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 .
The action is in response to the application filed on 3/22/2023. Claims 1-16 are pending and have been examined.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 3/22/2023 is in compliance with the provisions of 37 CFR 1.97, 1.98, and MPEP § 609. It has been placed in the application file, and the information referred to therein has been considered as to the merits.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 1-16 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding Claims 1 and 9:
Claims 1 and 9 recite the limitation "to the target device" in line 5 from the bottom of the claim. There is no target device mentioned before this limitation. There is insufficient antecedent basis for this limitation in the claim.
Claims 1 and 9 recite the limitation "the compressed inference model received from the target device" in lines 3-4 from the bottom of the claim. There is no compressed model received from the target device before this limitation. There is insufficient antecedent basis for this limitation in the claim.
Regarding Claims 2-8 and 10-15:
Claims 2-8 and 10-15 are rejected as being dependent on a rejected base claim without curing any of the deficiencies.
Claim Objections
Claims 5 and 13 are objected to because of the following informalities: “based on the priority at least a predetermined criterion” should read “based on the priority of at least a predetermined criterion”. 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 1-16 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:
The claim recites a method which is one of the four statutory categories of patentable subject matter.
Step 2A prong 1:
The claim recites an abstract idea configuring a set of compression methods to be applied to the inference model and a set of parameters for the set of compression methods on the basis of the constraint which amounts to a mental process as it can be performed in a human mind.
The claim recites an additional abstract idea applying a first compression method to be applied to the inference model and a set of parameters for the set of compression methods on the basis of the constraint, through a… which amounts to a mental process as it can be performed in a human mind.
The claim recites an additional abstract idea determining whether a compressed inference model is generated from the inference model through the… which amounts to a mental process as it can be performed in a human mind.
The claim recites an additional abstract idea when it is determined that the compressed inference model is not generated, applying a second compression method included in the set of compression methods, and a second parameter among the set of parameters to the inference model, following the first compression method, through the… which amounts to a mental process as it can be performed in a human mind.
The claim recites an additional abstract idea determining an optimal set of parameters on the basis of the performance of the compressed inference model received from the target device wherein the performance of the compressed inference model is measured by the target device using the dataset which amounts to a mental process as it can be performed in a human mind.
Step 2A prong 2:
The additional element of using a computing device is a generic computer component used to implement the abstract idea, therefore does not integrate the abstract idea into practical application MPEP 2106.05(f).
The additional element of using at least one processor is a generic computer component used to implement the abstract idea, therefore does not integrate the abstract idea into practical application MPEP 2106.05(f).
The additional element of using a compression pipeline is a generic computer component used to implement the abstract idea, therefore does not integrate the abstract idea into practical application MPEP 2106.05(f).
The additional element of receiving an inference model, a dataset, and a constraint does not integrate the abstract idea into practical application because receiving graph data is considered an insignificant extra solution activity of “mere data gathering” MPEP 2106.05(g).
The additional element of when it is determined that the compressed inference model is generated, transmitting the compressed inference model to the target device does not integrate the abstract idea into practical application because receiving graph data is considered an insignificant extra solution activity of “mere data gathering” MPEP 2106.05(g).
Step 2B:
The additional element of using a computing device is a generic computer component used to implement the abstract idea, therefore does not amount to significantly more MPEP 2106.05(f).
The additional element of using at least one processor is a generic computer component used to implement the abstract idea, therefore does not amount to significantly more MPEP 2106.05(f).
The additional element of using a compression pipeline is a generic computer component used to implement the abstract idea, therefore does not amount to significantly more MPEP 2106.05(f).
The additional element of receiving an inference model, a dataset, and a constraint data does not amount to significantly more because the additional element is an insignificant extra solution activity and further is a well understood routine and conventional activity. See MPEP 2106.05(d)(II)(i), (buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network)).
The additional element of when it is determined that the compressed inference model is generated, transmitting the compressed inference model to the target device data does not amount to significantly more because the additional element is an insignificant extra solution activity and further is a well understood routine and conventional activity. See MPEP 2106.05(d)(II)(i), (buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network)).
Therefore, the claim is ineligible.
