DETAILED ACTION
1. This communication is in response to the Application No. 18/585,396 filed on February 23, 2024 in which Claims 1-20 are presented for examination.
Notice of Pre-AIA or AIA Status
2. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
3. The information disclosure statement submitted on 06/10/2025 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Rejections - 35 USC § 101
4. 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.
5. Claims 19-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claim does not fall within at least one of the four categories of patent eligible subject matter because the broadest reasonable interpretation of the “computer readable storage medium” encompasses signals per se. Applicant’s specification Par. [0068] and Par. [0070] merely recite the term “computer readable storage medium” without stating that the “computer readable storage medium” excludes propagated and/or transitory signals per se. Furthermore, Claims 19-20 merely state that the “computer program product” comprises a “computer readable storage medium” containing instructions executable by one or more processors – however, the one or more processors themselves are seemingly not part of the computer program product of Claims 19-20, according to the currently drafted claim language. A claim whose BRI covers both statutory and non-statutory embodiments embraces subject matter that is not eligible for patent protection and therefore is directed to non-statutory subject matter. See MPEP 2106.03(II). It is suggested that Claims 19-20 be amended to recite a “non-transitory computer readable storage medium” to overcome this rejection.
6. Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Claim 1:
Step 1: Claim 1 is a method type claim. Therefore, Claims 1-9 are directed to either a process, machine, manufacture, or composition of matter.
2A Prong 1: If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation by mathematical calculation but for the recitation of generic computer components, then it falls within the “Mathematical Concepts” grouping of abstract ideas.
generating, from the set of fine-tuning tasks, a first task combination comprising a subset of the set of fine-tuning tasks (mental process – generating, from the set of fine-tuning tasks, a first task combination comprising a subset of the set of fine-tuning tasks may be performed manually by a user observing/analyzing the set of fine-tuning tasks and accordingly using judgement/evaluation to generate (with the aid of pen and paper) a first task combination comprising a subset of the set of fine-tuning tasks, based on said analysis of the set of fine-tuning tasks)
identifying a shared subspace within the subset of the set of fine-tuning tasks (mental process – identifying a shared subspace within the subset of the set of fine-tuning tasks may be performed manually by a user observing/analyzing the subset of the set of fine-tuning tasks and accordingly using judgement/evaluation to identify a shared subspace (with the aid of pen and paper) within the subset of the set of fine-tuning tasks, based on said analysis. Alternatively, as supported by dependent claim 2, this identifying may be performed by mathematical process by identifying a low rank matrix which is common to the subset of the set of fine-tuning tasks of the first task combination)
2A Prong 2: This judicial exception is not integrated into a practical application.
Additional elements:
receiving a pre-trained large language model (Adding insignificant extra-solution activity to the judicial exception – see MPEP 2106.05(g))
receiving a set of fine-tuning tasks for the pre-trained large language model, the set of fine-tuning tasks comprising at least a first fine-tuning task and a second fine-tuning task (Adding insignificant extra-solution activity to the judicial exception – see MPEP 2106.05(g))
responsive to identifying the shared subspace, fine-tuning the pre-trained large language model jointly over the first task combination (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level recitation of training/tuning a machine learning model with previously determined data without significantly more)
2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Additional elements:
receiving a pre-trained large language model (MPEP 2106.05(d)(II) indicates that merely “Receiving or transmitting data over a network” is a well-understood, routine, conventional function when it is claimed in a merely generic manner (as it is in the present claim). Thereby, a conclusion that the claimed limitation is well-understood, routine, conventional activity is supported under Berkheimer)
receiving a set of fine-tuning tasks for the pre-trained large language model, the set of fine-tuning tasks comprising at least a first fine-tuning task and a second fine-tuning task (MPEP 2106.05(d)(II) indicates that merely “Receiving or transmitting data over a network” is a well-understood, routine, conventional function when it is claimed in a merely generic manner (as it is in the present claim). Thereby, a conclusion that the claimed limitation is well-understood, routine, conventional activity is supported under Berkheimer)
responsive to identifying the shared subspace, fine-tuning the pre-trained large language model jointly over the first task combination (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level recitation of training/tuning a machine learning model with previously determined data without significantly more. This cannot provide an inventive concept)
For the reasons above, Claim 1 is rejected as being directed to an abstract idea without significantly more. This rejection applies equally to dependent claims 2-9. The additional limitations of the dependent claims are addressed below.
