DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application is being examined under the pre-AIA first to invent provisions.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 05/31/2024 (3), 12/02/2024 (2), 03/14/2025 (2), 06/02/2025 (1), and 05/28/2026 (1) is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Rejections - 35 USC § 101
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 1-23 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
In reference to claim 1:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“A computer-implemented method for quantizing vectors of parameters to a bit width for processing by a neural network model, comprising: initializing a plurality of quantizers for the bit width;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could initialize a plurality of mental quantizers for the bit width.
“mapping the vectors into clusters based on quantization errors, wherein each one of the clusters is associated with one quantizer of the plurality of quantizers; and” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could map the vectors into clusters based on quantization errors.
“optimizing at least a portion of the quantizers in the plurality based on computed per- cluster quantization errors to produce optimized quantizers;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could optimize or adjust the mental quantizer models based on evaluation of quantization errors.
“quantizing the parameters of the vectors in each cluster using the optimized quantizer for the cluster to produce quantized vectors comprising quantized parameters at the bit width;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could quantize the parameters using the optimized mental quantizer to produce quantized vectors.
“and processing the quantized vectors by performing operations at the bit width [by a layer of the neural network model] to produce output values.”
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
“by a layer of the neural network model” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
“by a layer of the neural network model” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 2:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 1, wherein the parameters are at least one of weights or activations.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 3:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 1, wherein at least a portion of the parameters are activations that are dynamically calculated at each layer of the neural network model.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 4:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 1, wherein the parameters comprise vectors of elements within a single dimension of a multi-dimensional parameter tensor.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 5:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 1, wherein a Lloyd-Max algorithm is used to optimize each quantizer in the portion of the quantizers.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 6:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 1, wherein mapping the vectors into clusters based on quantization errors comprises, for each vector: computing the quantization errors resulting from quantizing each parameter in the vector using each quantizer to produce per-vector quantizer errors;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
“mapping the vector to the cluster associated with the one quantizer of the plurality of quantizers for which a minimal per-vector quantizer error is produced.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 7:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 1, wherein mapping the vectors into clusters based on quantization errors comprises, for each vector: determining a per-vector statistical proxy corresponding to quantization error, wherein each quantizer is associated with a quantizer statistical proxy definition; and” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
“mapping the vector to the cluster associated with the one quantizer of the plurality of quantizers producing a minimal difference between the per-vector statistical proxy and the quantizer statistical proxy definition.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 8:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 7, wherein the per-vector statistical proxy comprises at least one of a mean value of the parameters in the vector, an absolute mean value of the parameters in the vector, a median value of the parameters in the vector, an absolute median value of the parameters in the vector, or a maximum value and minimum value of the parameters in the vector.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 9:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 7, wherein initializing the plurality of quantizers comprises: for each vector, determining the per-vector statistical proxy;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
“identifying a centroid for each cluster by applying K-means clustering to the per- vector statistical proxies; and” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
“for each cluster, storing the centroid identified for the cluster as the quantizer statistical proxy definition for the quantizer associated with the cluster.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 10:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 9, wherein initializing the plurality of quantizers further comprises: optimizing at least a portion of the quantizers in the plurality based on computed initial per-cluster quantization errors to produce optimized quantizers; and” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
“for each cluster, storing a quantizer definition for the optimized quantizer associated with the cluster.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 11:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 1, wherein initializing the plurality of quantizers comprises setting the quantization levels to at least one of random values, pre-calibrated definitions, or precomputed definitions.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 12:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 1, wherein the quantization errors are mean-squared errors.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 13:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 1, further comprising, before the quantizing: replacing at least one quantizer in the plurality with one of the optimized quantizers;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
“and repeating the mapping and optimizing to update the optimized quantizers.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 14:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 13, further comprising determining that a criterion is met after the repeating and before the processing.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 15:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The computer-implemented method of claim 1, wherein at least one of the quantizers in the plurality quantizes using a fixed number format including one of a floating- point format including a single bit of exponent, a floating-point format including a single bit of mantissa, a four-bit integer format, or a bias-adjusted format.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 16:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
No
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
“The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed on a server or in a data center and the quantized vectors or the output values are streamed to a user device.” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
“The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed on a server or in a data center and the quantized vectors or the output values are streamed to a user device.” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 17:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
No
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
“The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed within a cloud computing environment.” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
“The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed within a cloud computing environment.” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 18:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
No
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
“The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed for training, testing, or certifying an additional neural network model employed in a machine, robot, or autonomous vehicle.” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
“The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed for training, testing, or certifying an additional neural network model employed in a machine, robot, or autonomous vehicle.” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 19:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
No
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
“The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed on a virtual machine comprising a portion of a graphics processing unit.” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
“The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed on a virtual machine comprising a portion of a graphics processing unit.” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 20:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a machine
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“initializing a plurality of quantizers for the bit width;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could initialize a plurality of mental quantizers for the bit width.
“mapping the vectors into clusters based on quantization errors, wherein each one of the clusters is associated with one quantizer of the plurality of quantizers; and” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could map the vectors into clusters based on quantization errors.
“optimizing at least a portion of the quantizers in the plurality based on computed per- cluster quantization errors to produce optimized quantizers;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could optimize or adjust the mental quantizer models based on evaluation of quantization errors.
“quantizing the parameters of the vectors in each cluster using the optimized quantizer for the cluster to produce quantized vectors comprising quantized parameters at the bit width;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could quantize the parameters using the optimized mental quantizer to produce quantized vectors.
“and processing the quantized vectors by performing operations at the bit width [by a layer of the neural network model] to produce output values.”
