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 .
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1, in step 1 of the 101 analysis set forth in MPEP 2106, the claim recites A computer-implemented method for accelerating a trained machine learning model, the method comprising:. The claim recites a method. A method is one of the four statutory categories of invention.
In Step 2A, Prong 1 of the 101 analysis set forth in MPEP 2106, the examiner has determined that the following limitations recite a process that, under broadest reasonable interpretation, covers a mental process or mathematical concept but for the recitation of generic computer components:
parsing the trained machine learning model to identify one or more layers of the trained machine learning model and, for each layer included in the one or more layers, one or more corresponding compression techniques that can be applied to compress the layer; (i.e., the broadest reasonable interpretation includes a step of observation, evaluation, and judgement and could be performed mentally or with pen and paper like identifying which layers can be pruned, which is either a mental process of observation/evaluation/judgement (MPEP 2106)).
performing,…one or more iterative operations to select, for each layer included in the one or more layers, a compression technique included in the one or more corresponding compression techniques and values of one or more parameters associated with the compression technique; (i.e., the broadest reasonable interpretation includes a step of observation, evaluation, and judgement and could be performed mentally or with pen and paper like determining which type of pruning to use, which is either a mental process of observation/evaluation/judgement (MPEP 2106)).
and compressing each layer included in the one or more layers using the compression technique that is selected for the layer and the values of the one or more parameters associated with the compression technique to generate a compressed trained machine learning model. (i.e., the broadest reasonable interpretation includes a step of observation, evaluation, and judgement and could be performed mentally or with pen and paper like performing the pruning like removing noisy neurons, which is either a mental process of observation/evaluation/judgement (MPEP 2106)).
If the claim limitations, under their broadest reasonable interpretation, covers activities classified under Mental processes: concepts performed in the human mind (including observation, evaluation, judgement, or opinion) (see MPEP 2106.04(a)(2), subsection (III)) or Mathematical concepts: mathematical relationships, mathematical formulas or equations, or mathematical calculations (see MPEP 2106.04(a)(2), subsection (I)). Accordingly, the claim recites an abstract idea.
In Step 2A, Prong 2 of the 101 analysis, set forth in MPEP 2106, the examiner has determined that the following additional elements do not integrate this judicial exception into a practical application:
…based on a hardware device on which the trained machine learning model is intended to execute,…(i.e., the generic computer components recited in this limitation merely add the words “apply it”, or an equivalent, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f))).
Since the claim does not contain any other additional elements, that amount to integration into a practical application, the claim is directed to an abstract idea.
In Step 2B of the 101 analysis set forth in the 2019 PEG, the examiner has determined that the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception:
Regarding limitation (IV), under the broadest reasonable interpretation, merely recite steps that apply generic computer components as a tool to perform judicial exceptions, which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Considering additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
Regarding claim 2, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 2 recites further comprising performing one or more quantization operations on the compressed trained machine learning model to generate a quantized trained machine learning model. Under the broadest reasonable interpretation, the limitations recite performing a quantization technique which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 2 does not solve the deficiencies of claim 1.
Regarding claim 3, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 3 recites wherein the one or more iterative operations comprise one or more reinforcement learning operations. Under the broadest reasonable interpretation, the limitations merely recite steps that apply generic reinforcement learning, which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Therefore, claim 3 does not solve the deficiencies of claim 1.
Regarding claim 4, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 4 recites where the compression technique included in the one or more corresponding compression techniques comprises at least one of a pruning technique, a decomposition technique, or an approximation technique. Under the broadest reasonable interpretation, the limitations recite pruning or removing values which is a step of observation, evaluation, and judgement which can be performed mentally or with pen and paper. The steps of observation, evaluation, and judgement are mental processes. Therefore, claim 4 does not solve the deficiencies of claim 1.
Regarding claim 5, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 5 recites wherein the one or more iterative operations are further based on at least one of a predefined accuracy constraint or a predefined execution speed constraint. Under the broadest reasonable interpretation, the limitations recite selecting a type of pruning based on an accuracy which is a step of observation, evaluation, and judgement which can be performed mentally or with pen and paper. The steps of observation, evaluation, and judgement are mental processes. Therefore, claim 5 does not solve the deficiencies of claim 1.
Regarding claim 6, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 6 recites further comprising, in response to determining that the compressed trained machine learning model does not satisfy a predefined accuracy constraint, performing one or more operations to re-train the compressed trained machine learning model. Under the broadest reasonable interpretation, the limitations merely recite steps that apply generic retraining of a model, which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Therefore, claim 6 does not solve the deficiencies of claim 1.