Regarding Claim 2:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 2 depends on.
identifying the shared subspace comprises identifying a low rank matrix which is common to the subset of the set of fine-tuning tasks of the first task combination (mathematical process – identifying the shared subspace by identifying a low rank matrix which is common to the subset of the set of fine-tuning tasks of the first task combination may be performed by mathematical process as outlined by Applicant’s specification Par. [0016]. Alternatively, this identifying may be performed manually by a user observing/analyzing the subset of the set of fine-tuning tasks of the first combination and accordingly using judgement/evaluation to identify a low rank matrix that is common to the subset (with the aid of pen and paper), based on said analysis))
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 3:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 3 depends on.
generating, from the set of fine-tuning tasks, a plurality of task combinations, wherein each task combination of the plurality of task combinations comprises a unique subset of the set of fine-tuning tasks (mental process – generating a plurality of task combinations, where each task combination comprises a unique subset of the set of fine-tuning tasks may be performed manually by a user observing/analyzing the set of fine-tuning tasks and accordingly using judgement/evaluation to generate a plurality of unique task combinations (with the aid of pen and paper), based on said analysis)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 4:
Step 2A Prong 1:
See the rejection of Claim 3 above, which Claim 4 depends on.
Step 2A Prong 2 & Step 2B:
generating a candidate fine-tuned model of the pre-trained large language model for each of the plurality of task combinations (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner's note: high level recitation of training a machine learning model with previously determined data without significantly more. This cannot provide an inventive concept)
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 5:
Step 2A Prong 1:
See the rejection of Claim 4 above, which Claim 5 depends on.
Step 2A Prong 2 & Step 2B:
evaluating an inference performance of each of the candidate fine-tuned models using labeled task-specific data (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner's note: high level recitation of training a machine learning model with previously determined data without significantly more. This cannot provide an inventive concept)
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 6:
Step 2A Prong 1: See the rejection of Claim 5 above, which Claim 6 depends on.
selecting, from the candidate fine-tuned models, a candidate having a highest performance metric (mental process – selecting a candidate having a highest performance metric may be performed manually by a user observing/analyzing the candidate fine-tuned model and accordingly using judgement/evaluation to select a candidate having a highest associated performance metric (with the aid of pen and paper), based on said analysis)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 7:
Step 2A Prong 1: See the rejection of Claim 6 above, which Claim 7 depends on.
updating at least one weight of the pre-trained large language model to match a respective weight of the candidate having the highest performance metric (mental process – updating at least one weight of the pre-trained large language model to match a respective weight of the candidate having the highest performance metric may be performed manually by a user observing/analyzing both the weights of the pre-trained large language model and the candidate having the highest performance metric and accordingly using judgement/evaluation to update at least one weight of the pre-trained large language model (with the aid of pen and paper) to match a respective weight of the candidate having the highest performance metric, based on said analysis)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 8:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 8 depends on.
fine-tuning the pre-trained large language model jointly over the first task combination comprises enforcing sparsity by taking proximal steps after each gradient update (mathematical process – enforcing sparsity by taking proximal steps after each gradient update may be performed by mathematical process, utilizing an algorithm such as proximal gradient descent)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 9:
Step 2A Prong 1:
See the rejection of Claim 8 above, which Claim 9 depends on.
Step 2A Prong 2 & Step 2B:
wherein sparsity is enforced according to a task-specific diagonal matrix having sparse diagonal entries (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that sparsity is enforced according to a task-specific diagonal matrix having sparse diagonal entries does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Independent Claim 10 recites substantially the same limitations as Claim 1, in the form of a system, including generic computer components. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
For the reasons above, Claim 10 is rejected as being directed to an abstract idea without significantly more. This rejection applies equally to dependent claims 11-18. The additional limitations of the dependent claims are addressed below.
