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 .
Claims 1-8, 10-18 and 20-25 are presented for examination.
Claim Objections
Claims 1, 8, 11, 18 and 21 are objected to because of the following informalities:
Claim 1 [line 13], claim 11 [line 14] and claim 21 [line 19]: “selecting a GNN from the plurality of GNNs and the plurality of new GNNs based on the evaluation, the GNN to be used for reducing precisions of quantizable parameters of a second DNN” is confusing and should be rephrased as “selecting a GNN from the plurality of GNNs and the plurality of new GNNs based on the evaluation, wherein the GNNs is used for reducing precisions of quantizable parameters of a second DNN”
Claim 8 [line 2]: “the another quantizable operation” should be “the other quantizable operation”,
Claim 18 [line 2]: “the another quantizable operation” should be “the other quantizable operation”
Appropriate corrections are 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 1-8, 10-18 and 20-25 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Independent claims
Step 1
Claim 1 is drawn to a method for optimizing multiple objectives of mixed-precision quantization, claim 11 is drawn to a non-transitory computer-readable media, and claim 21 is drawn to an apparatus that stores computer program instructions to perform the method of claim 1. Therefore, each of these groups falls under one of four categories of statutory subject matter (process/method, machines/product/apparatus, manufactures, and composition of matter).
Step 2A – Prong 1
Claims 1, 11 and 21 are directed to a judicially recognized exception of an abstract idea without significantly more.
Claim 1, 11 and 21 recite a method of generating a plurality of graph neural networks (GNNs) that under its broadest reasonable interpretation enumerates a mathematical concept. A human can perform the calculation using words or using mathematical symbols to create mathematical models such as GNNs. Therefore, the step of generating a plurality of GNNs is nothing more than a mathematical concept (MPEP 2106.04(a)(2)(I)).
Claim 1, 11 and 21 recite further a method of generating a plurality of new GNNs based on the plurality of GNNs that under its broadest reasonable interpretation enumerates a mathematical concept. A human can perform the calculation using words or using mathematical symbols to create mathematical models such as GNNs. Therefore, the step of generating a plurality of new GNNs is nothing more than a mathematical concept (MPEP 2106.04(a)(2)(I)).
Claim 1, 11 and 21 recite further a method of inputting the sequential graph into the plurality of GNNs and the plurality of new GNNs that under its broadest reasonable interpretation enumerates a mathematical concept. A human can perform the calculation using words or using mathematical symbols to feed data (the sequential graph) into mathematical models. Therefore, the step of inputting the sequential graph into the GNNs is nothing more than a mathematical concept (MPEP 2106.04(a)(2)(I)).
Claim 1, 11 and 21 recite further a method of evaluating outputs of the plurality of GNNs and the plurality of new GNNs based on conflicting objectives of reducing precisions of the quantizable parameters of the first DNN that under its broadest reasonable interpretation enumerates a mathematical concept. A human can perform the calculation using words or using mathematical symbols to optimize or to evaluate the outputs involving mathematical analysis. Therefore, the step of evaluating outputs of the plurality of GNNs is nothing more than a mathematical concept (MPEP 2106.04(a)(2)(I)).
Step 2A – Prong 2
Claims 1, 11 and 21 recite further a method of generating a sequential graph for a first deep neural network (DNN), the first DNN comprising a sequence of quantizable operations, each of which includes quantizable parameters and is represented by a different node in the sequential graph, wherein a quantizable operation in the sequence comprises an activation function and quantizable parameters of the quantizable operation comprise activations that fails to integrate the abstract idea into a practical application. The step of generating a sequential graph for a first DNN is a form of insignificant input and output solution activities, where generating a sequential graph wherein each of which includes quantizable parameters and wherein quantizable operation comprises an activation is necessary for all uses of the judicial exception. This additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (MPEP 2106.05(g)).