Regarding claim 7, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 7 recites further comprising performing one or more operations to fuse at least two layers included in the one or more layers to generate a fused layer. Under the broadest reasonable interpretation, the limitations recite combining two layers which is a step of observation, evaluation, and judgement which can be performed mentally or with pen and paper. The steps of observation, evaluation, and judgement are mental processes. Therefore, claim 7 does not solve the deficiencies of claim 1.
Regarding claim 8, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 8 recites further comprising updating, based on user input, the one or more corresponding compression techniques for at least one layer included in the one or more layers. Under the broadest reasonable interpretation, the limitations recite changing a parameter of a pruning technique which is a step of observation, evaluation, and judgement which can be performed mentally or with pen and paper. The steps of observation, evaluation, and judgement are mental processes. Therefore, claim 8 does not solve the deficiencies of claim 1.
Regarding claim 9, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 9 recites further comprising performing one or more operations to convert the compressed trained machine learning model to a binary format that is executable via the hardware device. Under the broadest reasonable interpretation, the limitations recite performing a binary conversion calculation which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 9 does not solve the deficiencies of claim 1.
Regarding claim 10, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 10 recites wherein the trained machine learning model comprises a trained artificial neural network. Under the broadest reasonable interpretation, the limitations merely recite steps that apply a generic neural network, which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Therefore, claim 10 does not solve the deficiencies of claim 1.
Regarding claim 11, in step 1 of the 101 analysis set forth in MPEP 2106, the claim recites One or more non-transitory computer readable media. The claim recites a non-transitory computer readable medium which is interpreted as an article of manufacture. An article of manufacture is one of the four statutory categories of invention. For the Step 2A/2B analyses, since claim 11 is similar to claim 1 it is rejected under the same rationales as claim 1.
The additional limitation below fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception.
One or more non-transitory computer readable media including instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of:… (i.e., the generic computer components recited in this limitation merely add the words “apply it”, or an equivalent, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f))).
Considering additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
Regarding claims 12-14, the claims are similar to claims 2, 4, and 5 and are rejected under the same rationales.
Regarding claim 15, it is dependent upon claim 11 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 15 recites where the compression technique included in the one or more corresponding compression techniques comprises at least one of a global pruning technique, a local pruning technique, a filter pruning via geometric median (FPGM) technique, a structured pruning technique, an unstructured pruning technique, an Energy Threshold (QR) technique, a Nystromformer technique, a Tucker tensor decomposition technique, a principle component analysis (PCA) decomposition technique, or a canonical polyadic tensor decomposition (CPD) technique. Under the broadest reasonable interpretation, the limitations merely recite steps that apply a generic compression technique, which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Therefore, claim 15 does not solve the deficiencies of claim 11.
Regarding claims 16-18, the claims are similar to claims 6-7 and 9 and are rejected under the same rationales.
Regarding claim 19, it is dependent upon claim 11 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 19 recites wherein the one or more iterative operations are further based on a predefined constraint on a size of the compressed trained machine learning model. Under the broadest reasonable interpretation, the limitations recite selecting the type of pruning based on the size of a model which is a step of observation, evaluation, and judgement which can be performed mentally or with pen and paper. The steps of observation, evaluation, and judgement are mental processes. Therefore, claim 19 does not solve the deficiencies of claim 11.
Regarding claim 20, in step 1 of the 101 analysis set forth in MPEP 2106, the claim recites A system comprising: one or more memories storing instructions; and one or more processors. The claim recites system with hardware components which is interpreted as a machine. A machine is one of the four statutory categories of invention. For the Step 2A/2B analyses, since claim 20 is similar to claim 1 it is rejected under the same rationales as claim 1.
The additional limitation below fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception.
A system comprising: one or more memories storing instructions; and one or more processors coupled to the one or more memories that, when executing the instructions, perform the steps of:… (i.e., the generic computer components recited in this limitation merely add the words “apply it”, or an equivalent, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f))).
Considering additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
Claim Rejections - 35 USC § 102
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 1-5, 8, 10-15, and 19-20 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Zhan, et al., Non-Patent Literature “Deep Model Compression Via Two-Stage Deep Reinforcement Learning” (“Zhan”).
