computer or processor non-transitory storage medium, such as for example a memory, a disk drive, or a USB flash memory, encoding, including or storing instructions, e.g., computer-executable instructions, which, when executed by a processor or controller, carry out methods disclosed therein.”)
Pham and Butvinik are analogous to the claimed invention because they both use continual learning to train a model. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date to have implemented Pham on data processing hardware. Thus, this would be applying a known technique (continual learning) to a known device (data processing hardware) ready for improvement to yield predictable results (train a model) (MPEP 2143 I. (C) Use of known technique to improve similar devices (methods, or products) in the same way).
Regarding claim 12, claim 12 recites substantially similar limitations to claim 2, and is therefore rejected under the same analysis.
Regarding claim 13, claim 13 recites substantially similar limitations to claim 3, and is therefore rejected under the same analysis.
Regarding claim 14, claim 14 recites substantially similar limitations to claim 4, and is therefore rejected under the same analysis.
Regarding claim 18, claim 18 recites substantially similar limitations to claim 8, and is therefore rejected under the same analysis.
Regarding claim 19, claim 19 recites substantially similar limitations to claim 9, and is therefore rejected under the same analysis.
Regarding claim 20, claim 20 recites substantially similar limitations to claim 20, and is therefore rejected under the same analysis.
Claim(s) 5 and 15 is/are rejected under 35 U.S.C. 103 as being unpatentable over Pham in view of Butvinik in further view of Sokar et al. (“Self-Attention Meta-Learner for Continual Learning”) (hereafter referred to as Sokar).
Regarding claim 5, Pham in view of Butvinik teaches the method of claim 4. Pham in view of Butvinik does not teach, but Sokar does teach
wherein the first layer and the second layer are each a self-attention layer (Sokar, page 2, Figure 1
PNG
media_image4.png
550
1257
media_image4.png
Greyscale
Examiner notes that the meta attention layers are the self-attention layers.).
Pham, Butvinik, and Sokar are considered analogous to the claimed invention because they use continual learning to train models. It would have been obvious to one having ordinary skill in the art prior to the effective filing date to have modified Pham in view of Butvinik to include self-attention layers like in Sokar. Doing so is advantageous because “SAM [Self-Attention Meta-Learner] learns a prior knowledge that can generalize to new distributions and learns to boost the features relevant to the input data” (Sokar, page 9, Conclusion).
Regarding claim 15, claim 15 recites substantially similar limitations to claim 5, and is therefore rejected under the same analysis.
Claim(s) 6-7 and 16-17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Pham in view of Butvinik in further view of Sokar et al. (“Self-Attention Meta-Learner for Continual Learning”) (hereafter referred to as Sokar).
Regarding claim 6, Pham in view of Butvinik teaches the method of claim 4. Pham in view of Butvinik does not teach, but Sokar does teach
wherein inserting the task-invariant prompt at the first layer of the pre-trained model comprises prepending the task-invariant prompt to an input … feature of the first layer (Sokar, page 2, Figure 1
PNG
media_image4.png
550
1257
media_image4.png
Greyscale
where “The input of the attention block is the feature maps X = {X1, X2, …, Xc} resulted from applying the convolutional operator F on the output of the previous layer” (Sokar, page 3, 2nd column, 3rd paragraph). Examiner notes that the task-invariant prompt is the query to learn the sample. Examiner further notes that the arrival of the input, or sample in Figure 1 is prior to X, an input feature that is used in the self-attention layer)
Pham, Butvinik, and Sokar are considered analogous to the claimed invention because they use continual learning to train models. It would have been obvious to one having ordinary skill in the art prior to the effective filing date to have modified Pham in view of Butvinik to prepend a prompt to a feature like in Sokar. Doing so is advantageous because “SAM [Self-Attention Meta-Learner] learns a prior knowledge that can generalize to new distributions and learns to boost the features relevant to the input data” (Sokar, page 9, Conclusion).
Pham in view of Butvinik and Sokar do not explicitly disclose that the input features are embeddings. However, Hu does disclose
input embedding feature (Hu, page 1, 2nd column, page 1, “The features U are first passed through a squeeze operation, which aggregates the feature maps across spatial dimensions HxW to produce a channel descriptor. This descriptor embeds the global distribution of channel-wise feature responses, enabling information from the global receptive field of the network to be leveraged by its lower layers.”)
Pham, Butvinik, Sokar and Hu are considered analogous to the claimed invention because they use self-attention layers to train machine learning models. It would have been obvious to one having ordinary skill in the art prior to the effective filing date to have modified Pham in view of Butvinik and Sokar embed features like in Hu. Thus would be a simple substitution of one known element (the self-attention layer in Sokar) for another (the self-attention layer in Hu) to obtain predictable results (train machine learning models) (MPEP 2143 I. (B) Simple substitution of one known element for another to obtain predictable results.).