Claims 1, 11 and 21 recite further a method of selecting a GNN from the plurality of GNNs and the plurality of new GNNs based on the evaluation, the GNN to be used for reducing precisions of quantizable parameters of a second DNN that fails to integrate the abstract idea into a practical application. The step of selecting a GNN from a plurality of GNNs is a form of insignificant input and output solution activities, where selecting a GNN based on the evaluation is necessary for all uses of the judicial exception. This additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (MPEP 2106.05(g)).
Step 2B
The additional elements in step 2A-Prong 2 those are forms of insignificant extra-solution activities, do not amount to significantly more than an abstract idea because the court decision have determined that these additional elements of generating a sequential graph wherein each of which includes quantizable parameters and wherein quantizable operation comprises an activation; and selecting a GNN based on the evaluation to be well-understood, routine, and conventional when claimed in a merely generic manner (MPEP 2106.05(d)(II)).
As such, claims 1, 11 and 21 are not patent eligible.
Dependent claims
Claims 2-8, 10, 12-18, 20 and 22-25 merely narrow the previously recited abstract idea limitations. For the reasons described above with respect to claims 1, 11 and 21, this judicial exception is not meaningfully integrated into a practical application, or significantly more than the abstract idea. The claims disclose similar limitations described for the independent claims above and do not provide anything more than the mental processes that are practically capable of being performed in the human mind with the assistance of pen. Therefore, claims 2-10 and 12-19 also recite abstract ideas that do not integrate into a practical application or amount to significantly more than the judicial exception, and are rejected under U.S.C. 101.
Step 1
Claims 2-8 and 10 are drawn to a method for optimizing multiple objectives of mixed-precision quantization, claims 12-18 and 20 are drawn to a non-transitory computer-readable media, and claims 22-25 are drawn to an apparatus that stores computer program instructions to perform the method of claims 2-8 and 10. Therefore, each of these groups falls under one of four categories of statutory subject matter (process/method, machines/product/apparatus, manufactures, and composition of matter).
Step 2A – Prong 1
Dependent claims 4, 14 and 23 recite further the mathematical process by wherein generating the plurality of new GNNs based on the plurality of GNNs comprises: generating new internal parameters based on internal parameters of the plurality of GNNs; and forming the plurality of new GNNs based on the new internal parameters and an architecture of neurons of the plurality of GNNs those are based on one or more features of the ML project (MPEP 2106.04(a)(2)(I)). Dependent claims 5, 15 and 24 recite further the mathematical process by wherein evaluating outputs of the plurality of GNNs and the plurality of new GNNs comprises: generating a Pareto optimal set from the plurality of GNNs and the plurality of new GNNs based on performances of the plurality of GNNs and the plurality of new GNNs in achieving the conflicting objectives, wherein the Pareto optimal set comprises one or more GNNs in the plurality of GNNs and the plurality of new GNNs those are based on one or more features of the ML project (MPEP 2106.04(a)(2)(I)).
Step 2A – Prong 2
Dependent claims 2, 12 and 22 recite further the insignificant extra solution activities by wherein the plurality of GNNs comprises a first species of GNNs and a second species of GNNs, the GNNs in the first species have a first architecture of neurons, and the GNNs in the second species have a second architecture of neurons that is different from the first architecture of neurons. 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 (MPEP 2106.05(g)).
Dependent claims 3 and 13 recite further the insignificant extra solution activities by wherein the GNNs in the first GNN species have different internal parameters. 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 (MPEP 2106.05(g)).
Dependent claims 6, 16 and 25 recite further the insignificant extra solution activities by wherein the GNN is configured to receive a sequential graph for the second DNN as an input and to output a bit-width probability distribution for each respective layer in the second DNN, the bit-width probability distribution comprising a plurality of probabilities, and each of the plurality of probabilities corresponds to a different bit-width. 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 (MPEP 2106.05(g)).
Dependent claims 7 and 17 recite further the insignificant extra solution activities by wherein a bit-width is to be selected from the bit-width probability distribution based on the plurality of probabilities and the bit-width is to be used to reduce precisions of quantizable parameters of the respective layer in the second DNN. 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 (MPEP 2106.05(g)).