Regarding Claim 2:
Claim 2 which incorporates the rejection of Claim 1, recites further abstract ideas selecting the first compression method and the first parameter for the first compression method and determining whether to further select a compression method and a parameter on the basis of the constraint and the first compression method and when it is determined to further select a compression method and a parameter, selecting the second compression method and the second parameter for the second compression method on the basis of the first compression method which are mental processes as they can be performed in a human mind. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 3:
Claim 3 which incorporates the rejection of Claim 1, recites further abstract ideas when the performance of the compressed inference model satisfies the constraint, the selected set of parameters is determined as the optimal parameter set and wherein when the performance of the compressed inference model does not satisfy the constraint… are repeatedly performed which are mental processes as they can be performed in a human mind. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 4:
Claim 4 incorporates the rejection of Claim 1. The claim further recites a description of the constraint from the steps of Claim 1 and is ineligible for the same reasons as set forth in Claim 1. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 5:
Claim 5 which incorporates the rejection of Claim 1, recites further abstract ideas a priority is assigned to each of a plurality of items included in the constraint and determining whether the compressed inference model satisfies the constraint based on the priority at least a predetermined criterion on the basis of the performance of the compressed inference model which are mental processes as they can be performed in a human mind. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 6:
Claim 6 which incorporates the rejection of Claim 1, recites a further abstract idea the set of compression methods are selected from a compression method pool, and wherein the compression method pool includes pruning, quantization, resolution change, and filter decomposition which is a mental process as it can be performed in a human mind. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 7:
Claim 7 incorporates the rejection of Claim 1. The claim further recites a description of the performance of the compressed inference model from the determining an optimal set of parameters step and is ineligible for the same reasons as set forth in Claim 1. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 8:
Claim 8 which incorporates the rejection of Claim 1, recites a further abstract idea the set of compression methods is configured on the basis of a predetermined rule, and the predetermined rule includes at least one of a first rule that a quantization-based compression method included in the optimal parameter set is to be positioned last in the compression pipeline or a second rule that an activation change-based compression method is to be positioned before the quantization-based compression method which is a mental process as it can be performed in a human mind. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 9:
Step 1:
The claim recites a computing device which is one of the four statutory categories of patentable subject matter.
Step 2A prong 1:
The claim recites an abstract idea configure a set of compression methods to be applied to the inference model and a set of parameters for the set of compression methods on the basis of the constraint which amounts to a mental process as it can be performed in a human mind.
The claim recites an additional abstract idea apply a first compression method to be applied to the inference model and a set of parameters for the set of compression methods on the basis of the constraint, through a… which amounts to a mental process as it can be performed in a human mind.
The claim recites an additional abstract idea determine whether a compressed inference model is generated from the inference model through the… which amounts to a mental process as it can be performed in a human mind.
The claim recites an additional abstract idea when it is determined that the compressed inference model is not generated, apply a second compression method included in the set of compression methods, and a second parameter among the set of parameters to the inference model, following the first compression method, through the… which amounts to a mental process as it can be performed in a human mind.
The claim recites an additional abstract idea determine an optimal set of parameters on the basis of the performance of the compressed inference model received from the target device wherein the performance of the compressed inference model is measured by the target device using the dataset which amounts to a mental process as it can be performed in a human mind.
Step 2A prong 2:
The additional element of using a computing device is a generic computer component used to implement the abstract idea, therefore does not integrate the abstract idea into practical application MPEP 2106.05(f).
The additional element of using a memory configured to store at least one instruction is a generic computer component used to implement the abstract idea, therefore does not integrate the abstract idea into practical application MPEP 2106.05(f).
The additional element of using at least one processor executing the at least one instruction is a generic computer component used to implement the abstract idea, therefore does not integrate the abstract idea into practical application MPEP 2106.05(f).
The additional element of using a compression pipeline is a generic computer component used to implement the abstract idea, therefore does not integrate the abstract idea into practical application MPEP 2106.05(f).
The additional element of receive an inference model, a dataset, and a constraint does not integrate the abstract idea into practical application because receiving graph data is considered an insignificant extra solution activity of “mere data gathering” MPEP 2106.05(g).
The additional element of when it is determined that the compressed inference model is generated, transmit the compressed inference model to the target device does not integrate the abstract idea into practical application because receiving graph data is considered an insignificant extra solution activity of “mere data gathering” MPEP 2106.05(g).
Step 2B:
The additional element of using a computing device is a generic computer component used to implement the abstract idea, therefore does not amount to significantly more MPEP 2106.05(f).
The additional element of using a memory configured to store at least one instruction is a generic computer component used to implement the abstract idea, therefore does not amount to significantly more MPEP 2106.05(f).