Regarding claim 7, Pham in view of Butvinik teaches the method of claim 4. Pham in view of Butvinik does not teach, but Sokar does teach
wherein inserting the respective task-specific prompt at the second layer of the pre-trained model comprises prepending the respective task- specific prompt to an input embedding feature of the second layer (Sokar, page 2, Figure 1
PNG
media_image4.png
550
1257
media_image4.png
Greyscale
where “The input of the attention block is the feature maps X = {X1, X2, …, Xc} resulted from applying the convolutional operator F on the output of the previous layer” (Sokar, page 3, 2nd column, 3rd paragraph). Examiner notes that the task-specific prompt is the arrival of a sample. Examiner further notes that the arrival of the input in Figure 1 is prior to X, an input feature that is used in the self-attention layer.)
Pham, Butvinik, and Sokar are considered analogous to the claimed invention because they use continual learning to train models. It would have been obvious to one having ordinary skill in the art prior to the effective filing date to have modified Pham in view of Butvinik to prepend a prompt to a feature like in Sokar. Doing so is advantageous because “SAM [Self-Attention Meta-Learner] learns a prior knowledge that can generalize to new distributions and learns to boost the features relevant to the input data” (Sokar, page 9, Conclusion).
Pham in view of Butvinik and Sokar do not explicitly disclose that the input features are embeddings. However, Hu does disclose
input embedding feature (Hu, page 1, 2nd column, page 1, “The features U are first passed through a squeeze operation, which aggregates the feature maps across spatial dimensions HxW to produce a channel descriptor. This descriptor embeds the global distribution of channel-wise feature responses, enabling information from the global receptive field of the network to be leveraged by its lower layers.”)
Pham, Butvinik, Sokar and Hu are considered analogous to the claimed invention because they use self-attention layers to train machine learning models. It would have been obvious to one having ordinary skill in the art prior to the effective filing date to have modified Pham in view of Butvinik and Sokar embed features like in Hu. Thus would be a simple substitution of one known element (the self-attention layer in Sokar) for another (the self-attention layer in Hu) to obtain predictable results (train machine learning models) (MPEP 2143 I. (B) Simple substitution of one known element for another to obtain predictable results.).
Regarding claim 16, claim 16 recites substantially similar limitations to claim 6, and is therefore rejected under the same analysis.
Regarding claim 17, claim 17 recites substantially similar limitations to claim 7, and is therefore rejected under the same analysis.
Response to Arguments
On pages 6-7, Applicant argues:
Specifically, the rejection appears to suggest that the terms "prepending" and "input embedding feature" do not provide a clear boundary for the scope of the claims.
Applicant respectfully submits that the specification provides a clear structural and mathematical description of an example "input embedding feature." Application describes that "Given a pre-trained vision transformer f with N consecutive MSA layers, the input embedding feature of the i-th MSA layer may be denoted as hCi), i=l, 2, ... , N." Para. [0025].
Furthermore, the Specification provides the dimensionality of this feature, stating that "the input to the MSA layer is hE:[R{LxD_" Id. These descriptions provide the exact nature of the feature and its mathematical properties. Thus, the term is not "vague" but is a specific technical term of art in the field of transformer architectures.
Applicant respectfully submits that the Term "prepending" describes a clear operation on embeddings. The Specification explains that prompts are used for "attaching fixed or learnable 'instructions."' Para. [0018]. Regarding the specific mechanism of insertion, the Specification states that "inserting the task-invariant prompt at the first layer of the pre-trained model may include prepending the task-invariant prompt to an input embedding feature of the first layer". Para. [0005].
The Specification further clarifies that "applying a prompting function may be viewed as modifying the inputs of the MSA layers". Para. [0031]. Within the context of "prompting functions" such as "Prom pt Tuning (Pro-T) and/ or Prefix Tuning (Pre-T)", the term "prepending" has a well-defined meaning. Id.
The relationship is further solidified by Equations (1) and (2), which define the prompting function hg (i)=f prompt(g,h (i)) used to "attach the prompts to the hidden embeddings." Para. [0026]. This explicitly shows the prompt g being combined with the input embedding feature h(i).
Applicant respectfully submits that the Specification provides direct support for the claimed limitations. It defines the input embedding feature as Mi)_ It describes prepending as a method of "attaching" or "inserting" prompts to these features via a defined prompting function f prompt.
Because the claim language is "clear and precise," Applicant respectfully requests that the 112 rejection be withdrawn.
Applicant’s arguments that the claim language is clear and precise have been fully considered and are persuasive. Specifically, the argument on pages 6-7 regarding that the prompting function as defined by equations 1 and 2 from the instant specification is used to attach the prompts to hidden embeddings through the equations was persuasive. The 112(b) rejections of the claims have been withdrawn.