Dependent claims 8 and 18 recite further the insignificant extra solution activities by wherein another quantizable operation in the sequence comprises a convolution and quantizable parameters of the another quantizable operation comprise weights. 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 (MPEP 2106.05(g)).
Dependent claims 10 and 20 recite further the insignificant extra solution activities by wherein the multiple objectives are selected from a group consisting of maximizing task performance of the DNN, minimizing model size of the DNN, and minimizing compute complexity of the DNN. 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 (MPEP 2106.05(g)).
As such, dependent claims 2-8, 10, 12-18, 20 and 22-25 are not patent eligible.
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 1-8, 10-18 and 20-25 are rejected under 35 U.S.C. 103 as being unpatentable over Liang et al (US 20200143243 A1) hereafter Liang, and further in view of Sharma et al (US 20200225996 A1) hereafter Sharma.
With respect to claim 1, Liang teaches a method for optimizing multiple objectives of mixed-precision quantization (a visualization of the AutoML implementation system is provided with the evolutionary algorithm for architecture called CoDeepNEAT which evolves both hyperparameters and network structure. CDN adapts multiobjective optimization to find minimal architectures [par. 0024-0027]), Liang teaches the method comprising:
generating a plurality of graph neural networks (GNNs) (CDN is based on NEAT for evolving deep neural network (DNN) architectures and hyperparameters. NEAT is the process which includes a population of chromosomes of minimal complexity, wherein each chromosome is represented as a graph and is referred as an individual. In NEAT, mutation involves randomly adding a node or a connection between two nodes. The population is divided into species (or subpopulations) based on some similarity metric [par. 0024-0027]);
generating a plurality of new GNNs based on the plurality of GNNs (each species grows proportionally to its fitness and evolution occurs separately in each species. The blueprint chromosome is a graph where each node contains a pointer to a particular module species. Each module chromosome is a graph that represents a small DNN. For each blueprint chromosome, each node in the blueprint’s graph is replaced with a module chosen randomly from the species to which that node points [par. 0024-0029]).
However, Liang does not disclose generating a sequential graph for a first deep neural network (DNN), the first DNN comprising a sequence of quantizable operations, each of which includes quantizable parameters and is represented by a different node in the sequential graph, wherein a quantizable operation in the sequence comprises an activation function and quantizable parameters of the quantizable operation comprise activations; inputting the sequential graph into the plurality of GNNs and the plurality of new GNNs; evaluating outputs of the plurality of GNNs and the plurality of new GNNs based on conflicting objectives of reducing precisions of the quantizable parameters of the first DNN; and selecting a GNN from the plurality of GNNs and the plurality of new GNNs based on the evaluation, the GNN to be used for reducing precisions of quantizable parameters of a second DNN.
In the same field of endeavor, Sharma teaches generating a sequential graph for a first deep neural network (DNN), the first DNN comprising a sequence of quantizable operations, each of which includes quantizable parameters and is represented by a different node in the sequential graph (the DNN workflow begins with defining a dataflow graph of the DNN using high-level API. The API allows the programmer to specify the precision for each operation in the DNN. The operation of the DNN model includes a dataflow analyzer component iterating over the nodes of the dataflow graph of the DNN [par. 0004, 0112-0114, 0172-0176]), wherein a quantizable operation in the sequence comprises an activation function and quantizable parameters of the quantizable operation comprise activations (a heterogenous architecture is disclosed for training quantized neural networks. Training operations includes quantized activations/weights in the forward phase favor dense execution due to the large overhead of zero-skipping for quantized activations. The quantized activations in the forward phase have precisions between 45-60% zeroes. Using mixed-precision allows the heterogenous architecture to reduce the high resource cost of the compute units. The backward propagation phase updates the original weights to reduce a loss function to improve classification. Weights are quantized from a first precision datatype into a second precision datatype. A convolution is performed that uses low-bit width fixed-point data for activations and weights [par. 0025, 0060-0066, 0073-0077]);
