DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 10/15/2025 have been entered and considered. Initialed copies of the PTO-1449 by the Examiner are attached.
Claim Objections
Claim 1 is objected to because of the following informalities:
In claim 1, lines 1-3, the phrase“ An apparatus for executing a machine learning model, comprising: one or more memories; and one or more processors coupled to the one or more memories and configured to: receive input data;” should read “An apparatus for executing a machine learning model, comprising: one or more memories; and one or more processors coupled to the one or more memories, where the one or more processor is .
Appropriate correction is required.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101
Regarding independent claim 1 and dependent claims 2-14
Step 1 Analysis: Claim 1 is directed to an apparatus, which falls within one of the four statutory categories.
Step 2A Prong 1 Analysis: Claim 1 recites, in part, “process the input data using ” and “process the intermediate data ”, as drafted, are elements that, under broadest reasonable interpretation, covers “mathematical concepts” grouping of abstract ideas.
The limitations of ““process the input data to generate intermediate data, the first ML block including a first set of parameters” and “process the intermediate data to generate processed data, including a second set of parameters matching the first set of parameters” is a mere manipulation of the input data using mathematical formulation and does not involve transformation of the data and thus fall under the concept of “mathematical concepts”.
Accordingly, the claim recites an abstract idea.
Step 2A Prong 2 Analysis: The judicial exception is not integrated into a practical application. Particularly, the claim recites the following additional limitations:
“one or more memories; and one or more processors coupled to the one or more memories”
“a machine learning model”, “a first machine learning (ML)” and “a second ML block”
The additional elements of “one or more memories” and “one or more processors coupled to the one or more memories” are recited at a high level of generality, i.e., as a generic processor and memory performing a generic computer function of processing data. This generic processor and memory limitations are no more than mere instructions to apply the exception using generic computer components.
The additional elements of “a machine learning model”, “a first machine learning (ML)” and “a second ML block” are recited at a high level of generality and is therefore a mere generic machine learning model. Note that the first and second block are sub models within the machine learning model. First, the machine learning model is not trained and thus the claim does not recite any details of the training of the machine learning model. Second, the advantage or improvement for implementing the machine learning model is not claimed.
Accordingly, these additional elements do not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. Please see MPEP §2106.04.(a)(2).III.C.
The claim element of “receiving input data” and “output the processed data” impart additional elements, the additional elements merely constitute pre-solution and post-solution activating involving receiving of input data and outputting the processed data. Such extra-solution activities do not integrate the abstract idea into a practical application. Please see MPEP §2106.05(g). Accordingly, these additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claim as a whole is directed to an abstract idea. Please see MPEP §2106.04.(a)(2).III.C.
In view of the of the foregoing, the additional step does not integrate the abstract idea into a practical application.
For all of the foregoing reasons, claim 1 does not comply with the requirements of 35 USC 101.
Accordingly, the dependent claims 2-14 do not provide elements that overcome the deficiencies of the independent claim 1.
Claim 2 recites in part “wherein the intermediate data generated by the first ML block is processed by a third ML block before being processed by the second ML block “ do not overcome the rejection of the parent claim 1 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 3 recites in part “wherein the first ML block and second ML block comprise a same type of ML block “do not overcome the rejection of the parent claim 1 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 4 recites in part “wherein the first ML block and second ML block comprise a feed-forward block” do not overcome the rejection of the parent claim 1 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 5 recites in part “wherein the first set of parameters and second set of parameters comprise parameters for at least one linear layer of the feed-forward block” do not overcome the rejection of the parent claim 4 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 6 recites in part “wherein the feed-forward block is a part of at least one of a transformer block, a conformer block, or a convnext block” do not overcome the rejection of the parent claim 4 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 7 recites in part “” do not overcome the rejection of the parent claim 1 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 8 recites in part “herein the adapter block includes one or more linear layers” do not overcome the rejection of the parent claim 7 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 9 recites in part “wherein parameters of the one or more linear layers of a first adapter block associated with the first ML block differ from parameters of the one or more linear layers of a second adapter block associated with the second ML block” do not overcome the rejection of the parent claim 8 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 10 recites in part “” do not overcome the rejection of the parent claim 8 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 11 recites in part “wherein the adapter block is trained with the first ML block and the second ML block” do not overcome the rejection of the parent claim 7 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 12 recites in part “wherein the adapter block, the first ML block, and the second ML block are trained using on-device training” do not overcome the rejection of the parent claim 7 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 13 recites in part “wherein the input data comprises image data.