The additional element of using at least one processor executing the at least one instruction is a generic computer component used to implement the abstract idea, therefore does not amount to significantly more MPEP 2106.05(f).
The additional element of using a compression pipeline is a generic computer component used to implement the abstract idea, therefore does not amount to significantly more MPEP 2106.05(f).
The additional element of receive an inference model, a dataset, and a constraint data does not amount to significantly more because the additional element is an insignificant extra solution activity and further is a well understood routine and conventional activity. See MPEP 2106.05(d)(II)(i), (buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network)).
The additional element of when it is determined that the compressed inference model is generated, transmit the compressed inference model to the target device data does not amount to significantly more because the additional element is an insignificant extra solution activity and further is a well understood routine and conventional activity. See MPEP 2106.05(d)(II)(i), (buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network)).
Therefore, the claim is ineligible.
Regarding Claim 10:
Claim 10 which incorporates the rejection of Claim 9, recites further abstract ideas select the first compression method and the first parameter for the first compression method and determine whether to further select a compression method and a parameter on the basis of the constraint and the first compression method and when it is determined to further select a compression method and a parameter, select the second compression method and the second parameter for the second compression method on the basis of the first compression method which are mental processes as they can be performed in a human mind. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 11:
Claim 11 which incorporates the rejection of Claim 9, recites further abstract ideas when the performance of the compressed inference model satisfies the constraint, the selected set of parameters is determined as the optimal parameter set and wherein when the performance of the compressed inference model does not satisfy the constraint, the processor is further configured to: repeatedly… which are mental processes as they can be performed in a human mind. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 12:
Claim 12 incorporates the rejection of Claim 9. The claim further recites a description of the constraint from the steps of Claim 9 and is ineligible for the same reasons as set forth in Claim 9. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 13:
Claim 13 which incorporates the rejection of Claim 9, recites further abstract ideas a priority is assigned to each of a plurality of items included in the constraint and determine whether the compressed inference model satisfies the constraint based on the priority at least a predetermined criterion on the basis of the performance of the compressed inference model which are mental processes as they can be performed in a human mind. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 14:
Claim 14 which incorporates the rejection of Claim 9, recites a further abstract idea select the set of compression methods from a compression method pool, and wherein the compression method pool includes pruning, quantization, resolution change, and filter decomposition which is a mental process as it can be performed in a human mind. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 15:
Claim 15 incorporates the rejection of Claim 9. The claim further recites a description of the performance of the compressed inference model from the determine an optimal set of parameters step and is ineligible for the same reasons as set forth in Claim 9. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
Regarding Claim 16:
Claim 16 which incorporates the rejection of Claim 9, recites a further abstract idea configure the set of compression methods on the basis of a predetermined rule, and the predetermined rule includes at least one of a first rule that a quantization-based compression method included in the optimal parameter set is to be positioned last in the compression pipeline or a second rule that an activation change-based compression method is to be positioned before the quantization-based compression method which is a mental process as it can be performed in a human mind. The claim does not recite any additional elements that integrate the abstract idea into practical application or amount to significantly more. The claim is ineligible.
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.
Claims 1-5, 7, 9-13, and 15 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Liu et al. “On-Demand Deep Model Compression for Mobile Devices: A Usage-Driven Model Selection Framework”, hereinafter “Liu”.
Regarding Claim 1:
Claim 1 recites the following contingent limitations: “when it is determined that the compressed influence model is not generated, applying a second compression method included in the set of compression methods, and a second parameter among the set of parameters to the inference model, following the first compression method, through the compression pipeline” and “when it is determined that the compressed inference model is generated, transmitting the compressed inference model to the target device”.
These limitations are contingent because they recite steps that are only required to be performed if their conditions precedent are met. Limitation applying a second compression method included in the set of compression methods, and a second parameter among the set of parameters to the inference model, following the first compression method, through the compression pipeline only needs to be performed if the compressed influence model is not generated, and limitation transmitting the compressed inference model to the target device only needs to be performed if the compressed inference model is generated. These conditions are mutually exclusive, and therefore only one of limitations applying a second compression method included in the set of compression methods, and a second parameter among the set of parameters to the inference model, following the first compression method, through the compression pipeline and transmitting the compressed inference model to the target device can be performed. Therefore, the BRI of Claim 1 requires limitation A method of determining an optimal parameter set that is performed by a computing device including at least one processor, the method comprising: receiving an inference model, a dataset, and a constraint; configuring a set of compression methods to be applied to the inference model and a set of parameters for the set of compression methods on the basis of the constraint; applying a first compression method included in the set of compression methods and a first parameter related to the first compression method to the inference model, through a compression pipeline; determining whether a compressed inference model is generated from the inference model through the compression pipeline and only one of either limitation applying a second compression method included in the set of compression methods, and a second parameter among the set of parameters to the inference model, following the first compression method, through the compression pipeline or limitation transmitting the compressed inference model to the target device.
