CHANNEL SPECIFIC NEURAL NETWORK PROCESSING

Notice of Pre-AIA or AIA Status 1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . DETAILED ACTION 2. This Office Action is sent in response to Applicant’s Communication received on 08/12/2026 for application number 18/529,102. Response to Amendments 3. The Amendment filed 08/12/2026 has been entered. Claims 1, 3, 9, 11, 13, 19, 21, 27, and 29 have been amended. Claims 1, 3, 6-9, 11, 13, 16-19, 21, 24-27, 29, and 32-34 remain pending in the application. 4. Applicant’s amendments to claims 1, 9, 11, 19, 27 have been fully considered and are persuasive. The amendments provided to overcome the §101 rejection set forth in the previous office action are sufficient. Accordingly, the 35 U.S.C § 101 rejection of claims 1-34 is respectfully withdrawn. Response to Arguments Applicant argues that the Office, however, failed to perform any obviousness analysis regarding why a person of ordinary skill in the art would modify the embodiment described in Lee's algorithm 1000 with another distinct embodiment described in Lee's algorithm 1100. Without this analysis, the Office has failed to establish a prima facie case of obviousness. However, Examiner respectfully disagrees and noted that the rejection does not require combining Lee’s algorithm 1000 with algorithm 1100. Lee’s discussion of algorithm 1000, together with paragraphs [0135]-[0138], already teaches the relevant arrangement. Specifically, Lee teaches different scale factors for different input channels and an output channel scale factor used in quantization. Applicant argues that Lee does not teach the amended feature in the amended independent claim 1. However, the argument is moot since this is a newly presented limitation, thus changes the scope of the claim. However, a newly found reference, Nagel, is applied. Claim Rejections – 35 USC § 103 5. The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action: A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102 of this title, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made. 6. Claims 1, 3, 6-9, 11, 13, 16-19, 21, 24-27, 29, and 32-34 are rejected under 35 U.S.C. 103 as being unpatentable over Lee et al. (U.S. Patent Application Pub. No. US 20210303972 A1) in view of Nagel et al. (U.S. Patent Application Pub. No. US 20200302299 A1). Claim 1: Lee teaches a method for channel specific neural network processing (i.e. The method may generate the second neural network by performing quantization for each channel of the determined layers of the first neural network; para. [0016]) executed by a processor (i.e. The processor 1210 controls an overall operation of the electronic system 1200. The processor 1210 may include a single processor core (Single Core), or a plurality of processor cores (Multi-Core); para. [0157]), comprising: (a) receiving current layer multi-channel output descriptors by next layer neurons (i.e. The second feature map FM2 of the convolution layer may be an input feature map of the next layer … Each of the inputs and outputs of each of the channels may be referred to as an input activation and an output activation, respectively. That is, the activation may be an output of one channel and may be a parameter corresponding to an input of channels included in the next layer; para. [0054, 0056, 0062]), wherein the current layer multi-channel output descriptors are provided by neurons of the current layer and the current layer and the next layer belong to a neural network (i.e. Each of the layers included in the neural network 2 may include a plurality of channels. A channel may correspond to a plurality of artificial nodes, processing elements (PEs), units, or similar terms. For example, as illustrated in FIG. 2, Layer 1 may include two channels (nodes), and each of the Layer 2 and Layer 3 may include three channels … the activation may be an output of one channel and may be a parameter corresponding to an input of channels included in the next layer; para. [0059-0062]); and each current layer multi-channel output descriptor includes multiple segments, each segment associated with a unique channel (i.e. fig. 1, the first and second feature maps FM1 and FM2 may be a high-dimensional matrix of two or more dimensions, and have respective activation parameters. When the first and second feature maps FM1 and FM2 correspond to, for example, three-dimensional feature maps, the first and second feature maps FM1, and FM2 have a width W (or column), a height H (or row), and a depth D. At this point, the depth D may correspond to the number of channels; para. [0054, 0055]); (b) processing, by the next layer neurons, the current layer multi-channel output descriptors to provide next layer multi-channel output descriptors (i.e. Each of the channels may determine its activation based on activations and weights received from channels included in the previous layer. The weight is a parameter used to operate an output activation in each channel, and may be a value assigned to a connection relationship between channels; para. [0063, 0064, 0069]), wherein the processing comprises: multiplying the current layer multi-channel output descriptors by channel compensated weights of the next layer neurons (i.e. weights included in the weight map WM may respectively be multiplied and added to all feature values in a region overlapping with the first feature map FM1 … when 6 is an activation function, wjk i is a weight from a kth channel included in an (i−1)th layer to a jth channel included in an