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
“A system for quantizing vectors of parameters to a bit width, comprising: a memory that stores at least a portion of the vectors; and one or more processors coupled to the memory to perform operations including:” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“by a layer of the neural network model” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
“A system for quantizing vectors of parameters to a bit width, comprising: a memory that stores at least a portion of the vectors; and one or more processors coupled to the memory to perform operations including:” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“by a layer of the neural network model” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 21:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a machine
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The system of claim 20, wherein a Lloyd-Max algorithm is used to optimize each quantizer in the portion of the quantizers.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 22:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a manufacture
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“initializing a plurality of quantizers for the bit width;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could initialize a plurality of mental quantizers for the bit width.
“mapping the vectors into clusters based on quantization errors, wherein each one of the clusters is associated with one quantizer of the plurality of quantizers; and” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could map the vectors into clusters based on quantization errors.
“optimizing at least a portion of the quantizers in the plurality based on computed per- cluster quantization errors to produce optimized quantizers;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could optimize or adjust the mental quantizer models based on evaluation of quantization errors.
“quantizing the parameters of the vectors in each cluster using the optimized quantizer for the cluster to produce quantized vectors comprising quantized parameters at the bit width;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could quantize the parameters using the optimized mental quantizer to produce quantized vectors.
“and processing the quantized vectors by performing operations at the bit width [by a layer of the neural network model] to produce output values.”
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
“A non-transitory computer-readable media storing computer instructions for quantizing vectors of parameters to a bit width for processing by a neural network model that, when executed by one or more processors, cause the one or more processors to perform the steps of:” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“by a layer of the neural network model” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
“A non-transitory computer-readable media storing computer instructions for quantizing vectors of parameters to a bit width for processing by a neural network model that, when executed by one or more processors, cause the one or more processors to perform the steps of:” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“by a layer of the neural network model” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 23:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a manufacture
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The non-transitory computer-readable media of claim 22,further comprising, before the quantizing: replacing at least one quantizer in the plurality with one of the optimized quantizers;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
“and repeating the mapping and optimizing to update the optimized quantizers.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 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.
Claim(s) 1-5, 12-15, 17-18, and 20-23 are rejected under 35 U.S.C. 103 as being unpatentable over Volodya Grancharov et al; US 20150051907 A1 filed Dec 12, 2012 (hereinafter “Grancharov”) in view of Ruiqi Guo et al; US 20210064634 A1 filed on Aug 25, 2020 (hereinafter “Guo”) in further view of Nikolaos Freris et al; US 20130031063 A1 (hereinafter “Freris”) in further view of Daniel Lo et al; US 20200264876 A1 filed on Feb 14, 2019 (hereinafter “Lo”)
Regarding claim 1, Grancharov teaches A computer-implemented method for quantizing vectors of parameters to a bit width for processing by a neural network model, comprising: initializing a plurality of quantizers for the bit width; (Grancharov Paragraph 0036; “In order to create a basic version of the specially designed advantageous CB, codevectors of a CB are split into two classes, here denoted C.sub.0 and C.sub.1 (this notation will be used both for the names of the classes, as well as for the corresponding centroids, cf. FIG. 1). To partition data into two classes a so-called K-means algorithm (Generalized Lloyd's algorithm) may be used. It is a well known technique, which takes an entire data set as input, and the desired number of classes, and outputs the centroids of the desired number of classes.” Grancharov Paragraph 0047; “The numbers in the table in FIG. 5 are derived under the assumption that the CB from FIG. 2 comprises four 7-bit segments (four segments with 128 codevectors each) of which two are "normal" or "physical" and two are "flipped" or "virtual").” Examiner notes a plurality of quantizers (centroids for each class in codebook used to quantize is the quantizer) is initialized/created for the bit width (7-bit segments))
mapping the vectors into clusters based on quantization errors, wherein each one of the clusters is associated with one quantizer of the plurality of quantizers; and (Grancharov Paragraph 0060; “The input target vector s is assigned, or concluded to belong to, the class to which it has the shortest distance, i.e. to which it is most similar, according to some distance measure (error measure).” Examiner notes that the vector (input vector) is mapped/assigned to the cluster associated with the one quantizer of the plurality of quantizers(class within codebook used by the quantizer) based on quantization errors (error measure between input vector and centroid reference vector of class))
Grancharov does not teach optimizing at least a portion of the quantizers in the plurality based on computed per- cluster quantization errors to produce optimized quantizers;
However, Guo does teach optimizing at least a portion of the quantizers in the plurality based on computed per- cluster quantization errors to produce optimized quantizers; (Guo Paragraph 0055; “The objective may be resolved through a k-Means style Lloyd's algorithm, which iteratively minimizes the new loss functions by assigning datapoints to partitions and updating the partition quantizer in each iteration.” Examiner notes that Lloyd-Max algorithm is used to optimize each quantizer in the portion of the quantizer (the partition quantizer; portion of the quantizer is interpreted to have 1 quantizer) based on computed per-cluster quantization errors (to minimizes the new loss functions))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov and Guo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. One of ordinary skill would have motivation to combine Grancharov and Guo to improves the ability of a machine learning model to perform a task “As one example, the techniques described herein enable quantization of a dataset according to a loss function that improves, relative to use of traditional loss functions, the ability of a machine-learned model to perform a task (e.g., an image processing, computer vision task, sensor data processing task, audio processing task, text processing task, classification task, detection task, recognition task, data search task, etc.).” (Guo Paragraph 0024).