inputting the sequential graph into the plurality of GNNs and the plurality of new GNNs (inference phase includes receiving input data for an input layer with the input data being quantized (quantized from a first precision datatype for input data into a second precision datatype). At operation 282, inputs are quantized or a mixed precision datatype. A convolution is performed on output from operation 280 and the inputs from operation 282 to generate a weight loss function [par. 0073-0086]);
evaluating outputs of the plurality of GNNs and the plurality of new GNNs based on conflicting objectives of reducing precisions of the quantizable parameters of the first DNN (at operation 210, the method includes performing a convolution operation of a convolution layer on the input data and weights. At operation 212, output from operation 210 is generated as the second precision datatype and quantized into a first precision datatype at operation 214. The output of an output layer is available for further processing [par. 0073-0086]); and
selecting a GNN from the plurality of GNNs and the plurality of new GNNs based on the evaluation, the GNN to be used for reducing precisions of quantizable parameters of a second DNN (the CU 600 includes n quantized mixed precision multipliers, each of which can multiply up to m-bit operands. While m depends on the minimum precision required by the MAC operations in convolution/fully-connected layers, n depends on the ratio of precision(max)/precision(min). the outputs of the n quantized multipliers are added to produce an output [par. 0073-0086, 0099]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated the concept of dynamically optimizing the circuit for forward and backward propagation phases of training for neural networks as suggested by Sharma into the concept of using multiobjective evolutionary algorithm to maximize the performance and to minimize the complexity of the evolved networks simultaneously by calculating the Pareto front given a group of individuals as suggested by Liang because both of the systems addressing the process of training DNNs to optimize architecture and parameters to maximize the performance of the DNN. Doing so would be desirable because the concept of Liang would be more efficient by quantizing the weights of the DNNs which reduces the bit widths for data and operations in a deep learning to yield increased performance and/or energy efficiency (Sharma, [par. 0003]).
With respect to claim 2¸ the combination of Liang and Sharma teaches wherein the plurality of GNNs comprises a first species of GNNs and a second species of GNNs, the GNNs in the first species have a first architecture of neurons, and the GNNs in the second species have a second architecture of neurons that is different from the first architecture of neurons (Liang, the method of multiple objective optimization includes initializing a first population of modules and blueprints, wherein for each species of modules in the first population and during each generation in a plurality of generations of subsequent populations, creating a set of empty species and a set of non-empty species. The population is divided into species (or subpopulations) based on similarity metric. The blueprint chromosome (or individual) is a graph where each node contains a pointer to a particular module species [par. 0012, 0027-0029]).
With respect to claim 3¸ the combination of Liang and Sharma teaches wherein the GNNs in the first GNN species have different internal parameters (Liang, each module chromosome is a graph that represents a small DNN. Each node contains a pointer to a particular module species. Each species grows proportionally to its fitness and evolution occurs separately in each species [par. 0027-0029]).
With respect to claim 4¸ the combination of Liang and Sharma teaches wherein generating the plurality of new GNNs based on the plurality of GNNs comprises:
generating new internal parameters based on internal parameters of the plurality of GNNs (Liang, NEAT is used to achieve DNN complexity minimization with application of multiobjective searching. Evolutionary elitism is applied in both the blueprint and the module populations. The elitism involves preserving the top fraction of the individuals within each species into the next generation based on their ranking within the species. The ranking is based on sorting the individuals by a single, primary objective fitness [par. 0027-0029, 0034-0038]); and
forming the plurality of new GNNs based on the new internal parameters and an architecture of neurons of the plurality of GNNs (Liang, for each blueprint chromosome, each node in the blueprint’s graph is replaced with a module chosen randomly from the species to which that node points. If multiple blueprint nodes point to the same module species, then the same module is used in all of them. After the nodes in the blueprint have been replaced, the individual is converted into a DNN [par. 0027-0029, 0034-0038]).