layer, bj i is a bias of the jth channel included in the layer, and aj i is an activation of the jth channel in the ith layer, the activation may be calculated by using Equation 1 … a scale factor of an input channel is pre-reflected on a weight by multiplying the weight by the reciprocal of the scale factor of the input channel (that is, by dividing the weight by the scale factor of the input channel); para. [0055, 0064, 0065, 0139]) to provide next layer products that compensate for estimated differences between scale factors associated with different channels of the current layer multi-channel output descriptors (i.e. At this point, one of SF I 1×SF W 1, SF I 2×SF W 1 and SF I 3×SF W 1 is selected as the scale factor SF O 1, information loss may occur in the process of quantizing with the unified selected scale factor. Also, the generated information loss may increase as the difference between the scale factors of each channel of the input feature map increases; para. [0134-0136]), different channels of an input feature map may have different scale factors, wherein: the channel compensated weights virtually align the scale factors associated with the different channels relative to one another (i.e. a scale factor of an input channel is pre-reflected on a weight by multiplying the weight by the reciprocal of the scale factor of the input channel (that is, by dividing the weight by the scale factor of the input channel) … the algorithm 1000 according to the related art may reduce the number of scale factors to be considered in the process of calculating the partial sum for each channel of an output feature map to a scale factor of a kernel (weight map) by pre-reflecting the scale factor of an input channel to the weight; para. [0139, 0143]); and multiplying a segment associated with a unique channel by a channel compensated weight that was compensated by a scale factor associated with the unique channel (i.e. Each of the channels may be processed by a computational unit or a processing element that outputs an output activation by receiving an input, and an input-output of each of the channels may be mapped. For example, when 6 is an activation function, wjk i is a weight from a kth channel included in an (i−1)th layer to a jth channel included in an layer, bj i is a bias of the jth channel included in the layer, and aj i is an activation of the jth channel in the ith layer, the activation may be calculated by using Equation 1 below. aji=σ(∑k⁢(wjki×aki-1)+bji):Equation⁢⁢1; para. [0064, 0065]) is executed by a multiplication element of the processor (i.e. Examples of hardware components that may be used to perform the operations described in this application where appropriate include controllers, sensors, generators, drivers, memories, comparators, arithmetic logic units, adders, subtractors, multipliers, dividers, integrators, and any other electronic components configured to perform the operations described in this application; para. [0167]), such that the processor applies a different quantization scale per output channel (i.e. If a quantization for applying a different scale factor for each channel is performed, a scale factor of a weight map used to obtain the first channel of an output feature map is SF W 1, the scale factor SF O 1 of the first channel of the output feature map needs to be selected as one of SF I 1×SF W 1, SF I 2×SF W 1 and SF I 3×SF W 1 … an input channel and output channel of each layer of the quantized neural network also need to be quantized; para. [0134, 0135, 0138]); and quantizing the next layer products by applying next layer output channel specific quantization (i.e. assuming that scale factors of three channels of an input feature map respectively are SF I 1, SF I 2 and SF I 3, and a scale factor of a weight map used to obtain the first channel of an output feature map is SF W 1, the scale factor SF O 1 of the first channel of the output feature map needs to be selected as one of SF I 1×SF W 1, SF I 2×SF W 1 and SF I 3×SF W 1; para. [0135-0138]). Lee does not explicitly teach without allocating dedicated hardware to align content of the different channels before the multiplying. However, Nagel teaches processing, by the next layer neurons, the current layer multi-channel output descriptors to provide next layer multi-channel output descriptors (i.e. The intermediate layer(s) inputs may multiply the incoming activation with a weight matrix 210 or may apply one or more weight factors and/or a bias to the black box inputs … Intermediate signals may be passed to other nodes or layers within the intermediate layer(s) 202 to produce the intermediate layer(s) activations that are ultimately passed as inputs to the output layer 204; para. [0039, 0040]), wherein the processing comprises: multiplying the current layer multi-channel output descriptors by channel compensated weights of the next layer neurons (i.e. scaling each of the output channels weights by a corresponding scaling factor, and scaling the next layer's corresponding input channel weights by the inverse of the corresponding scaling factor … Folding the scaling parameters into the weight matrices and the biases, i.e. {tilde over (W)}1=S−1W1, {tilde over (W)}2=W2S and {tilde over (b)}1=S−1b1,; para. [0033, 0049, 0055, 0056]) to provide next layer products that compensate for estimated differences between scale factors associated with different channels of the current layer multi- channel output descriptors (i.e. when the weight ranges of different