Grancharov in view of Guo does not teach quantizing the parameters of the vectors in each cluster using the optimized quantizer for the cluster to produce quantized vectors comprising quantized parameters at the bit width;
However, Freris does teach quantizing the parameters of the vectors in each cluster using the optimized quantizer for the cluster to produce quantized vectors comprising quantized parameters at the bit width; (Freris Paragraph 0037; “For each cluster, the MMSE quantization maps the pieces of data of the cluster to a smaller set of quantizers. The number of scalar quantizers used is fixed and equals the dimension of the data, i.e., each dimension is separately quantized by a scalar quantizer. The number of quantization levels depends on the contemplated compression rate. The values of the scalar quantizers are determined so as to minimize the sum of mean square errors (i.e., the square distance between the values of the pieces of data of cluster to the quantization level to which they are mapped). Such a quantization can be performed simply and offers a good trade-off between the compression rate and the preservation of quality of the data.” Freris Paragraph 0038; “The MMSE quantization may be 1-bit or multi-bit” Examiner notes that the parameter of the vectors in each cluster (pieces of data of the cluster for each cluster) is quantized using the optimized quantizer for the cluster (data is mapped to quantizer to minimize the sum of mean square errors) to produce quantized vectors comprising quantized parameters at the bit width (quantized dataset at the 1-bit or multi-bit))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, and Freris. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. One of ordinary skill would have motivation to combine Grancharov, Guo, and Freris to reduce the space taken by data to improve the computer “the method reduces the space taken by the data on the memory of a computer that executes the method. The method thereby improves the use of the computer, from a hardware point of view.” (Freris Paragraph 0013).
Grancharov in view of Guo in further view of Freris does not teach and processing the quantized vectors by performing operations at the bit width by a layer of the neural network model to produce output values.
However, Lo does teach and processing the quantized vectors by performing operations at the bit width by a layer of the neural network model to produce output values. (Lo Paragraph 0052; “An input tensor for the given layer can be converted from a normal-precision floating-point format to a quantized-precision floating-point format. A tensor operation can be performed using the converted input tensor having lossy or non-uniform mantissas. For example, during a back-propagation mode, the input tensor can be an output error term from a layer adjacent to (e.g., following) the given layer or weights of the given layer.” Examiner notes that the quantized vectors (quantized input tensor) is processed by performing operations at the bit width (a tensor operation using normal-precision floating-point format of tensor) by a layer of the neural network model (given layer) to produce output values (output error term))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, and Lo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. One of ordinary skill would have motivation to combine Grancharov, Guo, Freris, and Lo to allow for faster hardware, reduce memory overhead, and reduce energy use “Use of quantized formats can improve neural network processing by, for example, allowing for faster hardware, reduced memory overhead, simpler hardware design, reduced energy use, reduced integrated circuit area, cost savings and other technological improvements.” (Lo Paragraph 0051).
Regarding claim 2, Grancharov in view of Guo in further view of Freris does not teach The computer-implemented method of claim 1, wherein the parameters are at least one of weights or activations.
However, Lo does teach The computer-implemented method of claim 1, wherein the parameters are at least one of weights or activations. (Lo Paragraph 0176; “The weights for the given layer can be communicated from the normal-precision floating-point domain to the quantized floating-point domain through the quantizer 1442.” Examiner notes that parameters are weights)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, and Lo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. One of ordinary skill would have motivation to combine Grancharov, Guo, Freris, and Lo to allow for faster hardware, reduce memory overhead, and reduce energy use “Use of quantized formats can improve neural network processing by, for example, allowing for faster hardware, reduced memory overhead, simpler hardware design, reduced energy use, reduced integrated circuit area, cost savings and other technological improvements.” (Lo Paragraph 0051).
Regarding claim 3, Grancharov in view of Guo in further view of Freris does not teach The computer-implemented method of claim 1, wherein at least a portion of the parameters are activations that are dynamically calculated at each layer of the neural network model.
However, Lo does teach The computer-implemented method of claim 1, wherein at least a portion of the parameters are activations that are dynamically calculated at each layer of the neural network model. (Lo Paragraph 0054; “software functions can be provided that allow applications to define neural networks including weights, biases, activation functions, node values, and interconnections between layers of a neural network.” Lo Paragraph 0057; “during training forward propagation, once activation values for a next layer in the NN have been calculated, those values may not be accessed until for propagation through all layers has completed. Such activation values can be stored in such a bulk memory.” Examiner notes that the parameters are activations (activation values) are dynamically calculated at each layer of the neural network (activation functions are used to obtain activation values at each layer))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, and Lo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. One of ordinary skill would have motivation to combine Grancharov, Guo, Freris, and Lo to allow for faster hardware, reduce memory overhead, and reduce energy use “Use of quantized formats can improve neural network processing by, for example, allowing for faster hardware, reduced memory overhead, simpler hardware design, reduced energy use, reduced integrated circuit area, cost savings and other technological improvements.” (Lo Paragraph 0051).
Regarding claim 4, Grancharov in view of Guo in further view of Freris does not teach The computer-implemented method of claim 1, wherein the parameters comprise vectors of elements within a single dimension of a multi-dimensional parameter tensor.
However, Lo does teach The computer-implemented method of claim 1, wherein the parameters comprise vectors of elements within a single dimension of a multi-dimensional parameter tensor. (Lo Paragraph 0045; “the term “tensor” refers to a multi-dimensional array that can be used to represent properties of a NN and includes one-dimensional vectors as well as two-, three-, four-, or larger dimension matrices.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, and Lo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. One of ordinary skill would have motivation to combine Grancharov, Guo, Freris, and Lo to allow for faster hardware, reduce memory overhead, and reduce energy use “Use of quantized formats can improve neural network processing by, for example, allowing for faster hardware, reduced memory overhead, simpler hardware design, reduced energy use, reduced integrated circuit area, cost savings and other technological improvements.” (Lo Paragraph 0051).
Regarding claim 5, Grancharov does not teach The computer-implemented method of claim 1, wherein a Lloyd-Max algorithm is used to optimize each quantizer in the portion of the quantizers.