Regarding claim 1, Zhan discloses:
A computer-implemented method for accelerating a trained machine learning model, (Zhan, pg. 2, “In this paper, we propose to develop a novel two-stage DRL framework for deep model compression [method for accelerating a trained machine learning model,].”, and Zhan, pg. 9, “All experiments were performed using TensorFlow, allowing for automatic differentiation through the gradient updates [1], on 8 NVIDIA Tesla K80 GPUs [A computer-implemented]”).
the method comprising: parsing the trained machine learning model to identify one or more layers of the trained machine learning model and, for each layer included in the one or more layers, one or more corresponding compression techniques that can be applied to compress the layer; (Zhan, pg. 2, “In particular, the proposed framework integrates layer-wise [the method comprising: parsing the trained machine learning model to identify one or more layers of the trained machine learning model and,] pruning rate learning based on testing accuracy and FLOPs, element-wise variational pruning, and per-layer bits representation learning. In the pruning stage, we first conduct channel pruning that will prune the input channel dimension (i.e., C dimension) with minimized accumulated error in feature maps with the obtained per-layer pruning rate [for each layer included in the one or more layers, one or more corresponding compression techniques that can be applied to compress the layer;].”).
performing, based on a hardware device on which the trained machine learning model is intended to execute, one or more iterative operations to select, for each layer included in the one or more layers, a compression technique included in the one or more corresponding compression techniques and values of one or more parameters associated with the compression technique; (Zhan, pg. 4, “a continuous reinforcement learning control strategy [performing…one or more iterative operations to select, for each layer included in the one or more layers,] is needed to get a more stabilized scalar continuous action space, which can be represented as at = {prt|prt ∈ [prh,prl]}, where prl and prh are the lowest and highest and pruning rates, respectively [a compression technique included in the one or more corresponding compression techniques and values of one or more parameters associated with the compression technique;]. The compression rate in each layer is taken as a replacement of high-dimensional discrete masks at each weight of the kernels.”, and Zhan, abstract, “For example, employing deep neural networks on mobile systems requires the design of accurate yet fast CNN for low latency in classification and object detection [based on a hardware device on which the trained machine learning model is intended to execute].”).
and compressing each layer included in the one or more layers using the compression technique that is selected for the layer and the values of the one or more parameters associated with the compression technique to generate a compressed trained machine learning model. (Zhan, pg. 2, “In this paper, we propose to develop a novel two-stage DRL framework for deep model compression [to generate a compressed trained machine learning model.]. In particular, the proposed framework integrates layer-wise pruning rate learning based on testing accuracy and FLOPs, element-wise variational pruning, and per-layer bits representation learning [and compressing each layer included in the one or more layers using the compression technique that is selected for the layer and the values of the one or more parameters associated with the compression technique].”).
Regarding claim 2, Zhan discloses the computer-implemented method of claim 1. Zhan further discloses further comprising performing one or more quantization operations on the compressed trained machine learning model to generate a quantized trained machine learning model. (Zhan, abstract, “The first stage of compression, i.e., pruning, is achieved via exploiting deep reinforcement learning (DRL) to co learn the accuracy and the FLOPs updated after layer-wise channel pruning and element-wise variational pruning via information dropout. The second stage, i.e., quantization, is achieved via a similar DRL approach but focuses on obtaining the optimal bits representation for individual layers [further comprising performing one or more quantization operations on the compressed trained machine learning model to generate a quantized trained machine learning model.].”).
Regarding claim 3, Zhan discloses the computer-implemented method of claim 1. Zhan further discloses wherein the one or more iterative operations comprise one or more reinforcement learning operations. (Zhan, pg. 4, “a continuous reinforcement learning control strategy [wherein the one or more iterative operations comprise one or more reinforcement learning operations.] is needed to get a more stabilized scalar continuous action space, which can be represented as at = {prt|prt ∈ [prh,prl]}, where prl and prh are the lowest and highest and pruning rates, respectively.”).
Regarding claim 4, Zhan discloses the computer-implemented method of claim 1. Zhan further discloses where the compression technique included in the one or more corresponding compression techniques comprises at least one of a pruning technique, a decomposition technique, or an approximation technique. (Zhan, abstract, “The first stage of compression, i.e., pruning, is achieved via exploiting deep reinforcement learning (DRL) to co learn the accuracy and the FLOPs updated after layer-wise channel pruning and element-wise variational pruning via information dropout [where the compression technique included in the one or more corresponding compression techniques comprises at least one of a pruning technique].”).
Regarding claim 5, Zhan discloses the computer-implemented method of claim 1. Zhan further discloses wherein the one or more iterative operations are further based on at least one of a predefined accuracy constraint or a predefined execution speed constraint. (Zhan, pg. 2, “In this paper, we propose to develop a novel two-stage DRL framework for deep model compression. In particular, the proposed framework integrates layer-wise pruning rate learning based on testing accuracy and FLOPs [wherein the one or more iterative operations are further based on at least one of a predefined accuracy constraint or a predefined execution speed constraint.], element-wise variational pruning, and per-layer bits representation learning.”).