Further regarding Claim 1, Liu teaches:
A method of determining an optimal parameter set that is performed by a computing device including at least one processor (Liu implements their method in Python demonstrating that Liu performs their method on a computer, in which processor, memory, and storage devices are inherent, p. 393, col. 2, paragraph 3, “We implement AdaDeep with TensorFlow [14] in Python”), the method comprising:
receiving an inference model, a dataset, and a constraint (p. 393, col. 2, paragraph 3, “selects an initial DNN from a pool of three state-of-the-art DNN models, including LeNet [29], AlexNet [23], and VGG [37], according to the size of samples in Dt”, p. 390, col. 2, paragraph 3, “User demands are expressed as a set of constraints on accuracy, energy, latency and storage”);
configuring a set of compression methods to be applied to the inference model (p. 390, col. 1, paragraph 3, “framework that automatically selects a combination of DNN compression techniques to adapt to user-specified performance requirements and platform imposed resource constraints on accuracy, latency, storage and energy consumption”) and a set of parameters for the set of compression methods on the basis of the constraint (p. 394, col. 2, paragraph 5, “The parameters (i.e., k inW1f ,W1c andW2, the depth multiplier α in C2, the sparse random multiplier θ in C3) are empirically optimized by comparing the performance improvement on the layer where the compression technique is applied”);
applying a first compression method included in the set of compression methods and a first parameter related to the first compression method to the inference model, through a compression pipeline (pipeline is DRL optimizer algorithm, p.393, Algorithm 1, line 5, “Select at for ot by Q value (ϵ − дreedy) and observe Rt”, at is a specific compression technique as shown in Table 1, p.394, “The details of the compression techniques are as follows…” this section shows specific parameters of each technique which is the first parameter based on which technique is chosen during the algorithm);
determining whether a compressed inference model is generated from the inference model through the compression pipeline (p. 393, col. 1, paragraph 2, “After taking an action, we can observe the reward R1 for G, and R2 for H. Their interaction and balance guide the selection process.”, p. 393, Algorithm 1, compressed inference model is not generated until all iterations of episodes are complete);
when it is determined that the compressed inference model is not generated (p. 393, Algorithm 1, the compressed inference model is not generated until loop is done), applying a second compression method included in the set of compression methods, and a second parameter among the set of parameters to the inference model, following the first compression method, through the compression pipeline (After action taken in previous iteration another action is taken, p.393, Algorithm 1 line 5, “Select at for ot by Q value (ϵ − дreedy) and observe Rt”, this second selection will use a specific second parameter related to compression technique, p.394, “The details of the compression techniques are as follows…” this section shows specific parameters of each technique);
when it is determined that the compressed inference model is generated, transmitting the compressed inference model to the target device (p. 393, col. 2, paragraph 3, “The compressed DNNs generated by AdaDeep are then loaded into the target platforms and evaluated as Android projects”); and
determining an optimal set of parameters on the basis of the performance of the compressed inference model received from the target device and wherein the performance of the compressed inference model is measured by the target device using the dataset (optimal set of parameters are based on specific hardware information during algorithm and determined when used on target device and tested, p. 390, col. 2, paragraph 3, “AdaDeep aims to solve the following constrained optimization problem… where A, E, T and S denote the measured accuracy, energy cost, latency and storage of a given DNN running on a specific mobile platform”, p. 397, col. 2, paragraph 3, “evaluates AdaDeep across twelve different mobile devices”, p.394, col. 2, paragraph 6, “For each layer compression technique, we load the compressed DNN on Device 1 to process the test data 10 times, and obtain the mean and variance of the inference performance and resource utilization cost”, p. 397, Table 5).