With respect to claim 5¸ the combination of Liang and Sharma teaches wherein evaluating outputs of the plurality of GNNs and the plurality of new GNNs comprises:
generating a Pareto optimal set from the plurality of GNNs and the plurality of new GNNs based on performances of the plurality of GNNs and the plurality of new GNNs in achieving the conflicting objectives (Liang, Pareto front of each non-empty species is determined in accordance with at least a first and second objective. Removing some individuals in the Pareto front of each non-empty species and adding them to the first set of empty choices to form one or more sets of new species. Calculating Pareto front for MCDN to maximize the performance and minimize the complexity of the evolved networks simultaneously [par. 0012, 0013, 0038-0040]),
wherein the Pareto optimal set comprises one or more GNNs in the plurality of GNNs and the plurality of new GNNs (Liang, using Pareto front to determine the best blueprint chromosome as a graph that represents a DNN. The ranking of MCDN is computed using multiple fitness values for each individual to generate successive Pareto fronts from the individuals based on primary objective and/or secondary objective value [par. 0012, 0013, 0027-0029, 0035]).
With respect to claim 6¸ the combination of Liang and Sharma teaches wherein the GNN is configured to receive a sequential graph for the second DNN as an input and to output a bit-width probability distribution for each respective layer in the second DNN, the bit-width probability distribution comprising a plurality of probabilities, and each of the plurality of probabilities corresponds to a different bit-width (Sharma, quantization reduces bit widths for data and operations in deep learning model. Both the data-representations and precision for activations, weights and gradients vary between different DNN models. The heterogenous architecture of DNN utilizes properties of quantization in the bit-heterogeneous architecture to deliver significant improvement in performance and energy efficiency. The method includes receiving input data for an input layer with the input data being quantized or a mixed-precision datatype [par. 0003, 0062-0066, 0073-0084]).
With respect to claim 7¸ the combination of Liang and Sharma teaches wherein a bit-width is to be selected from the bit-width probability distribution based on the plurality of probabilities and the bit-width is to be used to reduce precisions of quantizable parameters of the respective layer in the second DNN (Sharma, quantization reduces bit widths for data and operations in deep learning model. Both the data-representations and precision for activations, weights and gradients vary between different DNN models. The gradients for the convolution operations may require either high bit width fixed-point or floating-point datatypes depending on the quantization algorithm. the activations for the convolution operation and weights for the convolution operation may require low bit width fixed-point representation [par. 0073-0084]).
With respect to claim 8¸ the combination of Liang and Sharma teaches wherein another quantizable operation in the sequence comprises a convolution and quantizable parameters of the another quantizable operation comprise weights (Sharma, at operation 246, a convolution is performed on output from operation 240 and the second precision datatype weights from operation 242 to generate input loss at operation 248. Convolution uses low-bit width fixed-point data for activations and weights. Each convolution may require mixed precision data types for gradients [par. 0073-0084]).
With respect to claim 10¸ the combination of Liang and Sharma teaches wherein the multiple objectives are selected from a group consisting of maximizing task performance of the DNN, minimizing model size of the DNN, and minimizing compute complexity of the DNN (Liang, an important goal of DNN optimization is to minimize the complexity or size of a network while simultaneously maximizing its performance. MCDN can be used to maximize the performance and minimize the complexity of the evolved networks simultaneously [par. 0007, 0039]).
With respect to claim 11, it is a non-transitory computer-readable media claim that is corresponding to the method of claim 1. Therefore, it is rejected for the same reason as claimed in claim 1 above.
With respect to claim 12, it is a non-transitory computer-readable media claim that is corresponding to the method of claim 2. Therefore, it is rejected for the same reason as claimed in claim 2 above.
With respect to claim 13, it is a non-transitory computer-readable media claim that is corresponding to the method of claim 3. Therefore, it is rejected for the same reason as claimed in claim 3 above.
With respect to claim 14, it is a non-transitory computer-readable media claim that is corresponding to the method of claim 4. Therefore, it is rejected for the same reason as claimed in claim 4 above.
With respect to claim 15, it is a non-transitory computer-readable media claim that is corresponding to the method of claim 5. Therefore, it is rejected for the same reason as claimed in claim 5 above.