Regarding claim 8, Zhan discloses the computer-implemented method of claim 1. Zhan further discloses further comprising updating, based on user input, the one or more corresponding compression techniques for at least one layer included in the one or more layers. (Zhan, pg. 7, “The C-dimension channel pruning can be formulated as: C arg min β,W 1 2N Y − i=1 2 βXiWT i F +λ β 1 (4) subject to β 0 ≤ pr ×C Wi F =1,∀i, where pr is the pruning rate, Xi and Y are the input volume and the output volume in each layer, Wi is the weights, β is the coefficient vector of length C for channel selection, and λ is a positive weight to be selected by users [further comprising updating, based on user input, the one or more corresponding compression techniques for at least one layer included in the one or more layers.].”).
Regarding claim 10, Zhan discloses the computer-implemented method of claim 1. Zhan further discloses wherein the trained machine learning model comprises a trained artificial neural network. (Zhan, abstract, “For example, employing deep neural networks [wherein the trained machine learning model comprises a trained artificial neural network.] on mobile systems requires the design of accurate yet fast CNN for low latency in classification and object detection.”).
Regarding claim 11, the claim is similar to claim 1 and rejected under the same rationales. Zhan further discloses the additional limitations of One or more non-transitory computer readable media including instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of: (Zhan, pg. 9, “All experiments were performed using TensorFlow, allowing for automatic differentiation through the gradient updates [1], on 8 NVIDIA Tesla K80 GPUs [One or more non-transitory computer readable media including instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of:]”).
Regarding claims 12-14, the claims are similar to claims 2, 4, and 5 and are rejected under the same rationales.
Regarding claim 15, Zhan discloses the one or more non-transitory computer readable media of claim 11. Zhan further discloses where the compression technique included in the one or more corresponding compression techniques comprises at least one of a global pruning technique, a local pruning technique, a filter pruning via geometric median (FPGM) technique, a structured pruning technique, an unstructured pruning technique, an Energy Threshold (QR) technique, a Nystromformer technique, a Tucker tensor decomposition technique, a principle component analysis (PCA) decomposition technique, or a canonical polyadic tensor decomposition (CPD) technique. (Zhan, abstract, “The first stage of compression, i.e., pruning, is achieved via exploiting deep reinforcement learning (DRL) to co learn the accuracy and the FLOPs updated after layer-wise channel pruning and element-wise variational pruning via information dropout [where the compression technique included in the one or more corresponding compression techniques comprises at least one of…a structured pruning technique, an unstructured pruning technique].”).
Regarding claim 19, Zhan discloses the one or more non-transitory computer readable media of claim 11. Zhan further discloses wherein the one or more iterative operations are further based on a predefined constraint on a size of the compressed trained machine learning model. (Zhan, pg. 5, “In the proposed model compression method, we learn the Pareto front of a set of models with two-dimensional outputs (model size and accuracy) [are further based on a predefined constraint on a size of the compressed trained machine learning model.] such that at least one output is better than (or at least as good as) all other outputs. We adopt a popular asynchronous actor critic [22] RL framework to compress [wherein the one or more iterative operations] a pre-trained network in each layer sequentially.”).
Regarding claim 20, the claim is similar to claim 1 and rejected under the same rationales. Zhan further discloses the additional limitations of A system comprising: one or more memories storing instructions; and one or more processors coupled to the one or more memories that, when executing the instructions, perform the steps of: (Zhan, pg. 9, “All experiments were performed using TensorFlow, allowing for automatic differentiation through the gradient updates [1], on 8 NVIDIA Tesla K80 GPUs [A system comprising: one or more memories storing instructions; and one or more processors coupled to the one or more memories that, when executing the instructions, perform the steps of:]”).
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 6 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Zhan, et al., Non-Patent Literature “Deep Model Compression Via Two-Stage Deep Reinforcement Learning” (“Zhan”) in view of Chandakkar, et al., Non-Patent Literature “Strategies for Re-training a Pruned Neural Network in an Edge Computing Paradigm” (“Chandakkar”).
Regarding claim 6, Zhan discloses the computer-implemented method of claim 1. While Zhan discloses a system that compresses a neural network on a per-layer basis, Zhan does not explicitly teach further comprising, in response to determining that the compressed trained machine learning model does not satisfy a predefined accuracy constraint, performing one or more operations to re-train the compressed trained machine learning model.