” do not overcome the rejection of the parent claim 1 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 14 recites in part “further comprising one or more cameras configured to capture the image data.” do not overcome the rejection of the parent claim 14 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Regarding independent claim 15 and dependent claims 16-20
Step 1 Analysis: Claim 15 is directed to a method, which falls within one of the four statutory categories.
Step 2A Prong 1 Analysis: Claim 15 recites, in part, “processing the input data and “processing the intermediate data as drafted, are elements that, under broadest reasonable interpretation, covers “mathematical concepts” grouping of abstract ideas.
The limitations of “process the input data to generate intermediate data, the first ML block including a first set of parameters” and “process the intermediate data to generate processed data, including a second set of parameters matching the first set of parameters” is a mere manipulation of the input data using mathematical formulation and does not involve transformation of the data and thus fall under the concept of “mathematical concepts”.
Accordingly, the claim recites an abstract idea.
Step 2A Prong 2 Analysis: The judicial exception is not integrated into a practical application. Particularly, the claim recites the following additional limitations:
“a machine learning model”, “a first machine learning (ML)” and “a second ML block”
Note that the first and second blocks are sub models within the machine learning model. The additional element includes generic machine learning model recited at high level of generality without limiting further, in details, on how machine learning model is trained to arrive at such output, these additional elements are recited as a mere attempt to implement the abstract ideas/judicial exceptions using generic machine learning models. Furthermore, the advantage or improvement for implementing the machine learning model is not claimed.
The claim element of “receiving input data” and “output the processed data” impart additional elements, the additional elements merely constitute pre-solution and post-solution activating involving receiving of input data and outputting the processed data. Such extra-solution activities do not integrate the abstract idea into a practical application. Please see MPEP §2106.05(g). Accordingly, these additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claim as a whole is directed to an abstract idea. Please see MPEP §2106.04.(a)(2).III.C.
In view of the of the foregoing, the additional step does not integrate the abstract idea into a practical application.
For all of the foregoing reasons, claim 15 does not comply with the requirements of 35 USC 101.
Accordingly, the dependent claims 16-20 do not provide elements that overcome the deficiencies of the independent claim 15.
Claim 16 recites in part “wherein the intermediate data generated by the first ML block is processed by a third ML block before being processed by the second ML block “ do not overcome the rejection of the parent claim 15 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 17 recites in part “wherein the first ML block and second ML block comprise a same type of ML block “do not overcome the rejection of the parent claim 15 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 18 recites in part “wherein the first ML block and second ML block comprise a feed-forward block” do not overcome the rejection of the parent claim 17 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 19 recites in part “wherein the first set of parameters and second set of parameters comprise parameters for at least one linear layer of the feed-forward block” do not overcome the rejection of the parent claim 18 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim 20 recites in part “wherein the feed-forward block is a part of at least one of a transformer block, a conformer block, or a convnext block” do not overcome the rejection of the parent claim 18 as stated above because the additional elements does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis 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 of this title, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-6 and 13-14 are rejected under 35 U.S.C. 103 as being unpatentable over Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP) in view of Versace et al (Pub No.: US20210216865).
Regarding claim 1, Kobayashi teaches an apparatus for executing a machine learning model (analyze the input contextualization effects of feed-forward (FF) blocks by rendering them in the attention maps as a human-friendly visualization scheme – see abstract), comprising: receive input data (input vector – see left Fig 1); process the input data using a first machine learning (ML) block (for e.g. normalization layer LN1 – see left Fig 1) to generate intermediate data (output data from normalization layer, LN1 being fed into the multi-head attention layer and then into the second normalization layer, LN2 – see left Fig 1), the first ML block including a first set of parameters (Note that the layer normalization first normalizes the input representation, then multiplies a weight vector γ element-wise – see section 6.2, [p][001] and left Fig 1); process the intermediate data using a second ML block (second normalization layer, LN2 – see left Fig 1) to generate processed data (output from second normalization layer, LN2 – see left Fig 1), the second ML block including a second set of parameters matching the first set of parameters (note that both are normalization layers – see left Fig 1); and output the processed data (final output layer with y3 – see Fig 1).
Kobayashi does not explicitly teach one or more memories; and one or more processors coupled to the one or more memories.