Claim 11 recites substantially the same limitations as Claim 2, in the form of a system, including generic computer components. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim 12 recites substantially the same limitations as Claim 3, in the form of a system, including generic computer components. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim 13 recites substantially the same limitations as Claim 4, in the form of a system, including generic computer components. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim 14 recites substantially the same limitations as Claim 5, in the form of a system, including generic computer components. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim 15 recites substantially the same limitations as Claim 6, in the form of a system, including generic computer components. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim 16 recites substantially the same limitations as Claim 7, in the form of a system, including generic computer components. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim 17 recites substantially the same limitations as Claim 8, in the form of a system, including generic computer components. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim 18 recites substantially the same limitations as Claim 9, in the form of a system, including generic computer components. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Independent Claim 19 recites substantially the same limitations as Claim 1, in the form of a computer program product, including generic computer components. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
For the reasons above, Claim 19 is rejected as being directed to an abstract idea without significantly more. This rejection applies equally to dependent claim 20. The additional limitations of the dependent claims are addressed below.
Claim 20 recites substantially the same limitations as Claim 2, in the form of a computer program product, including generic computer components. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim Rejections - 35 USC § 102
7. The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
8. Claims 1-2, 10-11, and 19-20 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Panda et al. (hereinafter Panda) (US PG-PUB 20250005370).
Regarding Claim 1, Panda teaches a method (Panda, Par. [0003], “Principles of the invention provide systems and techniques for multitask prompt tuning for parameter-efficient transfer learning. In one aspect, an exemplary method includes the operations of decomposing, using a hardware processor, a source task prompt of each of a plurality of source tasks as a multiplication of a shared prompt matrix shared across the source tasks and a low-rank task-specific matrix; […]”, therefore, a method is disclosed) comprising:
receiving a pre-trained large language model (Panda, Par. [0040], “Given a pre-trained language model with parameters Θ and one target task
PNG
media_image1.png
38
29
media_image1.png
Greyscale
with training data (X, Y)={xi, yi}i=1 N, directly finetuning all the parameters by maximizing the conditional probability P(Y|X; Θ) is expensive and often tends to overfit on small datasets. An alternative to finetuning, which is more parameter-efficient, is prompt tuning (PT), which randomly initializes a small number of learnable prompt vectors (i.e., soft prompts) to be prepended to the input embeddings of the PLM while freezing model parameters Θ.”, therefore, a pre-trained large language model is received/obtained);
receiving a set of fine-tuning tasks for the pre-trained large language model, the set of fine-tuning tasks comprising at least a first fine-tuning task and a second fine-tuning task (Panda, Par. [0033], “FIG. 1A is a high-level block diagram of an example conventional approach for transferring prompt vectors from different tasks. The method first trains soft prompts 218 on multiple source tasks 212 and then retrieves or aggregates 220 the soft prompts 218 to generate task prompts 224. Pretrained prompts are then used to initialize the corresponding prompt 224 for further finetuning on a target task 228 based on a (potentially learned) similarity measure.”, thus, a plurality/multiple source tasks are received for the pre-trained large language model, the set of fine-tuning tasks comprising at least a first fine-tuning task and a second fine-tuning task, as better depicted by Panda Figure 3 which shows two according source tasks in one embodiment);