Chandakkar teaches further comprising, in response to determining that the compressed trained machine learning model does not satisfy a predefined accuracy constraint, performing one or more operations to re-train the compressed trained machine learning model. (Chandakkar, pg. 246 col. 1, “We train the DNN to convergence and apply global pruning with a user-defined threshold T. Then we re-train the pruned network till convergence [further comprising, in response to determining that the compressed trained machine learning model does not satisfy a predefined accuracy constraint,]. During re-training, no additional modification/pruning to weights is performed i.e. the indices of the pruned weight elements stay constant during re-training [performing one or more operations to re-train the compressed trained machine learning model.]. Thus this method is the simplest to implement.”).
Zhan and Chandakkar are both in the same field of endeavor (i.e. model compression). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Zhan and Chandakkar to teach the above limitation(s). The motivation for doing so is that retraining a model improves the model’s performance to changing data (cf. Chandakkar, abstract, “To this end, weight pruning for DNNs has been proposed to reduce their storage footprint by an order of magnitude. However, it is yet unclear as to how to update/re-train DNNs once they are deployed on mobile devices. In this paper, we introduce the concept of re-training of pruned networks that should aid personalization of smart devices as well as increase their fault tolerance.”).
Regarding claim 16, the claim is similar to claim 6 and is rejected under the same rationales.
Claims 7 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Zhan, et al., Non-Patent Literature “Deep Model Compression Via Two-Stage Deep Reinforcement Learning” (“Zhan”) in view of O’Neill, et al., Non-Patent Literature “Layer-Wise Neural Network Compression via Layer Fusion” (“O’Neill”).
Regarding claim 7, Zhan discloses the computer-implemented method of claim 1. While Zhan discloses a system that compresses a neural network on a per-layer basis, Zhan does not explicitly teach further comprising performing one or more operations to fuse at least two layers included in the one or more layers to generate a fused layer.
O’Neill teaches further comprising performing one or more operations to fuse at least two layers included in the one or more layers to generate a fused layer. (O’Neill, abstract, “This paper proposes layer fusion- a model compression technique that discovers which weights to combine and then fuses weights of similar fully-connected, convolutional and attention layers [further comprising performing one or more operations to fuse at least two layers included in the one or more layers to generate a fused layer.].”).
Zhan and O’Neill are both in the same field of endeavor (i.e. model compression). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Zhan and O’Neill to teach the above limitation(s). The motivation for doing so is that fusing layers together reduces the computational overhead of the model (cf. O’Neill, abstract, “Layer fusion can significantly reduce the number of layers of the original network with little additional computation overhead, while maintaining competitive performance.”).
Regarding claim 17, the claim is similar to claim 7 and is rejected under the same rationales.
Claims 9 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Zhan, et al., Non-Patent Literature “Deep Model Compression Via Two-Stage Deep Reinforcement Learning” (“Zhan”) in view of Courbariaux, et al., Non-Patent Literature “Binarized Neural Networks: Training Neural Networks with Weights and Activations Constrained to +1 or −1” (“Courbariaux”).
Regarding claim 9, Zhan discloses the computer-implemented method of claim 1. While Zhan discloses a system that compresses a neural network on a per-layer basis, Zhan does not explicitly teach further comprising performing one or more operations to convert the compressed trained machine learning model to a binary format that is executable via the hardware device.
Courbariaux teaches further comprising performing one or more operations to convert the compressed trained machine learning model to a binary format that is executable via the hardware device. (Courbariaux, abstract, “We introduce a method to train Binarized Neural Networks (BNNs)- neural networks with binary weights and activations at run-time. At training-time the binary weights and activations are used for computing the parameters gradients. During the forward pass, BNNs drastically reduce memory size and accesses, and replace most arithmetic operations with bit-wise operations, which is expected to substantially improve power-efficiency [further comprising performing one or more operations to convert the compressed trained machine learning model to a binary format that is executable via the hardware device.].”).
Zhan and Courbariaux are both in the same field of endeavor (i.e. neural network optimization). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Zhan and Courbariaux to teach the above limitation(s). The motivation for doing so is that converting models to binary format reduces the computational resources used by a model (cf. Courbariaux, pg. 6 col. 2, “Importantly, we can see that memory accesses typically consume more energy than arithmetic operations, and memory access’ cost augments with memory size. In comparison with 32-bit DNNs, BNNs require 32 times smaller memory size and 32 times fewer memory accesses. This is expected to reduce energy consumption drastically (i.e., more than 32 times).”).
Regarding claim 18, the claim is similar to claim 9 and is rejected under the same rationales.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NICHOLAS S WU whose telephone number is (571)270-0939. The examiner can normally be reached Monday - Friday 8:00 am - 4:00 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.
/N.S.W./Examiner, Art Unit 2148 /MICHELLE T BECHTOLD/Supervisory Patent Examiner, Art Unit 2148
Prosecution Insights