Regarding Claim 2, Liu teaches the method as referenced above in Claim 1. Liu further teaches:
wherein the configuring the set of compression methods and the set of parameters includes:
selecting the first compression method and the first parameter for the first compression method (first compression method is selected during iteration in Algorithm 1, p. 393, Algorithm 1, line 5, “Select at for ot by Q value (ϵ − дreedy) and observe Rt”, at is a specific compression technique as shown in Table 1, p.394, “The details of the compression techniques are as follows…” this section shows specific parameters of each technique which is the first parameter based on which technique is chosen during the algorithm);
determining whether to further select a compression method and a parameter on the basis of the constraint and the first compression method (specific subsequent compression methods during iterations are determined on optimization of previous iteration compression methods, p. 393, Algorithm 1, line 5 shows action selected based on Q value line 8 shows Q value updated based on each iteration, reward functions are based on constraints, p. 390, col. 1, paragraph 3, “framework that automatically selects a combination of DNN compression techniques to adapt to user-specified performance requirements and platform imposed resource constraints on accuracy, latency, storage and energy consumption”); and
when it is determined to further select a compression method and a parameter, selecting the second compression method and the second parameter for the second compression method on the basis of the first compression method (Next iteration after selecting first compression method, p. 393, Algorithm 1, line 5 shows action selected based on Q value line 8 shows Q value updated based on each iteration).
Regarding Claim 3, Liu teaches the method as referenced above in Claim 1. Liu further teaches:
wherein when the performance of the compressed inference model satisfies the constraint, the selected set of parameters is determined as the optimal parameter set (After model is optimized based on constraint it is sent to target devices and tested against other methods, p. 397, col. 1, paragraph 2, “DRL optimizer outperforms the other two schemes in terms of the storage size, latency, and energy consumption… performance metrics (A, S, T and E) and the resource constraints (S and T ) are systematically included in the reward value and adaptively feedback to the DRL decision process”, p.397, Table 5), and
wherein when the performance of the compressed inference model does not satisfy the constraint (AdaDeep aims to optimize based on constraints and it is not satisfied until Algorithm 1 is finished, p. 390, col. 2, paragraph 3, “AdaDeep aims to solve the following constrained optimization problem… where A, E, T and S denote the measured accuracy, energy cost, latency and storage of a given DNN running on a specific mobile platform”), the configuring the set of compression methods and the set of parameters, the applying the first compression method and the first parameter to the inference model, the determining whether the compressed inference model is generated, the applying the second compression method and the second parameter to the inference model (p.393, Algorithm 1 repeats the process of applying a first compression method at with a first parameter then evaluates and determines whether the inference model is generated followed by applying a second method in the next iteration then applies second compression method at with a second parameter), the transmitting the compressed inference model to the target device, and the determining the optimal parameter set are repeatedly performed (p. 399, Table 7, p. 397, col. 2, paragraph 3, “This experiment evaluates AdaDeep across twelve different mobile devices”, p. 390, col. 2, paragraph 3, “AdaDeep aims to solve the following constrained optimization problem… where A, E, T and S denote the measured accuracy, energy cost, latency and storage of a given DNN running on a specific mobile platform”, Algorithm is performed on multiple devices and tested on target device to determine optimal parameters).
Regarding Claim 4, Liu teaches the method as referenced above in Claim 1. Liu further teaches:
wherein the constraint includes a value of at least one item among device, accuracy, model size, latency, compression time, and energy consumption (p. 390, col. 2, paragraph 3, “AdaDeep aims to solve the following constrained optimization problem. arдmax Js ∈Jall μ1N(A − Amin) + μ2N(Emax − E) s.t. T ≤ Tbдt , S ≤ Sbдt , (1) where A, E, T and S denote the measured accuracy, energy cost, latency and storage of a given DNN running on a specific mobile platform”).
Regarding Claim 5, Liu teaches the method as referenced above in Claim 1. Liu further teaches:
wherein a priority is assigned to each of a plurality of items included in the constraint (priority of constraints is based on coefficients based on device specification, p. 397, col. 2, paragraph 4, “different devices have different resource constraints, which lead to different performance and budget demands and thus require different coefficients μ1 ∼ μ4 in Eq. (8). Specifically, we empirically optimize μ1 ∼ μ4 for different devices to be: μ2 = max{ 4000−Ebattery 4000 , 0.6}, μ1 = 1 − μ2, μ4 = max{ 8−SCache 8 , 0.6}, and μ3 = 1 − μ4”), and
wherein the determining the optimal set of parameters includes:
determining whether the compressed inference model satisfies the constraint based on the priority at least a predetermined criterion on the basis of the performance of the compressed inference model (p. 395, col. 2, paragraph 7, “We set the scaling coefficients in Eq. (8) to be μ1 = 0.6 and μ2 = 0.4 considering that the battery capacity in RedMi 3S is relatively large and thus the energy consumption is of lower priority, and we set μ3 = 0.5 and μ4 = 0.5 in Eq. (8) because their corresponding constraints (i.e., C and Sp ) are equally important”, Model optimization when determining whether constraint is satisfied is based on reward functions Equation 8 in Algorithm 1, line 7,).