generating, from the set of fine-tuning tasks, a first task combination comprising a subset of the set of fine-tuning tasks (Panda, Par. [0048], “Thus, the total number of tunable parameters for a single target task 228 is (l×d)+(l+d). For a group of target tasks 228, the total number of tunable parameters is (l×d)+(l+d)τ, where τ is the number of target tasks 228. Different methods in terms of the number of trainable parameters are listed and compared in the table of FIG. 7 . (Pearson Correlation was adopted as evaluation metrics. “param/task” represents a number of trainable parameters for each task in a first conventional sentence classification benchmark. The top part of the table of FIG. 7 denotes model adaptation to each target task 228 (so param/task for MPT is just (l×d)+(l+d)). The bottom part (marked by *) denotes model adaptation to a group of tasks, where the param/task for MPT* is (l×d)/τ+(l+d). See section entitled Source Training and Target Adaptation.)”, thus, a first task combination/grouping is generated, comprising a subset of the set of fine-tuning tasks – See also Panda Par. [0037] which outlines how multi-task learning is utilized, comprising the combination of multiple different source tasks);
identifying a shared subspace within the subset of the set of fine-tuning tasks (Panda, Par. [0041], “Specifically, task prompts for source tasks 212 are decomposed into a task-shared component 232 and a low-rank task-specific component 214 (prompt decomposition), where the former is shared across all tasks 212, and the latter is task-specific. Prompt distillation is also used to better transfer multitask knowledge to the shared component by distilling knowledge from multiple tasks-specific source prompts. Once learned, the shared prompt matrix is adapted to a downstream target task 228 via low-rank multiplicative updates.”, therefore, a shared subspace (task-shared component and shared prompt matrix) within the subset of the set of fine-tuning tasks is identified); and
responsive to identifying the shared subspace, fine-tuning the pre-trained large language model jointly over the first task combination (Panda, Par. [0075], “In one example embodiment, a teacher prompt Pk t for a k-th source task using prompt tuning is obtained and multitask training is performed on the source tasks 212 to jointly learn the single shared soft prompt 232 via a knowledge distillation loss function (Equation 5).” & Claim 1, “decomposing, using a hardware processor, a source task prompt of each of a plurality of source tasks as a multiplication of a shared prompt matrix shared across the source tasks and a low-rank task-specific matrix; performing, using the hardware processor, prompt distillation to transfer multitask knowledge to the shared prompt matrix by distilling knowledge from the source task prompts; performing, using the hardware processor, low-rank multiplicative updates to the shared prompt matrix to transfer the multitask knowledge to one or more target tasks; […]”, thus, responsive to identifying the shared subspace (shared prompt), the pre-trained large language model is fine-tuned/updated jointly over the first task combination).
Regarding Claim 2, Panda teaches the method of claim 1, wherein identifying the shared subspace comprises identifying a low rank matrix which is common to the subset of the set of fine-tuning tasks of the first task combination (Panda, Par. [0034], “The soft prompt of each source task 212 (which can be represented as a prompt matrix) is decomposed as a multiplication of a shared matrix (depicted as task-shared prompt 232 on the left side of FIG. 1B and, equivalently, as transferred shared prompt 232 on the right side of FIG. 1B) and a low-rank task-specific matrix (depicted as task-specific component 214 on the left side of FIG. 1A), and this decomposition is found to be more effective than simply sharing the prompt matrix across all tasks. This decomposition is learned through knowledge distillation from soft prompts obtained from regular prompt tuning for each source task 212. To transfer to new tasks 228, low-rank multiplicative updates are performed to the shared prompt matrix.”, therefore, identifying the shared subspace comprises identifying a low rank matrix which is common to the subset of the set of fine-tuning tasks of the first task combination (shared prompt matrix with low-rank multiplicative updates performed)).
Regarding Claim 10, Panda teaches a system having a memory, computer readable instructions, and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations (Panda, Par. [0083], “In one aspect, an apparatus comprises a memory and at least one processor, coupled to the memory, and operative to perform operations comprising decomposing a source task prompt 216 of each of a plurality of source tasks 212 as a multiplication of a shared prompt matrix shared across the source tasks 212 and a low-rank task-specific matrix; performing prompt distillation to transfer multitask knowledge to the shared prompt matrix by distilling knowledge from the source task prompts 216; and performing low-rank multiplicative updates to the shared prompt matrix to transfer the multitask knowledge to one or more target tasks 228.”, thus, a system having a memory, computer readable instructions, and one or more processors for executing instructions is disclosed) comprising: […]
The rest of the claim language in Claim 10 recites substantially the same limitations as Claim 1, in the form of a system, therefore it is rejected under the same rationale.
Claim 11 recites substantially the same limitations as Claim 2 in the form of a system, therefore it is rejected under the same rationale.