On page 8, Applicant argues:
The claims are not directed to an abstract idea such as a "mathematical algorithm" or" mental process." Instead, they describe a physical and functional implementation of a model trainer. The specification provides that this framework "explicitly decouples prompt parameters into a task-invariant prompt (i.e., a general-prompt or G-Prompt) for learning task-invariant knowledge and a task-specific prompt (i.e., an expert-prompt or E-Prompt) for learning task specific knowledge." Para. [0020].
This specific decoupling of prompts is a "simple yet novel and effective framework that largely improves the continual learning practicality without data or memory access concerns." Para. [0020]. Unlike abstract data manipulation, this can involve the act of "inserting the task invariant prompt at a first layer ... and inserting the respective task-specific prompt at a second layer." Para. [0023]. This arrangement allows "different instructions [to] interact with the corresponding representations more effectively." Para. [0029]. Such internal architectural modifications are technical in nature and do not represent a "mental process" or "mathematical algorithm."
Regarding the Applicant’s argument that the claims integrate any abstract idea into a practical application, Examiner respectfully disagrees. Specifically, Examiner respectfully notes that the specification provides a bare assertion of improvement in paragraphs 0020 and 0023 without the detail necessary to be apparent to a person of ordinary skill in the art, and thus cannot provide an improvement (MPEP 2106.04(d)(1)). Additionally, the integration into a practical application cannot come from the abstract idea alone (MPEP 2106.04(d)). As such, the additional elements do not provide an integration into a practical application because “A computer-implemented method when executed by data processing hardware causes the data processing hardware to perform operations comprising” and “training a model by attaching the task-invariant prompt at a first layer of the model and attaching the selected respective task-specific prompt at a second layer of the model” amounts to mere “apply it on a computer” (see MPEP 2106.05(f)). “Obtaining a set of training samples, each training sample in the set of training samples associated with a respective task of a plurality of different tasks”, “obtaining a task-invariant prompt representative of learned knowledge common to each respective task of the plurality of different tasks”, and “for each respective task of the plurality of different tasks, obtaining a respective task-specific prompt representative of learned knowledge specific to the respective task” recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).
On pages 8-9, Applicant argues:
Even if the claims were characterized as involving an abstract idea, they provide an inventive concept that is "significantly more" than conventional practices. The Specification explains that conventional prompt work "simply places prompts only at a first MSA layer or at every MSA layer." Id.
In contrast, the claims provide "the two types of prompts more flexibility to attach to the most proper positions in a decoupled way." Id. This unconventional and flexible attachment mechanism allows for "rehearsal-free continual learning." Para. [0035]. By encoding knowledge in prompts rather than a buffer, the model trainer 110 "outperforms rehearsal-based methods even with relatively large buffer size." Para. [0020]. This technical efficiency in memory usage and learning capacity represents a "substantial' and "practical" improvement over prior systems.
Regarding Applicant’s argument that the claims provide significantly more, Examiner respectfully disagrees. Specifically, Examiner notes that the claims do not reflect the statement of “implementations herein provide the two types of prompts more flexibility to attach to the most proper positions in a decoupled way” (MPEP 2106.05(a)). Additionally, the improvements cannot come from the abstract idea alone (MPEP 2106.05(a)). As such, the additional elements do not provide significantly more because “A computer-implemented method when executed by data processing hardware causes the data processing hardware to perform operations comprising” and “training a model by attaching the task-invariant prompt at a first layer of the model and attaching the selected respective task-specific prompt at a second layer of the model” use a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)). “Obtaining a set of training samples, each training sample in the set of training samples associated with a respective task of a plurality of different tasks”, “obtaining a task-invariant prompt representative of learned knowledge common to each respective task of the plurality of different tasks”, and “for each respective task of the plurality of different tasks, obtaining a respective task-specific prompt representative of learned knowledge specific to the respective task” are well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
On page 9, Applicant argues:
B. Independent Claim 11
For at least the reasons discussed above with respect to claim 1, Applicant submits that claim 11 is also patent eligible.
C. Dependent Claims
For at least the reasons discussed above with respect to the above independent claims, Applicant submits that the claims respectively dependent thereon are also patent eligible. Applicant also notes that the patentability of the dependent claims certainly does not hinge on the patentability of independent claims. Some or all of these claims may possess features that are independently patentable, regardless of the patentability of the independent claims.
Regarding the Applicant’s argument that the dependent claims are allowable at least due in part to their dependency on the independent claims, the Examiner respectfully disagrees and notes the instant rejections and response to arguments regarding the independent claims above. Examiner also notes that claim 11 is rejected for similar reasons as claim 1 and therefore, maintains its rejection as well.