Versace explicitly teaches one or more memories (non-volatile memory – see [p][0056]); and one or more processors (see [p][0056]) coupled to the one or more memories (see [p][0056]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kobayashi of an apparatus for executing a machine learning model, comprising:
one or more memories; and one or more processors coupled to the one or more memories and configured to: receive input data with the teachings of Versace one or more memories; and one or more processors coupled to the one or more memories.
Wherein having Kobayashi one or more memories; and one or more processors coupled to the one or more memories.
. The motivation behind the modification would have been for input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis thus reducing training time since both Kobayashi and Versace relates to training machine learning models, wherein Kobayashi performs input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis while Versace uses much less data to build robust networks, has dramatically shorter training time (Please see Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP), section 7, [p][001] and Versace et al (Pub No.: US20210216865), see Abstract).
Regarding claim 2, Kobayashi in view of Versace teach the apparatus of claim 1, Kobayashi explicitly teaches wherein the intermediate data generated by the first ML block is processed by a third ML block (for e.g. multi-head attention – left see Fig 1) before being processed by the second ML block (note that the attention head is place between LN1 and LN2 – see left Fig 1).
Regarding claim 3, Kobayashi in view of Versace teach the apparatus of claim 1, Kobayashi explicitly teaches wherein the first ML block and second ML block comprise a same type of ML block (note that both are normalization layers – see left Fig 1).
Regarding claim 4, Kobayashi in view of Versace teach the apparatus of claim 3, Kobayashi explicitly teaches wherein the first ML block and second ML block comprise a feed-forward block (note that both ATB and FFB block contains feed forward connections, RES1 and RES2 – see let Fig 1).
Regarding claim 5, Kobayashi in view of Versace teach the apparatus of claim 4, Kobayashi explicitly teaches wherein the first set of parameters and second set of parameters comprise parameters for at least one linear layer of the feed-forward block (note that both ATB and FFB block contains feed forward connections, RES1 and RES2 – see let Fig 1).
Regarding claim 6, Kobayashi in view of Versace teach the apparatus of claim 4, Kobayashi explicitly teaches wherein the feed-forward block is a part of at least one of a transformer block (note that Fig 1 is a transformer - see Fig 1), a conformer block, or a convnext block.
Regarding claim 13, Kobayashi in view of Versace teach the apparatus of claim 1, Kobayashi does not explicitly teach wherein the input data comprises image data.
However, Versace explicitly teaches wherein the input data comprises image data (acquiring a first image with a camera – see [p][0029]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kobayashi of an apparatus for executing a machine learning model, comprising:
one or more memories; and one or more processors coupled to the one or more memories and configured to: receive input data with the teachings of Versace wherein the input data comprises image data
Wherein having Kobayashi wherein the input data comprises image data.
. The motivation behind the modification would have been for input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis thus reducing training time since both Kobayashi and Versace relates to training machine learning models, wherein Kobayashi performs input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis while Versace uses much less data to build robust networks, has dramatically shorter training time (Please see Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP), section 7, [p][001] and Versace et al (Pub No.: US20210216865), see Abstract).
Regarding claim 14, Kobayashi in view of Versace teach the apparatus of claim 13, Kobayashi does not explicitly teach further comprising one or more cameras configured to capture the image data.
However, Versace explicitly teaches further comprising one or more cameras configured to capture the image data (acquiring a first image with a camera – see [p][0029]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kobayashi of an apparatus for executing a machine learning model, comprising:
one or more memories; and one or more processors coupled to the one or more memories and configured to: receive input data with the teachings of Versace further comprising one or more cameras configured to capture the image data.
Wherein having Kobayashi further comprising one or more cameras configured to capture the image data.
. The motivation behind the modification would have been for input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis thus reducing training time since both Kobayashi and Versace relates to training machine learning models, wherein Kobayashi performs input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis while Versace uses much less data to build robust networks, has dramatically shorter training time (Please see Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP), section 7, [p][001] and Versace et al (Pub No.: US20210216865), see Abstract).
Claims 7-9 and 11 are rejected under 35 U.S.C. 103 as being unpatentable over Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP) in view of Versace et al (Pub No.: US20210216865) as applied to claim 1 further in view of Dong et al (NPL titled: Efficient Adaptation of Large Vision Transformer via Adapter Re-Composing)
Regarding claim 7, Kobayashi in view of Versace does not explicitly teaches the apparatus of claim 4, wherein the feed-forward block includes an adapter block.