Regarding Claim 19, Panda teaches a computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by one or more processors to cause the one or more processors to perform operations (Panda, Claim 14, “A computer program product, comprising:
one or more tangible computer-readable storage media and program instructions stored on at least one of the one or more tangible computer-readable storage media, the program instructions executable by a processor, the program instructions comprising: […]”, therefore, a computer program product comprising a computer readable storage medium having program instructions embodied, executable by a processor is disclosed) comprising: […]
The rest of the claim language in Claim 19 recites substantially the same limitations as Claim 1, in the form of a computer program product, therefore it is rejected under the same rationale.
Claim 20 recites substantially the same limitations as Claim 2 in the form of a computer program product, therefore it is rejected under the same rationale.
Claim Rejections - 35 USC § 103
9. 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.
10. Claims 3-7 and 12-16 are rejected under 35 U.S.C. 103 as being unpatentable over Panda et al. (hereinafter Panda) (US PG-PUB 20250005370), in view of Ilharco et al. (hereinafter Ilharco) (“Editing Models with Task Arithmetic”).
Regarding Claim 3, Panda teaches the method of claim 1.
Panda does not explicitly disclose generating, from the set of fine-tuning tasks, a plurality of task combinations, wherein each task combination of the plurality of task combinations comprises a unique subset of the set of fine-tuning tasks.
However, Ilharco teaches generating, from the set of fine-tuning tasks, a plurality of task combinations, wherein each task combination of the plurality of task combinations comprises a unique subset of the set of fine-tuning tasks (Ilharco, Pg. 5, “Adding task vectors builds multi-task models for image classification tasks. Accuracy is averaged over all downstream tasks. When more task vectors can be built. Each point represents an experiment with a subset of the eight tasks we study, and the solid line connects the average performance for each subset size […] Beyond pairs of tasks, we explore adding task vectors for all possible subsets of the tasks (28 in total)”, thus, a plurality of task combinations are generated from the set of fine-tuning tasks, where each task combination of the plurality comprises a unique subset of the set of fine-tuning tasks).
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of claim 1, as disclosed by Panda to include generating, from the set of fine-tuning tasks, a plurality of task combinations, wherein each task combination of the plurality of task combinations comprises a unique subset of the set of fine-tuning tasks, as disclosed by Ilharco. One of ordinary skill in the art would have been motivated to make this modification to enable generating a plurality of unique task combinations, which may improve multi-task learning and enhance performance on tasks where data is scarce (Ilharco, Pg. 1, “We show that we can edit a variety of models with task arithmetic—performing simple arithmetic operations on task vectors (Figure 1b-d). For example, negating a vector can be used to remove undesirable behaviors or unlearn tasks, while adding task vectors leads to better multi-task models, or even improves performance on a single task. Finally, when tasks form an analogy relationship, task vectors can be combined to improve performance on tasks where data is scarce.”)
Regarding Claim 4, Panda in view of Ilharco teaches the method of claim 3, further comprising generating a candidate fine-tuned model of the pre-trained large language model for each of the plurality of task combinations (Ilharco, Pg. 3, “We focus on three arithmetic expressions over task vectors, as illustrated in Figure 1: negating a task vector, adding task vectors together, and combining task vectors to form analogies. All operations are applied element-wise to the weight vectors. When negating a task vector τ, applying the resulting vector τnew = −τ corresponds to extrapolating between the fine-tuned model and the pre-trained model. The resulting model is worse at the target task, with little change in performance on control tasks (Section 3). Adding two or more task vectors τi yields τnew = iτi, and results in a multi-task model proficient in all tasks, sometimes even with gains over models fine-tuned on individual tasks (Section 4). Finally, when tasks A, B, C and Dform an analogy in the form “A is to B as C is to D”, the task vector τnew = τC + (τB − τA) improves performance on task D, even if there is little or no data for that task (Section 5). For all operations, the model weights obtained by applying τnew are given by θnew = θ+λτnew, where the scaling term λ is determined using held-out validation sets”, therefore, a candidate fine-tuned model of the pre-trained large language model is generated (model is edited, creating a different instance of the large language model) for each of the plurality of task combinations (different task arithmetic operations, which uniquely combine tasks)).