layers or channels differ significantly, it is challenging to accomplish per-tensor quantization of neural network weight tensors … A number of different factors or techniques may be used to determine the corresponding scaling factor, including heuristics, equalization of dynamic ranges, equalization of range extrema (minima or maxima), differential learning using straight through estimator (STE) methods and a local or global loss, and/or by using a metric for the quantization error … the computing device may be configured to select S in a way that it makes the dynamic ranges of each channel as similar as possible; para. [0028, 0033, 0057]), wherein: the channel compensated weights virtually align the scale factors associated with the different channels relative to one another (i.e. To reduce outliers and nearly equalize the ranges of weight tensors or channel weights within a layer, the computing system 300 may move scaling of the first layer to the adjacent second layer … the computing device may be configured to select S in a way that it makes the dynamic ranges of each channel as similar as possible. This heuristic ensures that the maximum/minimum per channel, which also defines the quantization range, are equally spread across the layers that share the same channel; para. [0033, 0044, 0057]); and multiplying a segment associated with a unique channel by a channel compensated weight that was compensated by a scale factor associated with the unique channel is executed by a multiplication element of the processor (i.e. If scaling factor si scales ci in the first layer 302, the computing device may instead factor it out and multiply the input channel di in second layer 304, moving the scaling factors for each layer to equalize the ranges of the channels of both W1 and W2 … An inverse scaling may be applied to the second layer 304 corresponding input channel di weights … the processor may scale each of the adjacent layer's corresponding input channel weights (i.e., the input channel weights of the i+1 layer) by applying an inverse of the corresponding scaling factor to each of input channel weights; para. [0045, 0049, 0065]), such that the processor applies a different quantization scale per output channel (i.e. This heuristic ensures that the maximum/minimum per channel, which also defines the quantization range, are equally spread across the layers that share the same channel; para. [0033, 0052, 0057]) without allocating dedicated hardware multipliers to align content of the different channels before the multiplying (i.e. Folding the batch normalization parameters into a preceding layer improves inference time because per channel scale and shift operations do not have to be performed—those operations are subsumed into the weight and bias parameters; para. [0050, 0056, 0072]); and quantizing the next layer products by applying next layer output channel specific quantization (i.e. the result of the inferences reflect the 32-bit fidelity in the model, so the results undergo re-quantization 108 to provide 8-bit unsigned value (uint8) output values; para. [0025, 0052, 0057]). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the invention of Lee to include the feature of Nagel. One would have been motivated to make this modification because it improves quantization performance, reduces quantization error caused by differing channel ranges, and improves inference efficiency. Claim 3: Lee and Nagel teach the method according to claim 1. Lee further teaches wherein each current layer multi-channel output descriptor segment is associated with a unique channel out of multiple current layer multi-channel output descriptor channels (i.e. The first and second feature maps FM1 and FM2 may be a high-dimensional matrix of two or more dimensions, and have respective activation parameters. When the first and second feature maps FM1 and FM2 correspond to, for example, three-dimensional feature maps, the first and second feature maps FM1, and FM2 have a width W (or column), a height H (or row), and a depth D. At this point, the depth D may correspond to the number of channels; para. [0054, 0055]). Claim 6: Lee and Nagel teach the method according to claim 1. Lee further teaches wherein the estimated differences are learned by monitoring reference current layer neurons of a reference current layer of a reference neural network that is approximated by the neural network (i.e. the first neural network may correspond to a neural network quantized from a third neural network having layers of floating-point parameters of a third bit precision greater than the first bit precision and having layers of fixed-point parameters of a first bit precision. In other words, the first neural network may correspond to a pre-quantized neural network that is generated as quantization of the third neural network is performed; para. [0114-0116]). Claim 7: Lee and Nagel teach the method according to claim 6. Lee further teaches comprising: determining the channel compensated weights by: training the neural network to provide trained weights (i.e. The pre-trained neural network may refer to a neural network generated by repeatedly training (or learning) a given initial neural network; para. [0084, 0087, 0092]); and modifying the trained weights by correction factors that represent the estimated differences (i.e. If a quantization for applying a different scale factor for each channel is performed, after a convolution operation between the input feature map and the weight map, the