However, Guo does teach The computer-implemented method of claim 1, wherein a Lloyd-Max algorithm is used to optimize each quantizer in the portion of the quantizers. (Guo Paragraph 0055; “The objective may be resolved through a k-Means style Lloyd's algorithm, which iteratively minimizes the new loss functions by assigning datapoints to partitions and updating the partition quantizer in each iteration.” Examiner notes that Lloyd-Max algorithm is used to optimize each quantizer in the portion of the quantizer (the partition quantizer; portion of the quantizer is interpreted to have 1 quantizer))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov and Guo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. One of ordinary skill would have motivation to combine Grancharov and Guo to improves the ability of a machine learning model to perform a task “As one example, the techniques described herein enable quantization of a dataset according to a loss function that improves, relative to use of traditional loss functions, the ability of a machine-learned model to perform a task (e.g., an image processing, computer vision task, sensor data processing task, audio processing task, text processing task, classification task, detection task, recognition task, data search task, etc.).” (Guo Paragraph 0024).
Regarding claim 12, Grancharov does not teach The computer-implemented method of claim 1, wherein the quantization errors are mean-squared errors.
However, Guo does teach The computer-implemented method of claim 1, wherein the quantization errors are mean-squared errors. (Guo Paragraph 0073; “Various loss functions can be used such as mean squared error”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov and Guo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. One of ordinary skill would have motivation to combine Grancharov and Guo to improves the ability of a machine learning model to perform a task “As one example, the techniques described herein enable quantization of a dataset according to a loss function that improves, relative to use of traditional loss functions, the ability of a machine-learned model to perform a task (e.g., an image processing, computer vision task, sensor data processing task, audio processing task, text processing task, classification task, detection task, recognition task, data search task, etc.).” (Guo Paragraph 0024).
Regarding claim 13, Grancharov does not teach The computer-implemented method of claim 1, further comprising, before the quantizing: replacing at least one quantizer in the plurality with one of the optimized quantizers;
However, Guo does teach The computer-implemented method of claim 1, further comprising, before the quantizing: replacing at least one quantizer in the plurality with one of the optimized quantizers; (Guo Paragraph 0055; “The objective may be resolved through a k-Means style Lloyd's algorithm, which iteratively minimizes the new loss functions by assigning datapoints to partitions and updating the partition quantizer in each iteration.” Examiner notes that updating the partition quantizer in each iteration is replacing the at least one quantizer in the plurality with the optimized/updated quantizer)
and repeating the mapping and optimizing to update the optimized quantizers. (Examiner refers to previous mapping to show that the mapping and optimizing (k-Means style Lloyd's algorithm) is repeated (iteratively performed) to update the optimized quantizers (updating the partition quantizer in each iteration))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov and Guo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. One of ordinary skill would have motivation to combine Grancharov and Guo to improves the ability of a machine learning model to perform a task “As one example, the techniques described herein enable quantization of a dataset according to a loss function that improves, relative to use of traditional loss functions, the ability of a machine-learned model to perform a task (e.g., an image processing, computer vision task, sensor data processing task, audio processing task, text processing task, classification task, detection task, recognition task, data search task, etc.).” (Guo Paragraph 0024).
Regarding claim 14, Grancharov does not teach The computer-implemented method of claim 13, further comprising determining that a criterion is met after the repeating and before the processing.
However, Guo does teach The computer-implemented method of claim 13, further comprising determining that a criterion is met after the repeating and before the processing. (Guo Paragraph 0056; “Since Equation (28) is a convex function of {tilde over (x)}, there exists an optimal solution for Equation (28). The update rule given a fixed partitioning can be found by setting the partial derivative of Equation (28) with respect to each codebook entry to zero. This algorithm provably converges in a finite number of steps.” Examiner notes that a criterion (partial derivative of Equation (28) with respect to each codebook entry to zero) is determined to be met/converges after the repeating and the before the processing)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov and Guo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. One of ordinary skill would have motivation to combine Grancharov and Guo to improves the ability of a machine learning model to perform a task “As one example, the techniques described herein enable quantization of a dataset according to a loss function that improves, relative to use of traditional loss functions, the ability of a machine-learned model to perform a task (e.g., an image processing, computer vision task, sensor data processing task, audio processing task, text processing task, classification task, detection task, recognition task, data search task, etc.).” (Guo Paragraph 0024).
Regarding claim 15, Grancharov does not teach The computer-implemented method of claim 1, wherein at least one of the quantizers in the plurality quantizes using a fixed number format including one of a floating- point format including a single bit of exponent, a floating-point format including a single bit of mantissa, a four-bit integer format, or a bias-adjusted format.
However, Lo does teach The computer-implemented method of claim 1, wherein at least one of the quantizers in the plurality quantizes using a fixed number format including one of a floating- point format including a single bit of exponent, a floating-point format including a single bit of mantissa, a four-bit integer format, or a bias-adjusted format. (Lo Paragraph 0108; “Q2( ) is a quantization function to further compress values to a second floating-point format (for example, a format having a fewer number of mantissa bits, a fewer number of exponent bits, a different exponent sharing scheme, having lossy or non-uniform mantissas, or outlier mantissa values for some of the values)” Examiner notes that one of the quantizers in the plurality of quantizers uses a fixed number format including one of a floating point format including a single bit of exponent (a format having a fewer number of mantissa bits, a fewer number of exponent bits (ie including single bit exponent)))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, and Lo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. One of ordinary skill would have motivation to combine Grancharov, Guo, Freris, and Lo to allow for faster hardware, reduce memory overhead, and reduce energy use “Use of quantized formats can improve neural network processing by, for example, allowing for faster hardware, reduced memory overhead, simpler hardware design, reduced energy use, reduced integrated circuit area, cost savings and other technological improvements.” (Lo Paragraph 0051).
Regarding claim 17, Grancharov does not teach The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed within a cloud computing environment.
However, Lo does teach The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed within a cloud computing environment. (Lo Paragraph 0220; “For example, the disclosed methods can be executed on processing units 2110 located in the computing environment 2130, or the disclosed methods can be executed on servers located in the computing cloud 2190.” Examiner notes that the steps is performed within a cloud computing environment (computing cloud))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, and Lo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. One of ordinary skill would have motivation to combine Grancharov, Guo, Freris, and Lo to allow for faster hardware, reduce memory overhead, and reduce energy use “Use of quantized formats can improve neural network processing by, for example, allowing for faster hardware, reduced memory overhead, simpler hardware design, reduced energy use, reduced integrated circuit area, cost savings and other technological improvements.” (Lo Paragraph 0051).