With respect to claim 16, it is a non-transitory computer-readable media claim that is corresponding to the method of claim 6. Therefore, it is rejected for the same reason as claimed in claim 6 above.
With respect to claim 17, it is a non-transitory computer-readable media claim that is corresponding to the method of claim 7. Therefore, it is rejected for the same reason as claimed in claim 7 above.
With respect to claim 18, it is a non-transitory computer-readable media claim that is corresponding to the method of claim 8. Therefore, it is rejected for the same reason as claimed in claim 8 above.
With respect to claim 20, it is a non-transitory computer-readable media claim that is corresponding to the method of claim 10. Therefore, it is rejected for the same reason as claimed in claim 10 above.
With respect to claim 21, it is an apparatus claim that is corresponding to the method of claim 1. Therefore, it is rejected for the same reason as claimed in claim 1 above.
With respect to claim 22, it is an apparatus claim that is corresponding to the method of claim 2. Therefore, it is rejected for the same reason as claimed in claim 2 above.
With respect to claim 23, it is an apparatus claim that is corresponding to the method of claim 4. Therefore, it is rejected for the same reason as claimed in claim 4 above.
With respect to claim 24, it is an apparatus claim that is corresponding to the method of claim 5. Therefore, it is rejected for the same reason as claimed in claim 5 above.
With respect to claim 25, it is an apparatus claim that is corresponding to the method of claim 6. Therefore, it is rejected for the same reason as claimed in claim 6 above.
Response to Arguments
Applicant’s arguments, see page 9 of the remark filed 06/30/2026, with respect to the rejection(s) of claim(s) 1-8, 10-18 and 20-25 under 35 U.S.C 103 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made (see rejection above).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Cavatassi et al (US 20230079744 A1) disclosed an apparatus for feature-based communications is provided that includes a probabilistic encoder and a transmitter. The probabilistic encoder is configured to encode source information into a set of probability distributions over a latent space. Each probability distribution represents one or more aspects of a subject of the source information. The transmitter is configured to transmit over a transmission channel, to a receiving electronic device, a set of transmission features representing the subject. Each transmission feature provides information about a respective one of the probability distributions in the latent space.
Zhang et al (US 20210158155 A1) disclosed a graph neural network for average power estimation of netlists is trained with register toggle rates over a power window from an RTL simulation and gate level netlists as input features. Combinational gate toggle rates are applied as labels. The trained graph neural network is then applied to infer combinational gate toggle rates over a different power window of interest and/or different netlist.
Lie et al (US 20200380344 A1) disclosed techniques in advanced deep learning provide improvements in one or more of accuracy, performance, and energy efficiency. An array of processing elements performs flow-based computations on wavelets of data. Each processing element has a respective compute element and a respective routing element. Each compute element has memory. At least a first single neuron is implemented using resources of a plurality of the array of processing elements. At least a portion of a second neuron is implemented using resources of one or more of the plurality of processing elements. In some usage scenarios, the foregoing neuron implementation enables greater performance by enabling a single neuron to use the computational resources of multiple processing elements and/or computational load balancing across the processing elements while maintaining locality of incoming activations for the processing elements.
Yang et al (US 20240232594 A1) disclosed methods for globally tuning and generating ML hardware accelerators. A design system selects an architecture representing a baseline processor configuration. An ML cost model of the system generates performance data about the architecture at least by modeling how the architecture executes computations of a neural network that includes multiple layers. Based on the performance data, the architecture is dynamically tuned to satisfy a performance objective when the architecture implements the neural network and executes machine-learning computations for a target application. In response to dynamically tuning the architecture, the system generates a configuration of an ML accelerator that specifies customized hardware configurations for implementing each of the multiple layers of the neural network.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Quoc Phung whose telephone number is (703) 756 1330. The examiner can normally be reached on Monday through Friday from 9am to 5pm PT.
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, Jennifer Welch can be reached on 571-272-7212. 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.
/Q.L.P./Examiner, Art Unit 2143
/JENNIFER N WELCH/Supervisory Patent Examiner, Art Unit 2143
Prosecution Insights