Regarding Claim 7, Liu teaches the method as referenced above in Claim 1. Liu further teaches:
The method of claim 1, wherein the performance of the compressed inference model includes a value of at least one item among latency, accuracy, and energy consumption (p. 397, col. 1, paragraph 2, “DRL optimizer outperforms the other two schemes in terms of the storage size, latency, and energy consumption… performance metrics (A, S, T and E) and the resource constraints (S and T ) are systematically included in the reward value and adaptively feedback to the DRL decision process”).
Regarding Claim 9, Liu teaches:
A computing device for determining an optimal parameter set, the computing device comprising:
a memory configured to store at least on instruction ; and at least one processor executing the at least one instruction (Liu implements their method in Python demonstrating that Liu performs their method on a computer, in which processor, memory, and storage devices are inherent, p. 393, col. 2, paragraph 3, “We implement AdaDeep with TensorFlow [14] in Python”), wherein the processor is configured to:
receive an inference model, a dataset, and a constraint (p. 393, col. 2, paragraph 3, “selects an initial DNN from a pool of three state-of-the-art DNN models, including LeNet [29], AlexNet [23], and VGG [37], according to the size of samples in Dt”, p. 390, col. 2, paragraph 3, “User demands are expressed as a set of constraints on accuracy, energy, latency and storage”);
configure a set of compression methods to be applied to the inference model (p. 390, col. 1, paragraph 3, “framework that automatically selects a combination of DNN compression techniques to adapt to user-specified performance requirements and platform imposed resource constraints on accuracy, latency, storage and energy consumption”) and a set of parameters for the set of compression methods on the basis of the constraint (p. 394, col. 2, paragraph 5, “The parameters (i.e., k inW1f ,W1c andW2, the depth multiplier α in C2, the sparse random multiplier θ in C3) are empirically optimized by comparing the performance improvement on the layer where the compression technique is applied”);
apply a first compression method included in the set of compression methods and a first parameter related to the first compression method to the inference model, through a compression pipeline (pipeline is DRL optimizer algorithm, p.393, Algorithm 1, line 5, “Select at for ot by Q value (ϵ − дreedy) and observe Rt”, at is a specific compression technique as shown in Table 1, p.394, “The details of the compression techniques are as follows…” this section shows specific parameters of each technique which is the first parameter based on which technique is chosen during the algorithm);
determine whether a compressed inference model is generated from the inference model through the compression pipeline (p. 393, col. 1, paragraph 2, “After taking an action, we can observe the reward R1 for G, and R2 for H. Their interaction and balance guide the selection process.”, p. 393, Algorithm 1, compressed inference model is not generated until all iterations of episodes are complete);
when it is determined that the compressed inference model is not generated (p. 393, Algorithm 1, the compressed inference model is not generated until loop is done), apply a second compression method included in the set of compression methods, and a second parameter among the set of parameters to the inference model, following the first compression method, through the compression pipeline (After action taken in previous iteration another action is taken, p.393, Algorithm 1 line 5, “Select at for ot by Q value (ϵ − дreedy) and observe Rt”, this second selection will use a specific second parameter related to compression technique, p.394, “The details of the compression techniques are as follows…” this section shows specific parameters of each technique);
when it is determined that the compressed inference model is generated, transmit the compressed inference model to the target device (p. 393, col. 2, paragraph 3, “The compressed DNNs generated by AdaDeep are then loaded into the target platforms and evaluated as Android projects”); and
determine an optimal set of parameters on the basis of the performance of the compressed inference model received from the target device and wherein the performance of the compressed inference model is measured by the target device using the dataset (optimal set of parameters are based on specific hardware information during algorithm and determined when used on target device and tested, p. 390, col. 2, paragraph 3, “AdaDeep aims to solve the following constrained optimization problem… where A, E, T and S denote the measured accuracy, energy cost, latency and storage of a given DNN running on a specific mobile platform”, p. 397, col. 2, paragraph 3, “evaluates AdaDeep across twelve different mobile devices”, p.394, col. 2, paragraph 6, “For each layer compression technique, we load the compressed DNN on Device 1 to process the test data 10 times, and obtain the mean and variance of the inference performance and resource utilization cost”, p. 397, Table 5).