Dong explicitly teaches wherein the feed-forward block includes an adapter block (see section 3.1 – where described a transformer which includes a feed forward network and a novel approach to create a unified linear space across different adapters to enhance parameter efficiency and adaptation performance – see section 3.1 and 3.2).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kobayashi in view of Versace of an apparatus for executing a machine learning model, comprising: one or more memories; and one or more processors coupled to the one or more memories and configured to: receive input data with the teachings of Dong wherein the feed-forward block includes an adapter block.
Wherein having Kobayashi wherein the feed-forward block includes an adapter block.
. The motivation behind the modification would have been for input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis thus reducing the number of new parameters while maintaining satisfactory performance since both Kobayashi and Dong relates to training machine learning models, wherein Kobayashi performs input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis while Dong effectively re-compose layer-adaptive adapters which allows for further reduce the number of new parameters while maintaining satisfactory performance (Please see Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP), section 7, [p][001] and Dong et al (NPL titled: Efficient Adaptation of Large Vision Transformer via Adapter Re-Composing), see Abstract).
Regarding claim 8, Kobayashi in view of Versace does not explicitly teaches the apparatus of claim 7, wherein the adapter block includes one or more linear layers.
Dong explicitly teaches wherein the adapter block includes one or more linear layers (linear down-projection, layer-specific re-scaling coefficients – see section 3.2).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kobayashi in view of Versace of an apparatus for executing a machine learning model, comprising: one or more memories; and one or more processors coupled to the one or more memories and configured to: receive input data with the teachings of Dong wherein the adapter block includes one or more linear layers.
Wherein having Kobayashi wherein the adapter block includes one or more linear layers.
. The motivation behind the modification would have been for input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis thus reducing the number of new parameters while maintaining satisfactory performance since both Kobayashi and Dong relates to training machine learning models, wherein Kobayashi performs input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis while Dong effectively re-compose layer-adaptive adapters which allows for further reduce the number of new parameters while maintaining satisfactory performance (Please see Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP), section 7, [p][001] and Dong et al (NPL titled: Efficient Adaptation of Large Vision Transformer via Adapter Re-Composing), see Abstract).
Regarding claim 9, Kobayashi in view of Versace does not explicitly teaches the apparatus of claim 8, wherein parameters of the one or more linear layers of a first adapter block associated with the first ML block differ from parameters of the one or more linear layers of a second adapter block associated with the second ML block.
Dong explicitly teaches wherein parameters of the one or more linear layers of a first adapter block associated with the first ML block differ from parameters of the one or more linear layers of a second adapter block associated with the second ML block (see section 3.2)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kobayashi in view of Versace of an apparatus for executing a machine learning model, comprising: one or more memories; and one or more processors coupled to the one or more memories and configured to: receive input data with the teachings of Dong wherein parameters of the one or more linear layers of a first adapter block associated with the first ML block differ from parameters of the one or more linear layers of a second adapter block associated with the second ML block.
Wherein having Kobayashi wherein parameters of the one or more linear layers of a first adapter block associated with the first ML block differ from parameters of the one or more linear layers of a second adapter block associated with the second ML block.
. The motivation behind the modification would have been for input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis thus reducing the number of new parameters while maintaining satisfactory performance since both Kobayashi and Dong relates to training machine learning models, wherein Kobayashi performs input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis while Dong effectively re-compose layer-adaptive adapters which allows for further reduce the number of new parameters while maintaining satisfactory performance (Please see Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP), section 7, [p][001] and Dong et al (NPL titled: Efficient Adaptation of Large Vision Transformer via Adapter Re-Composing), see Abstract).
Regarding claim 11, Kobayashi in view of Versace does not explicitly teaches the apparatus of claim 7, wherein the adapter block is trained with the first ML block and the second ML block.
Dong explicitly teaches wherein the adapter block is trained with the first ML block and the second ML block (see section 4.3).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kobayashi in view of Versace of an apparatus for executing a machine learning model, comprising: one or more memories; and one or more processors coupled to the one or more memories and configured to: receive input data with the teachings of Dong wherein the adapter block is trained with the first ML block and the second ML block.
Wherein having Kobayashi wherein the adapter block is trained with the first ML block and the second ML block.