Regarding claim 18, Grancharov does not teach The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed within a cloud computing environment.
However, Guo does teach The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed for training, testing, or certifying an additional neural network model employed in a machine, robot, or autonomous vehicle. (Guo Paragraph 0071; “The user computing device 102 and/or the server computing system 130 can train the models 120 and/or 140 via interaction with the training computing system 150 that is communicatively coupled over the network 180.” Guo Paragraph 0078; “one or more of the machine-learned models 120 and 140 may include parameter values quantized according to the present disclosure. For example, model inputs, weights, activations, scale factors, biases, accumulators, outputs, and/or other parameter values used in a model may be quantized according to the present disclosure.” Examiner notes that the steps is performed for training an additional neural network (model 120) employed in a machine (user computing system))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov and Guo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. One of ordinary skill would have motivation to combine Grancharov and Guo to improves the ability of a machine learning model to perform a task “As one example, the techniques described herein enable quantization of a dataset according to a loss function that improves, relative to use of traditional loss functions, the ability of a machine-learned model to perform a task (e.g., an image processing, computer vision task, sensor data processing task, audio processing task, text processing task, classification task, detection task, recognition task, data search task, etc.).” (Guo Paragraph 0024).
Regarding claim 20, Grancharov teaches A system for quantizing vectors of parameters to a bit width, comprising: a memory that stores at least a portion of the vectors; and one or more processors coupled to the memory to perform operations including: (Grancharov Paragraph 0044; “FIG. 2 illustrates a CB where the codevectors of the left part are stored in a memory” Grancharov Paragraph 0071; “Furthermore the arrangement 1200 comprises at least one computer program product 1208 in the form of a non-volatile memory, e.g. an EEPROM, a flash memory and a hard drive. The computer program product 1208 comprises a computer program 1210, which comprises code means, which when run in the processing unit 1206 in the arrangement 1200 causes the arrangement to perform the actions of a procedure described earlier in conjunction with FIGS. 9a-c.”)
Claim 20 is machine claim of method claim 1 and is accordingly rejected using substantially similar rationale as to that which is set for with respect to claim 1.
Regarding claim 21, claim 21 has similar limitations as of claim 5, except it is a machine claim (Grancharov 0044, 0071), therefore it is rejected under the same rationale as claim 5.
Regarding claim 22, Grancharov teaches A non-transitory computer-readable media storing computer instructions for quantizing vectors of parameters to a bit width for processing by a neural network model that, when executed by one or more processors, cause the one or more processors to perform the steps of: (Grancharov Paragraph 0071; “Furthermore the arrangement 1200 comprises at least one computer program product 1208 in the form of a non-volatile memory, e.g. an EEPROM, a flash memory and a hard drive. The computer program product 1208 comprises a computer program 1210, which comprises code means, which when run in the processing unit 1206 in the arrangement 1200 causes the arrangement to perform the actions of a procedure described earlier in conjunction with FIGS. 9a-c.”)
Claim 22 is manufacture claim of method claim 1 and is accordingly rejected using substantially similar rationale as to that which is set for with respect to claim 1.
Regarding claim 23, claim 23 has similar limitations as of claim 13, except it is a manufacture claim (Grancharov 0071), therefore it is rejected under the same rationale as claim 13.
Claim(s) 6 is rejected under 35 U.S.C. 103 as being unpatentable over Volodya Grancharov et al; US 20150051907 A1 filed Dec 12, 2012 (hereinafter “Grancharov”) in view of Ruiqi Guo et al; US 20210064634 A1 filed on Aug 25, 2020 (hereinafter “Guo”) in further view of Nikolaos Freris et al; US 20130031063 A1 (hereinafter “Freris”) in further view of Daniel Lo et al; US 20200264876 A1 filed on Feb 14, 2019 (hereinafter “Lo”) in further view of Jes Thyssen et al; US 20030135367 A1 filed on Aug 12, 2002 (hereinafter “Thyssen”)
Regarding claim 6, Grancharov teaches mapping the vector to the cluster associated with the one quantizer of the plurality of quantizers for which a minimal per-vector quantizer error is produced. (Grancharov Paragraph 0060; “The input target vector s is assigned, or concluded to belong to, the class to which it has the shortest distance, i.e. to which it is most similar, according to some distance measure (error measure).” Examiner notes that the vector (input vector) is mapped/assigned to the cluster associated with the one quantizer of the plurality of quantizers(class within codebook used by the quantizer) for which a minimal per-vector quantizer error is produced (distance measure (error measure) between input vector and centroid reference vector of class))
Grancharov does not teach The computer-implemented method of claim 1, wherein mapping the vectors into clusters based on quantization errors comprises, for each vector: computing the quantization errors resulting from quantizing each parameter in the vector using each quantizer to produce per-vector quantizer errors; and
However, Thyssen does teach The computer-implemented method of claim 1, wherein mapping the vectors into clusters based on quantization errors comprises, for each vector: computing the quantization errors resulting from quantizing each parameter in the vector using each quantizer to produce per-vector quantizer errors; and (Thyssen Paragraph 0163; “Quantizer 1058 quantizes input signal u(n) to produce a quantized signal uq(n) (also referred to as a quantizer output signal) associated with a quantization noise or error signal q(n). Combiner 1064 combines (that is, differences) signals u(n) and uq(n) to produce the quantization error or noise signal q(n).” Thyssen Paragraph 0244; “Many different kinds of LSP quantizers can be used in block 16.” Thyssen Paragraph 0334; “Each of the VQ input vectors u(n) corresponds to one of N VQ error vectors q(n).” Examiner notes that the quantization errors (error signal q(n)) is computed from quantizing each parameter in the vector (input signal u(n)) using each quantizer (quantizer of LSP quantizers) to produce per-vector quantizer errors (N VQ error vectors q(n)))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, Lo, and Thyssen. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. Thyssen teaches a method for efficient excitation quantization in noise feedback coding. One of ordinary skill would have motivation to Grancharov, Guo, Freris, Lo, and Thyssen to minimize mathematical operations to minimize power consumption and maximize processing bandwidth “It is advantageous to minimize the number of mathematical operations in order to minimize a power consumption, and maximize a processing bandwidth, of the signal processing device.” (Thyssen Paragraph 0008).