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of claim 3, as disclosed by Panda in view of Ilharco to include generating a candidate fine-tuned model of the pre-trained large language model for each of the plurality of task combinations, as disclosed by Ilharco. One of ordinary skill in the art would have been motivated to make this modification to enable efficient knowledge transfer without the need for additional training and/or access to additional training data (Ilharco, Pg. 4, “We now turn our attention to adding task vectors, either to build multi-task models that are proficient on multiple tasks simultaneously, or to improve single-task performance. This operation allows us to reuse and transfer knowledge either from in-house models, or from the multitude of publicly available fine-tuned models, without additional training or access to training data. We explore addition on various image classification and natural language processing tasks”).
Regarding Claim 5, Panda in view of Ilharco teaches the method of claim 4, further comprising evaluating an inference performance of each of the candidate fine-tuned models using labeled task-specific data (Ilharco, Pg. 6, Tables 3 & 4 which depict the performance evaluated for each of the candidate/edited fine-tuned models using labeled task-specific data (supported by Appendix section D.7 which details the data used, comprising labeled sentences for evaluating natural language processing tasks)).
The reasons of obviousness have been noted in the rejection of Claims 3 and 4 above and applicable herein.
Regarding Claim 6, Panda in view of Ilharco teaches the method of claim 5, further comprising selecting, from the candidate fine-tuned models, a candidate having a highest performance metric (Ilharco, Pg. 2, “Adding task vectors results in better multi-task models, or improved perfor mance on a single task. In Section 4, we add task vectors from various image models (CLIP, Radford et al. [78]) and compare the performance of the resulting model with using multiple specialized fine-tuned models. We find that the single resulting model can be competitive with using multiple specialized models. Adding two task vectors maintains 98.9% of the accuracy, and the average performance on the entire set of tasks increases as more task vectors are added. Moreover, adding a task vector from a different task can improve performance on a target task using text models (T5, Raffel et al. [79]).”, therefore, a candidate/edited model having a highest performance metric (greatest accuracy, for example maintaining 98.9% of accuracy) is selected for editing the original pre-trained model).
The reasons of obviousness have been noted in the rejection of Claims 3 and 4 above and applicable herein.
Regarding Claim 7, Panda in view of Ilharco teaches the method of claim 6, further comprising updating at least one weight of the pre-trained large language model to match a respective weight of the candidate having the highest performance metric (Ilharco, Pg. 1, Abstract, “A task vector specifies a direction in the weight space of a pre-trained model, such that movement in that direction improves performance on the task. We build task vectors by subtracting the weights of a pre-trained model from the weights of the same model after fine-tuning on a task. We show that these task vectors can be modified and combined together through arithmetic operations such as negation and addition, and the behavior of the resulting model is steered accordingly. Negating a task vector decreases performance on the target task, with little change in model behavior on control tasks. Moreover, adding task vectors together can improve performance on multiple tasks at once. Finally, when tasks are linked by an analogy relationship of the form “A is to B as C is to D”, combining task vectors from three of the tasks can improve performance on the fourth, even when no data from the fourth task is used for training. Overall, our experiments with several models, modalities and tasks show that task arithmetic is a simple, efficient and effective way of editing models”, therefore, at least one weight of the pre-trained large language model is updated to match a respective weight of the candidate/edited model having the highest performance).
The reasons of obviousness have been noted in the rejection of Claims 3 and 4 above and applicable herein.
Claim 12 recites substantially the same limitations as Claim 3 in the form of a system, therefore it is rejected under the same rationale.
Claim 13 recites substantially the same limitations as Claim 4 in the form of a system, therefore it is rejected under the same rationale.
Claim 14 recites substantially the same limitations as Claim 5 in the form of a system, therefore it is rejected under the same rationale.
Claim 15 recites substantially the same limitations as Claim 6 in the form of a system, therefore it is rejected under the same rationale.