. The motivation behind the modification would have been for input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis thus reducing the number of new parameters while maintaining satisfactory performance since both Kobayashi and Dong relates to training machine learning models, wherein Kobayashi performs input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis while Dong effectively re-compose layer-adaptive adapters which allows for further reduce the number of new parameters while maintaining satisfactory performance (Please see Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP), section 7, [p][001] and Dong et al (NPL titled: Efficient Adaptation of Large Vision Transformer via Adapter Re-Composing), see Abstract).
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP) in view of Versace et al (Pub No.: US20210216865) as applied to claim 1 further in view of Dong et al (NPL titled: Efficient Adaptation of Large Vision Transformer via Adapter Re-Composing) as applied to claim 8 further in view of Lei et al (Conditional Adapters: Parameter-efficient Transfer Learning with Fast Inference).
Regarding claim 10, Kobayashi in view of Versace and Dong does not explicitly teaches the apparatus of claim 8, wherein the adapter block further comprises a layer normalization layer, an activation function, and a sigmoid function.
Lei explicitly teaches wherein the adapter block further comprises a layer normalization layer (see section 3.1 subsection Learned Rounder), an activation function (see section 2, [p][006]), and a sigmoid function (sigmoid activation function – see section 5, [p][005]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kobayashi in view of Versace and Dong of an apparatus for executing a machine learning model, comprising: one or more memories; and one or more processors coupled to the one or more memories and configured to: receive input data with the teachings of Lei wherein the adapter block further comprises a layer normalization layer, an activation function, and a sigmoid function.
Wherein having Kobayashi wherein the adapter block further comprises a layer normalization layer, an activation function, and a sigmoid function.
. The motivation behind the modification would have been for input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis to enable a new way of balancing speed and accuracy using conditional computation, since both Kobayashi and Lei relates to training machine learning models, wherein Kobayashi performs input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis while Lei enable a new way of balancing speed and accuracy using conditional computation (Please see Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP), section 7, [p][001] and Lei et al (Conditional Adapters: Parameter-efficient Transfer Learning with Fast Inference), see abstract).
Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP) in view of Versace et al (Pub No.: US20210216865) as applied to claim 1 further in view of Dong et al (NPL titled: Efficient Adaptation of Large Vision Transformer via Adapter Re-Composing) as applied to claim 7 further in view of Li et al (Pub No.: US20230316085).
Regarding claim 12, Kobayashi in view of Versace does not explicitly teaches the apparatus of claim 7, wherein the adapter block, the first ML block, and the second ML block are trained using on-device training.
Li explicitly teaches wherein the adapter block, the first ML block, and the second ML block are trained using on-device training (the adapter type/number selection algorithm can easily apply to the multi-target case by wrapping switched layers and learning all of α, β, γ on device – see [p][0140]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kobayashi in view of Versace and Dong of an apparatus for executing a machine learning model, comprising: one or more memories; and one or more processors coupled to the one or more memories and configured to: receive input data with the teachings of Li wherein the adapter block, the first ML block, and the second ML block are trained using on-device training.
Wherein having Kobayashi wherein the adapter block, the first ML block, and the second ML block are trained using on-device training.
. The motivation behind the modification would have been for inputting contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis and adding at least one adapter module to the pre-trained machine learning model to create a local machine learning model, since both Kobayashi and Li relates to training machine learning models, wherein Kobayashi performs input contextualization through the lens of a refined attention map by leveraging the existing norm-based analysis while Li adds at least one adapter module to the pre-trained machine learning model to create a local machine learning model, (Please see Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP), section 7, [p][001] and Li et al (Pub No.: US20230316085), see [p][0006]).
Claim Rejections - 35 USC § 102
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 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)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(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.
Claims 15-20 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Kobayashi et al (NPL titled: ANALYZING FEED-FORWARD BLOCKS IN TRANSFORMERS THROUGH THE LENS OF ATTENTION MAP)
Regarding independent claim 15, Kobayashi teaches a method for executing a machine learning model (analyze the input contextualization effects of feed-forward (FF) blocks by rendering them in the attention maps as a human-friendly visualization scheme – see abstract),comprising: receiving input data (input vector – see left Fig 1); process the input data using a first machine learning (ML) block (for e.g. normalization layer LN1 – see left Fig 1) to generate intermediate data (output data from normalization layer, LN1 being fed into the multi-head attention layer and then into the second normalization layer, LN2 – see left Fig 1), the first ML block including a first set of parameters (Note that the layer normalization first normalizes the input representation, then multiplies a weight vector γ element-wise – see section 6.2, [p][001] and left Fig 1); processing the intermediate data using a second ML block (second normalization layer, LN2 – see left Fig 1) to generate processed data (output from second normalization layer, LN2 – see left Fig 1), the second ML block including a second set of parameters matching the first set of parameters (note that both are normalization layers – see left Fig 1); and outputting the processed data (final output layer with y3 – see Fig 1).