Claim(s) 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over Volodya Grancharov et al; US 20150051907 A1 filed Dec 12, 2012 (hereinafter “Grancharov”) in view of Ruiqi Guo et al; US 20210064634 A1 filed on Aug 25, 2020 (hereinafter “Guo”) in further view of Nikolaos Freris et al; US 20130031063 A1 (hereinafter “Freris”) in further view of Daniel Lo et al; US 20200264876 A1 filed on Feb 14, 2019 (hereinafter “Lo”) in further view of Vilim Bartolucci et al; US 20030072366 A1 filed on Sep 20, 2002 (hereinafter “Bartolucci”)
Regarding claim 7, Grancharov teaches wherein each quantizer is associated with a quantizer statistical proxy definition; and (Grancharov Paragraph 0036; “In order to create a basic version of the specially designed advantageous CB, codevectors of a CB are split into two classes, here denoted C.sub.0 and C.sub.1 (this notation will be used both for the names of the classes, as well as for the corresponding centroids, cf. FIG. 1). To partition data into two classes a so-called K-means algorithm (Generalized Lloyd's algorithm) may be used. It is a well known technique, which takes an entire data set as input, and the desired number of classes, and outputs the centroids of the desired number of classes.” Examiner notes that each quantizer (Vector Quantizer) is associated with a quantizer statistical proxy definition (codebook with code vectors))
mapping the vector to the cluster associated with the one quantizer of the plurality of quantizers producing a minimal difference between the per-vector statistical proxy and the quantizer statistical proxy definition. (Grancharov Paragraph 0060; “The input target vector s is assigned, or concluded to belong to, the class to which it has the shortest distance, i.e. to which it is most similar, according to some distance measure (error measure).” Examiner notes that the vector (input vector) is mapped/assigned to the cluster associated with the one quantizer of the plurality of quantizers(class within codebook used by the quantizer) producing a minimal difference between the per-vector statistical proxy and the quantizer statistical proxy definition (distance measure (error measure) between input vector and centroid reference vector of class))
Grancharov does not teach The computer-implemented method of claim 1, wherein mapping the vectors into clusters based on quantization errors comprises, for each vector: determining a per-vector statistical proxy corresponding to quantization error,
However, Bartolucci does teach The computer-implemented method of claim 1, wherein mapping the vectors into clusters based on quantization errors comprises, for each vector: determining a per-vector statistical proxy corresponding to quantization error, (Bartolucci Paragraph 0121; “The discretized absolute mean value of the vector X.sub.i is obtained by multiplying the value .mu..sub.q by the pitch used previously.” Bartolucci Paragraph 0127; “Experiments show that, once a certain scale factor has been exceeded, it is no longer convenient to transmit the vector because the quantization error is much greater than the predicted error.” Examiner notes that a per-vector statistical proxy (absolute mean of the vector) is determined/obtained where it is corresponding to quantization error (paragraph 127 shows that the mean value is corresponding to a vector and the vector is corresponding to a quantization error))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, Lo, and Bartolucci. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. Bartolucci teaches a method for compression of digital video signals. One of ordinary skill would have motivation to Grancharov, Guo, Freris, Lo, and Bartolucci to reduce the redundancy in data and discard information not important for more efficient processing “To obtain a high compression ratio it is of fundamental importance to reduce the redundancy in the signal and to discard the information that is not important for the user.” (Bartolucci Paragraph 0006).
Regarding claim 8, Grancharov does not teach The computer-implemented method of claim 7, wherein the per-vector statistical proxy comprises at least one of a mean value of the parameters in the vector, an absolute mean value of the parameters in the vector, a median value of the parameters in the vector, an absolute median value of the parameters in the vector, or a maximum value and minimum value of the parameters in the vector.
However, Bartolucci does teach The computer-implemented method of claim 7, wherein the per-vector statistical proxy comprises at least one of a mean value of the parameters in the vector, an absolute mean value of the parameters in the vector, a median value of the parameters in the vector, an absolute median value of the parameters in the vector, or a maximum value and minimum value of the parameters in the vector. (Bartolucci Paragraph 0121; “The discretized absolute mean value of the vector X.sub.i is obtained by multiplying the value .mu..sub.q by the pitch used previously.” Bartolucci Paragraph 0127; “Experiments show that, once a certain scale factor has been exceeded, it is no longer convenient to transmit the vector because the quantization error is much greater than the predicted error.” Examiner notes that a per-vector statistical proxy (absolute mean of the vector) comprises the absolute mean value of the parameters in the vector)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, Lo, and Bartolucci. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. Bartolucci teaches a method for compression of digital video signals. One of ordinary skill would have motivation to Grancharov, Guo, Freris, Lo, and Bartolucci to reduce the redundancy in data and discard information not important for more efficient processing “To obtain a high compression ratio it is of fundamental importance to reduce the redundancy in the signal and to discard the information that is not important for the user.” (Bartolucci Paragraph 0006).