Claim 16 recites substantially the same limitations as Claim 7 in the form of a system, therefore it is rejected under the same rationale.
11. Claims 8-9 and 17-18 are rejected under 35 U.S.C. 103 as being unpatentable over Panda et al. (hereinafter Panda) (US PG-PUB 20250005370), in view of Ding et al. (hereinafter Ding) (“Sparse Low-rank Adaptation of Pre-trained Language Models”).
Regarding Claim 8, Panda teaches the method of claim 1.
Panda does not explicitly disclose wherein fine-tuning the pre-trained large language model jointly over the first task combination comprises enforcing sparsity by taking proximal steps after each gradient update.
However, Ding teaches wherein fine-tuning the pre-trained large language model jointly over the first task combination comprises enforcing sparsity by taking proximal steps after each gradient update (Ding, Pg. 1, Abstract, “Fine-tuning pre-trained large language mod els in a parameter-efficient manner is widely studied for its effectiveness and efficiency. The popular method of low-rank adaptation (LoRA) offers a notable approach, hypothe sizing that the adaptation process is intrinsi cally low-dimensional. Although LoRA has demonstrated commendable performance, it is implemented with a fixed and unalterable in trinsic rank that might not always be the ideal choice. Recognizing the need for more flexi ble adaptation, we extend the methodology of LoRA to an innovative approach we call sparse low-rank adaptation (SoRA) that enables dy namic adjustments to the intrinsic rank during the adaptation process. We achieve this through the incorporation of a gate unit optimized with proximal gradient method in the training stage, controlling the cardinality of rank under the sparsity of the gate. In the subsequent inference stage, we eliminate the parameter blocks cor responding to the zeroed-out ranks, to reduce each SoRA module back to a concise yet rank optimal LoRA.”, therefore, the pre-trained large language model may be fine-tuned jointly over the first task combination (varying tasks as disclosed by Ding Pg. 8 section 4.6 Efficiency Analysis) by enforcing sparsity by taking proximal steps after each gradient update (proximal gradient method)).
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of claim 1, as disclosed by Panda to include wherein fine-tuning the pre-trained large language model jointly over the first task combination comprises enforcing sparsity by taking proximal steps after each gradient update, as disclosed by Ding. One of ordinary skill in the art would have been motivated to make this modification to enable the use of a proximal gradient method which may efficiently tame parameters by updating sparsely, hence improving model generalization and overall performance (Ding, Pg. 1, Abstract, “Our approach strengthens the representation power of LoRA by initializing it with a higher rank, while efficiently taming a temporarily increased number of parameters via updating in a sparse way. We further intro duce a sparsifying scheduler for SoRA, aiming to examine the impact of the number of non zero parameters on the model’s memorization and generalization. Our experimental results demonstrate that SoRA can outperform other baselines even with 70% retained parameters and 70% training time.”)
Regarding Claim 9, Panda in view of Ding teaches the method of claim 8, wherein sparsity is enforced according to a task-specific diagonal matrix having sparse diagonal entries (Ding, Pg. 2, “In contrast, we eliminate this requirement and instead selectively filter low-rank components by controlling the intermediate diagonal matrix. We detailedly compare SoRA and related methods in Section 3.”, thus, as also shown by Figure 1 and equations 2 & 3 on Pg. 3, sparsity is enforced according to a task-specific diagonal matrix having sparse diagonal entries).
The reasons of obviousness have been noted in the rejection of Claim 8 above and applicable herein.
Claim 17 recites substantially the same limitations as Claim 8 in the form of a system, therefore it is rejected under the same rationale.
Claim 18 recites substantially the same limitations as Claim 9 in the form of a system, therefore it is rejected under the same rationale.
Conclusion
12. Any inquiry concerning this communication or earlier communications from the examiner should be directed to Devika S Maharaj whose telephone number is (571)272-0829. The examiner can normally be reached Monday - Thursday 8:30am - 5:30pm.
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, Alexey Shmatov can be reached at (571)270-3428. 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.
/DEVIKA S MAHARAJ/Examiner, Art Unit 2123
Prosecution Insights