Regarding claim 16, Kobayashi explicitly teaches the method of claim 15, wherein the intermediate data generated by the first ML block is processed by a third ML block (for e.g. multi-head attention – left see Fig 1) before being processed by the second ML block (note that the attention head is place between LN1 and LN2 – see left Fig 1).
Regarding claim 17, Kobayashi explicitly teaches the method of claim 15, wherein the first ML block and second ML block comprise a same type of ML block (note that both are normalization layers – see left Fig 1).
Regarding claim 18, Kobayashi explicitly teaches the method of claim 17, wherein the first ML block and second ML block comprise a feed-forward block (note that both ATB and FFB block contains feed forward connections, RES1 and RES2 – see let Fig 1).
Regarding claim 19, Kobayashi explicitly teaches the method of claim 18, wherein the first set of parameters and second set of parameters comprise parameters for at least one linear layer of the feed-forward block (note that both ATB and FFB block contains feed forward connections, RES1 and RES2 – see let Fig 1).
Regarding claim 17, Kobayashi explicitly teaches the method of claim 18, wherein the feed-forward block is a part of at least one of a transformer block (note that Fig 1 is a transformer - see left Fig 1), a conformer block, or a convnext block.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Lee et al (Pub No.: 20230368499) discloses a method of extracting image features based on a vision transformer, a method of performing embedding on an input image in units of patches and extracting visual features through global attention. An apparatus for extracting an image feature based on a vision transformer according to an embodiment of the disclosure includes a memory configured to store data and a processor configured to control the memory, wherein the processor is configured to perform embedding on multi-patches for an input image, extract feature maps for the embedding multi-patches, perform transformer encoding based on a neural network using the extracted feature maps, extract a feature of the input image through a final feature map extracted through the transformer encoding, and wherein the patches have different sizes.
Li et al (Pub No.: 20230103997) discloses a vision transformer includes L layers, and H attention heads in each layer. An h′ of the attention heads include an attention mask added before a Softmax operation, and an h of the attention heads include unmasked attention heads in which H=h′+h. Each attention mask multiplies a Query vector and a Key vector for form element-wise products. At least one attention mask is a hard mask that selects closest neighbors of a patch and ignores patches further away than the closest neighbors of the patch. Alternatively, at least one attention mask includes a soft mask that multiplies weights of closest neighbors of a patch by a magnification factor and passes weights of patches that are further away than the closest neighbors of the patch. A learnable bias α may be added to diagonal elements of the at least one attention map.
Jiao et al (Pub No.: 20220067533) discloses a transformer-based neural network includes at least one mask attention network (MAN). The MAN computes an original attention data structure that expresses influence between pairs of data items in a sequence of data items. The MAN then modifies the original data structure by mask values in a mask data structure, to produce a modified attention data structure. Compared to the original attention data structure, the modified attention data structure better accounts for the influence of neighboring data items in the sequence of data items, given a particular data item under consideration. The mask data structure used by the MAN can have static and/or machine-trained mask values. In one implementation, the transformer-based neural network includes at least one MAN in combination with at least one other attention network that does not use a mask data structure, and at least one feed-forward neural network.
Yin et al (Pub No.: 20230073835) discloses a method includes accessing a batch B of a plurality of images, wherein each image in the batch is part of a training set of images used to train a vision transformer comprising a plurality of attention heads. The method further includes determining, for each attention head A, a similarity between (1) the output of the attention head evaluated using each image in the batch and the (2) output of each attention head evaluated using each image in the batch. The method further includes determining, based on the determined similarities, an importance score for each attention head; and pruning, based on the importance scores, one or more attention heads from the vision transformer.
.
Inquiries
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ANDRAE S ALLISON whose telephone number is (571)270-1052. The examiner can normally be reached on Monday-Friday 9am-5pm 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, Chineyere Wills-Burns, can be reached on (571) 272-9752. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/ANDRAE S ALLISON/Primary Examiner, Art Unit 2673
April 26, 2026
Prosecution Insights