Regarding claim 9, Grancharov teaches identifying a centroid for each cluster by applying K-means clustering to the per- vector statistical proxies; and (Grancharov Paragraph 0036; “In order to create a basic version of the specially designed advantageous CB, codevectors of a CB are split into two classes, here denoted C.sub.0 and C.sub.1 (this notation will be used both for the names of the classes, as well as for the corresponding centroids, cf. FIG. 1). To partition data into two classes a so-called K-means algorithm (Generalized Lloyd's algorithm) may be used. It is a well known technique, which takes an entire data set as input, and the desired number of classes, and outputs the centroids of the desired number of classes.” Examiner notes that a centroid (centroids of the desired number of classes) are identified/outputted for each cluster (class) by applying k-means clustering to the per-vector proxies (input))
for each cluster, storing the centroid identified for the cluster as the quantizer statistical proxy definition for the quantizer associated with the cluster. (Examiner refers to previous mapping to show that for each cluster (class) the centroid identified for the cluster (centroid) is stored as the quantizer statistical proxy definition (codebook with centroid reference vector) for the quantizer associated with the cluster (Vector Quantizer using the codebook))
Grancharov does not teach The computer-implemented method of claim 7, wherein initializing the plurality of quantizers comprises: for each vector, determining the per-vector statistical proxy;
However, Bartolucci does teach The computer-implemented method of claim 7, wherein initializing the plurality of quantizers comprises: for each vector, determining the per-vector statistical proxy; (Bartolucci Paragraph 0121; “The discretized absolute mean value of the vector X.sub.i is obtained by multiplying the value .mu..sub.q by the pitch used previously.” Bartolucci Paragraph 0127; “Experiments show that, once a certain scale factor has been exceeded, it is no longer convenient to transmit the vector because the quantization error is much greater than the predicted error.” Examiner notes that a per-vector statistical proxy (absolute mean of the vector) is determined/obtained for each vector)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, Lo, and Bartolucci. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. Bartolucci teaches a method for compression of digital video signals. One of ordinary skill would have motivation to Grancharov, Guo, Freris, Lo, and Bartolucci to reduce the redundancy in data and discard information not important for more efficient processing “To obtain a high compression ratio it is of fundamental importance to reduce the redundancy in the signal and to discard the information that is not important for the user.” (Bartolucci Paragraph 0006).
Regarding claim 10, Grancharov teaches for each cluster, storing a quantizer definition for the optimized quantizer associated with the cluster. (Grancharov Paragraph 0036; “In order to create a basic version of the specially designed advantageous CB, codevectors of a CB are split into two classes, here denoted C.sub.0 and C.sub.1 (this notation will be used both for the names of the classes, as well as for the corresponding centroids, cf. FIG. 1). To partition data into two classes a so-called K-means algorithm (Generalized Lloyd's algorithm) may be used. It is a well known technique, which takes an entire data set as input, and the desired number of classes, and outputs the centroids of the desired number of classes.” Examiner notes that for each cluster (class) the centroid identified for the cluster (centroid) is stored as the quantizer statistical proxy definition (codebook with centroid reference vector) for the optimized quantizer associated with the cluster (Vector Quantizer using the codebook))
Grancharov does not teach The computer-implemented method of claim 9, wherein initializing the plurality of quantizers further comprises: optimizing at least a portion of the quantizers in the plurality based on computed initial per-cluster quantization errors to produce optimized quantizers; and
However, Guo does teach The computer-implemented method of claim 9, wherein initializing the plurality of quantizers further comprises: optimizing at least a portion of the quantizers in the plurality based on computed initial per-cluster quantization errors to produce optimized quantizers; and (Guo Paragraph 0055; “The objective may be resolved through a k-Means style Lloyd's algorithm, which iteratively minimizes the new loss functions by assigning datapoints to partitions and updating the partition quantizer in each iteration.” Examiner notes that Lloyd-Max algorithm is used to optimize each quantizer in the portion of the quantizer (the partition quantizer; portion of the quantizer is interpreted to have 1 quantizer) based on computed per-cluster quantization errors (to minimizes the new loss functions))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov and Guo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. One of ordinary skill would have motivation to combine Grancharov and Guo to improves the ability of a machine learning model to perform a task “As one example, the techniques described herein enable quantization of a dataset according to a loss function that improves, relative to use of traditional loss functions, the ability of a machine-learned model to perform a task (e.g., an image processing, computer vision task, sensor data processing task, audio processing task, text processing task, classification task, detection task, recognition task, data search task, etc.).” (Guo Paragraph 0024).
Claim(s) 11 is rejected under 35 U.S.C. 103 as being unpatentable over Volodya Grancharov et al; US 20150051907 A1 filed Dec 12, 2012 (hereinafter “Grancharov”) in view of Ruiqi Guo et al; US 20210064634 A1 filed on Aug 25, 2020 (hereinafter “Guo”) in further view of Nikolaos Freris et al; US 20130031063 A1 (hereinafter “Freris”) in further view of Daniel Lo et al; US 20200264876 A1 filed on Feb 14, 2019 (hereinafter “Lo”) in further view of Augustin Ion Gavrilescu et al; US 20050027521 A1 filed on Mar 30, 2004 (hereinafter “Gavrilescu”)
Regarding claim 11, Grancharov does not teach The computer-implemented method of claim 1, wherein initializing the plurality of quantizers comprises setting the quantization levels to at least one of random values, pre-calibrated definitions, or precomputed definitions.
However, Gavrilescu does teach The computer-implemented method of claim 1, wherein initializing the plurality of quantizers comprises setting the quantization levels to at least one of random values, pre-calibrated definitions, or precomputed definitions. (Gavrilescu Paragraph 0008; “The number of quantization levels may be freely selected e.g. seven or more or ten or more levels.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, Lo, and Gavrilescu. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. Gavrilescu teaches the method comprises quantizing a source digital signal to generate with different quantizations at least a first and a second bit-stream. One of ordinary skill would have motivation to combine Grancharov, Guo, Freris, Lo, and Gavrilescu to improve the error resilience of the model “The employed mechanism enables the control of the tradeoff between the coding efficiency and error-resilience, and provides an increased robustness by improving the error resilience in the most important layers of the embedded bit-streams.” (Gavrilescu Paragraph 0015).