Regarding Claim 10, the rejection of Claim 9 is incorporated and further, the claim is rejected
for the same reasons as set forth in Claim 2.
Regarding Claim 11, the rejection of Claim 9 is incorporated and further, the claim is rejected
for the same reasons as set forth in Claim 3.
Regarding Claim 12, the rejection of Claim 9 is incorporated and further, the claim is rejected
for the same reasons as set forth in Claim 4.
Regarding Claim 13, the rejection of Claim 9 is incorporated and further, the claim is rejected
for the same reasons as set forth in Claim 5.
Regarding Claim 15, the rejection of Claim 9 is incorporated and further, the claim is rejected
for the same reasons as set forth in Claim 7.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 6, 8, 14, and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Liu in view of Wang et al. “HAQ: Hardware-Aware Automated Quantization with Mixed Precision”, hereinafter “Wang”.
Regarding Claim 6, Liu teaches the method as referenced above in Claim 1. Liu further teaches:
wherein the set of compression methods are selected from a compression method pool (Table 1, “Action a∼As Selectable combinations of compression techniques”, p. 397, col. 2, col. 7, “We apply ten mainstream compression techniques from three categories, i.e., weight compression(W1f , W2, W3, W1c ), convolution decomposition (C1, C2, C3), and special architecture layers (L1, L2, L3)”), and
wherein the compression method pool includes pruning (p. 394, col. 1, paragraph 4, “W3: prune fc1 and fc2 using the magnitude based weight pruning strategy”)… resolution change (p. 394, col. 1, paragraph 5, “L3: replace the traditional fc layers, fci and fci+1, with a global average pooling layer [31]. It generates one feature map for each category in the last conv layer”), and filter decomposition (p. 394, col. 1, paragraph 7, “C1: decompose convi using convolution kernel sparse decomposition”).
Liu does not expressly teach:
quantization
However, Wang teaches:
quantization (p. 1, Abstract, “automatically determine the quantization policy”, p. 4, col. 2, paragraph 3, “quantize the weights and activations of each layer using the action ak”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use quantization as one of the compression methods in the compression options of Liu. The motivation to do so would be to accelerate inference (p. 1 Abstract, “quantization is a widely used technique to compress and accelerate deep neural network (DNN) inference”).
Regarding Claim 8, Liu in view of Wang teaches the method as referenced above in Claim 1. In the combination as set forth above, Liu in view of Wang teaches:
wherein the set of compression methods is configured on the basis of a predetermined rule (predetermined rule is user-specified performance requirements and platform imposed resource constraints, p. 390, col. 1, paragraph 3, “framework that automatically selects a combination of DNN compression techniques to adapt to user-specified performance requirements and platform imposed resource constraints on accuracy, latency, storage and energy consumption”), and
the predetermined rule includes at least one of a first rule that a quantization-based compression method included in the optimal parameter set is to be positioned last in the compression pipeline or a second rule that an activation change-based compression method is to be positioned before the quantization-based compression method (The optimizing process uses a combination of DNN compression techniques that has the best reward for the predetermined rule, p. 393, Algorithm 1, line 7 using reward functions and Equation 8 showing reward functions based on rule, p. 390, col. 1, paragraph 3, “framework that automatically selects a combination of DNN compression techniques to adapt to user-specified performance requirements and platform imposed resource constraints, the optimizing process chooses the best compression techniques which can cause them to be selected in any order including quantization to be positioned last and pruning method before it).
Regarding Claim 14, the rejection of Claim 9 is incorporated and further, the claim is rejected
for the same reasons as set forth in Claim 6.
Regarding Claim 16, the rejection of Claim 9 is incorporated and further, the claim is rejected
for the same reasons as set forth in Claim 8.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JESSE CHEN COULSON whose telephone number is (571)272-4716. The examiner can normally be reached Monday-Friday 8:30-5:30.
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, Kakali Chaki can be reached at (571) 272-3719. 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.
/JESSE C COULSON/
Examiner, Art Unit 2122
/KAKALI CHAKI/Supervisory Patent Examiner, Art Unit 2122