On page 10, Applicant argues:
Applicant respectfully submits that there is nothing in either reference or the knowledge of those skilled in the art that provides a motivation to combine two references that are fundamentally dependent on rehearsal data to arrive at the "rehearsal-free" method of Claim 1. The '143 Application describes a "rehearsal-free model" as one that "encodes learned knowledge from sequential tasks in small learnable parameters called prompts." Neither cited reference teaches this approach.
In response to applicant’s argument that there is no teaching, suggestion, or motivation to combine the references, the examiner recognizes that obviousness may be established by combining or modifying the teachings of the prior art to produce the claimed invention where there is some teaching, suggestion, or motivation to do so found either in the references themselves or in the knowledge generally available to one of ordinary skill in the art. See In re Fine, 837 F.2d 1071, 5 USPQ2d 1596 (Fed. Cir. 1988), In re Jones, 958 F.2d 347, 21 USPQ2d 1941 (Fed. Cir. 1992), and KSR International Co. v. Teleflex, Inc., 550 U.S. 398, 82 USPQ2d 1385 (2007). In this case, Pham and Butvinik are analogous to the claimed invention because they both use continual learning to train a model. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date to have implemented Pham on data processing hardware. Thus, this would be applying a known technique (continual learning) to a known device (data processing hardware) ready for improvement to yield predictable results (train a model) (MPEP 2143 I. (C) Use of known technique to improve similar devices (methods, or products) in the same way).
On pages 10-11, Applicant argues:
Pham utilizes a "slow learner" which is a "standard backbone network" and a "fast
learner" which is a "simple CNN". Sec. 2.3, 2.4. These represent entire neural network architectures, not "prompts" attached to a frozen model. Pham's "slow learner" focuses on learning "generic features" but does so through "Self-Supervised Learning (SSL)" on memory data, not via a task-invariant prompt. Introduction, page 2, first full paragraph. Moreover, these learners are not learnable parameters designed to be attached to the internal layers of a model.
Butvinik divides model parameters into a "subset of shared model parameters" and a "subset of task-specific model parameters." Para. [0025]. Butvinik's parameters are "generated by applying a propagator to the one or more training samples", which is a fundamentally different mechanism from the prompt-based architecture of Claim 1.
Accordingly, neither reference suggests the specific use of "prompts," for example to "instruct the model to properly reuse learned representations instead of learning new representations from scratch." Because Pham and Butvinik lack these specific prompt elements, they cannot suggest training a model by attaching the task-invariant prompt at a first layer of the model and attaching the selected respective task-specific prompt at a second layer of the model," as recited in claim 1,” as recited in claim 1.
Regarding Applicant’s argument that the prior art of record does not teach prompt, Examiner respectfully disagrees. Specifically, Examiner notes that the broadest reasonable interpretation of prompt can be a query or signal to continue processing. As such, Pham discloses this in, page 3, 3rd paragraph, “Second, the supervised learning phase happens whenever a labeled sample arrives, which triggers the fast learner to first query the representation from the slow learner and adapt it to learn this sample. The incurred loss will be backpropagated into both learners for supervised knowledge consolidation. Additionally, the fast learner’s adaptation is per-sample based and do not require additional information such as the task identifiers” where “(i) the slow learner is responsible for learning a general, task agnostic representation; and (ii) the fast learner learns with labeled data from the continuum to quickly capture the new information and then consolidate the knowledge to the slow learner” (Pham, page 3, 2nd paragraph). Examiner respectfully notes that the task-specific prompt is the arrival of the labeled sample to the fast learner to learn the sample and the task-invariant prompt is the query from the fast learner to the slow learner
On page 11, Applicant argues:
B. Independent Claim 11
For at least the reasons discussed above with respect to claim 1, Applicant submits that claim 11 is also patent eligible.
C. Dependent Claims
For at least the reasons discussed above with respect to the above independent claims, Applicant submits that the claims respectively dependent thereon are also patent eligible. Applicant also notes that the patentability of the dependent claims certainly does not hinge on the patentability of independent claims. Some or all of these claims may possess features that are independently patentable, regardless of the patentability of the independent claims.
Regarding the Applicant’s argument that the dependent claims are allowable at least due in part to their dependency on the independent claims, the Examiner respectfully disagrees and notes the instant rejections and response to arguments regarding the independent claims above. Examiner also notes that claim 11 is rejected for similar reasons as claim 1 and therefore, maintains its rejection as well.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Arani et al. (“Learning Fast, Learning Slow: A General Continual Learning Method Based on Complementary Learning System”) also discloses task-invariant and task-specific prompts in continual learning.
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 KAITLYN R HAEFNER whose telephone number is (571)272-1429. The examiner can normally be reached Monday - Thursday: 7:15 am - 5:15 pm EST.
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, Michelle Bechtold can be reached at (571) 431-0762. 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.
/K.R.H./Examiner, Art Unit 2148 /MICHELLE T BECHTOLD/Supervisory Patent Examiner, Art Unit 2148