Claim(s) 16 is rejected under 35 U.S.C. 103 as being unpatentable over Volodya Grancharov et al; US 20150051907 A1 filed Dec 12, 2012 (hereinafter “Grancharov”) in view of Ruiqi Guo et al; US 20210064634 A1 filed on Aug 25, 2020 (hereinafter “Guo”) in further view of Nikolaos Freris et al; US 20130031063 A1 (hereinafter “Freris”) in further view of Daniel Lo et al; US 20200264876 A1 filed on Feb 14, 2019 (hereinafter “Lo”) in further view of David Young Joon Pio; US 20190200083 A1 filed on Dec 21, 2017 (hereinafter “Pio”)
Regarding claim 16, Grancharov does not teach The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed on a server or in a data center
However, Lo does teach The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed on a server or in a data center (Lo Paragraph 0220; “For example, the disclosed methods can be executed on processing units 2110 located in the computing environment 2130, or the disclosed methods can be executed on servers located in the computing cloud 2190.” Examiner notes that the steps are performed on a server (servers located in the computing cloud))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, and Lo. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. One of ordinary skill would have motivation to combine Grancharov, Guo, Freris, and Lo to allow for faster hardware, reduce memory overhead, and reduce energy use “Use of quantized formats can improve neural network processing by, for example, allowing for faster hardware, reduced memory overhead, simpler hardware design, reduced energy use, reduced integrated circuit area, cost savings and other technological improvements.” (Lo Paragraph 0051).
Grancharov in view of Lo does not teach and the quantized vectors or the output values are streamed to a user device.
However, Pio does teach and the quantized vectors or the output values are streamed to a user device. (Pio Paragraph 0005; ” the quantization is performed in real-time as the frames are being streamed to the computing device.” Pio Paragraph 0054; “a user operating a computing device can access a single-stream content item to be presented through a viewport. The viewport may be provided through a display of the computing device (e.g., a virtual reality computing device). In this example, the single-stream content item may represent a view of a scene in which all directions have been encoded at a high quality… The frame compression module 214 can then obtain quantized residual data by decompressing the content item. This quantized residual data may be obtained on a per-frame basis, for example.” Examiner notes that the quantized vectors (quantized residual data in content item) are streamed to a user device (streamed to the computing device))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, Lo, and Pio. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. Pio teaches methods for presenting content. One of ordinary skill would have motivation to combine Grancharov, Guo, Freris, Lo, and Pio to improve streaming of the content “In some embodiments, predicted changes to the user's head orientation can be used to improve streaming of the content item” (Pio Paragraph 0052).
Claim(s) 19 is rejected under 35 U.S.C. 103 as being unpatentable over Volodya Grancharov et al; US 20150051907 A1 filed Dec 12, 2012 (hereinafter “Grancharov”) in view of Ruiqi Guo et al; US 20210064634 A1 filed on Aug 25, 2020 (hereinafter “Guo”) in further view of Nikolaos Freris et al; US 20130031063 A1 (hereinafter “Freris”) in further view of Daniel Lo et al; US 20200264876 A1 filed on Feb 14, 2019 (hereinafter “Lo”) in further view of Steven L. Tieg et al; US 11537870 B1 filed on Mar 14, 2018 (hereinafter “Pio”)
Regarding claim 19, Grancharov does not teach The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed on a virtual machine comprising a portion of a graphics processing unit.
However, Tieg does teach The computer-implemented method of claim 1, wherein at least one of the steps of initializing, mapping, optimizing, quantizing, and processing is performed on a virtual machine comprising a portion of a graphics processing unit. (Teig Column 12 Line 59; “As shown, the system 400 includes an input generator 405, an error calculator 410, an error propagator 415, and a weight modifier 425. In some embodiments, all of these modules execute on a single device, such as a server, desktop or laptop computer, a mobile device (e.g., a smartphone, tablet, etc.), a virtual machine, etc. In other embodiments, these modules may execute across multiple interconnected devices (or virtual machines), or separate instances may execute on multiple devices (or virtual machines) for additional computing power. In some embodiments, at least some of the operations are executed by one or more graphics processing units (GPUs) of such a computing device (or devices).” Examiner notes that steps are performed on a virtual machine comprising a portion of a graphics processing unit)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Grancharov, Guo, Freris, Lo, and Tieg. Grancharov teaches Vector Quantizer and method therein for efficient vector quantization. Guo teaches methods of quantizing a database with respect to a novel loss or quantization error function which applies a weight to an error measurement of quantized elements respectively corresponding to the datapoints in the database. Freris teaches compression of data partitioned into clusters. Lo teaches methods for training a neural network accelerator using quantized precision data formats are disclosed, and, in particular, for adjusting floating-point formats used to store activation values during training. Tieg teaches methods training a machine learning network. One of ordinary skill would have motivation to combine Grancharov, Guo, Freris, Lo, and Tieg to reduce variance of the estimated gradient, leading to smoother, faster, and higher-quality optimization “the propagation of distributions rather than sampling output values reduces the variance of the estimated gradient, which on the whole leads to improved solution quality. In addition, in the case that output values are quantized, propagating distributions allows for the network to compute a continuous function of the network parameters, avoiding discontinuities and infinite gradients that would appear under the assumption that the dot products are known with infinite precision.” (Tieg Column 19 Line 59).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DANIEL DUC TRAN whose telephone number is (571)272-6870. The examiner can normally be reached Mon-Fri 8:00-5:00 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, Viker Lamardo can be reached at (571) 270-5871. 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.
/D.D.T./Examiner, Art Unit 2147
/ERIC NILSSON/Primary Examiner, Art Unit 2151
Prosecution Insights