number of scale factors to be considered in the process of calculating a partial sum for each channel of an output feature map may be increased by the number of channels in the input feature map; para. [0095-0097, 0135, 0136, 0139]). Claim 8: Lee and Nagel teach the method according to claim 1. Lee further teaches wherein the neural network comprises multiple layers and the method comprises performing step (a) and step (b) for at least two pairs of consecutive layers of the multiple layers (i.e. the neural network 2 may be a DNN or an n-layer neural network, including two or more hidden layers. For example, as illustrated in FIG. 2, the neural network 2 may be a DNN, including an input layer (Layer 1), two hidden layers (Layer 2 and Layer 3), and an output layer (Layer 4); para. [0059]). Claim 9 is similar in scope to Claim 1 and is rejected under a similar rationale. Lee teaches a non-transitory computer readable medium (i.e. methods as described above, and any associated data, data files, and data structures, may be recorded, stored, or fixed in or on one or more non-transitory computer-readable storage media; para. [0170]) for channel specific neural network processing, the non-transitory computer readable medium storing instructions that, once executed by one or more processing circuits, causes the one or more processing circuits (i.e. The processor 110 performs an overall function for controlling the neural network quantization apparatus 10. For example, the processor 110 controls an overall operation of the neural network quantization apparatus 10 by executing programs stored in the memory 120 in the neural network quantization apparatus 10; para. [0080]). Claim 11 is similar in scope to Claim 1 and is rejected under a similar rationale. Lee teaches a neural network processor, comprising: one or more processing circuits that are configured to (i.e. The processor 110 performs an overall function for controlling the neural network quantization apparatus 10. For example, the processor 110 controls an overall operation of the neural network quantization apparatus 10 by executing programs stored in the memory 120 in the neural network quantization apparatus 10; para. [0080]). Claims 13, 16-18 are similar in scope to Claims 3, 6-8 and are rejected under a similar rationale. Claim 19 is similar in scope to Claim 1 and is rejected under a similar rationale. Lee teaches a method for channel specific neural network processing in a neural network including current layer neurons and next layer neurons (i.e. the neural network 2 may be a DNN or an n-layer neural network, including two or more hidden layers. For example, as illustrated in FIG. 2, the neural network 2 may be a DNN, including an input layer (Layer 1), two hidden layers (Layer 2 and Layer 3), and an output layer (Layer 4); para. [0059]). Claims 21 and 24-26 are similar in scope to Claims 3, 6-8 and are rejected under a similar rationale. Claim 27 is similar in scope to Claim 1 and is rejected under a similar rationale. Lee teaches a neural network processor for channel specific neural network processing in a neural network including current layer neurons and next layer neurons (i.e. The processor 110 performs an overall function for controlling the neural network quantization apparatus 10. For example, the processor 110 controls an overall operation of the neural network quantization apparatus 10 by executing programs stored in the memory 120 in the neural network quantization apparatus 10; para. [0059, 0080]). Claims 29 and 32-34 are similar in scope to Claims 3, 6-8 and are rejected under a similar rationale. Conclusion The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure. Kim et al. (Pub. No. US 10325352 B1), The method includes steps of: a computing device (a) generating k-th quantization loss values by referring to k-th initial weights of a k-th initial convolutional layer included in a k-th convolutional block, a (k−1)-th feature map outputted from the (k−1)-th convolutional block, and each of k-th scaling parameters; (b) determining each of k-th optimized scaling parameters by referring to the k-th quantization loss values; (c) generating a k-th scaling layer and a k-th inverse scaling layer by referring to the k-th optimized scaling parameters; and (d) transforming the k-th initial convolutional layer into a k-th integrated convolutional layer by using the k-th scaling layer and the (k−1)-th inverse scaling layer. Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a). A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action. It is noted that any citation to specific pages, columns, lines, or figures in the prior art references and any interpretation of the references should not be considered to be limiting in any way. A reference is relevant for all it contains and may be relied upon for all that it would have reasonably suggested to one having ordinary skill in the art. In re Heck, 699 F.2d 1331, 1332-33, 216 U.S.P.Q. 1038, 1039 (Fed. Cir. 1983) (quoting In re Lemelson, 397 F.2d 1006, 1009, 158 U.S.P.Q. 275, 277 (C.C.P.A. 1968)). Any inquiry concerning this communication or earlier communications from the examiner should be directed to TAN TRAN whose telephone number is (303)297-4266. The examiner can normally be reached on Monday - Thursday - 8:00 am - 5:00 pm MT. 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, Matt Ell can be reached on 571-270-3264. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000. /TAN H TRAN/Primary Examiner, Art Unit 2141