Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 27 January 2026 has been entered.
Claim Rejections - 35 USC § 101
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claims 1-31 is/are rejected under 35 U.S.C. 101 because they are directed to abstract ideas without significantly more.
Regarding claims 1-31:
Step 1: With respect to claims 1-8 and 17-24, applying step 1, the preamble of claims 1-8 and 17-24 claims a processor, which falls within the statutory category of an apparatus. With respect to claims 9-16, applying step 1, the preamble of claims 9-16 claims a machine-readable medium, which falls within the statutory category of a manufacture. With respect to claims 25- 31, applying step 1, the preamble of claims 25-31 claims a method, which falls within the statutory category of a process.
Regarding claim 1,
Step 2A – Prong One: Claim 1 recites:
One or more processors comprising:
circuity to use one or more neural networks to perform one or more operations including:
at least one operation to generate one or more feature maps for an image using two or more filters, wherein at least two filters of the two or more filters are to be adaptively adjusted to have different characteristics based, at least in part, on different features at different locations of the image; and
at least one operation to identify one or more features in the image based, at least in part, on the one or more feature maps
The broadest reasonable interpretation of the bolded limitations above are directed to mathematical concepts and mental processes. Generating one or more feature maps for an image using two or more filters is a mathematical calculation (see specification at [0057], states using filters consists of performing dot-product operations across all dimensions of an input image). Adaptively adjusting filters based on different features at different location is a mathematical calculation (see specification at [0057], filters are used for dot-product operations, adjustment of a filter is adjustment of a dot-product operand or matrix, which is a mathematical calculation). Identifying one or more features in the image based at least in part on the one or more feature maps is a process that can be performed in the human mind, with or without the physical aids of a pen and paper. A human could observe a map of features, and use evaluation and judgement to identify features in an image based on a feature map.
Step 2A – Prong One (Yes).
Step 2A – Prong Two: The additional elements in this claim are “one or more processors” and “circuitry to use one or more neural networks to perform one or more operations including”. These additional elements are mere instructions to apply the judicial exception on a generic computer (See MPEP 2106.05(f)). The computer is recited at a high level of generality and imposes no meaningful limitations on the claim.
Even when viewed in combination, the additional element does not integrate the recited judicial exception into practical application. Step 2A – Prong Two (No).
Step 2B: The additional elements in this claim are “one or more processors” and “circuitry to use one or more neural networks to perform one or more operations including”. These additional elements are mere instructions to apply the judicial exception on a generic computer (See MPEP 2106.05(f)). The computer is recited at a high level of generality and imposes no meaningful limitations on the claim. Thus, the additional elements do not amount to significantly more than the judicial exception. Step 2B (No).
Claim 1 is ineligible.
With respect to claims 9, 17, and 25,
These claims are similar to claim 1 and thus are rejected on similar rationale as above. The non-transitory machine readable medium recited in these claims is also a generic computing component.
Claims 9, 17, and 25 are ineligible.
Dependent claims:
Claims 2-4, 6, 8, 10-13, 15-16, 18-20, 22, 24, 27-28, 30-31: These claims recite further abstract ideas (mathematical concepts or mental processes) and thus are ineligible.
Claims 5, 14, 21, and 29: This claim recites “wherein the one or more neural networks comprise convolutional neural networks”. In Step 2A – Prong Two, this is a mere attempt to generally link the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)), and thus does not integrate the judicial exception into practical application. In Step 2B, this does not amount to significantly more than the judicial exception. Thus, claims 5, 14, 21, and 29 are ineligible.
Claim 7, 23, and 26: These claims recite further abstract ideas (mathematical concepts) but also recite “one or more neural networks”. In Step 2A – Prong Two, this is a mere attempt to generally link the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)), and thus does not integrate the judicial exception into practical application. In Step 2B, this does not amount to significantly more than the judicial exception. Thus, claims 7 and 26 are ineligible.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1-7, 9-15, 17-23, and 25-30 is/are rejected under 35 U.S.C. 103 as being unpatentable over Wolf et al. (US Pub. No. 2019/0042892, filed April 2017, hereinafter “Wolf”) in view of Matveev et al. (US Patent No. 10,902,318, filed Nov. 2018, hereinafter “Matveev”).
Regarding claim 1, Wolf teaches one or more processors, comprising:
to use one or more neural networks to perform one or more operations including:
at least one operation to generate one or more feature maps for an image using two or more filters (Wolf, [0006] – “ In first layer 51.sub.1, image 52 is convolved with each one of filters 54.sub.1 and 54.sub.2…. Each one of filters 54.sub.1 and 54.sub.2 corresponds to a feature to be identified in the image. The sizes of the filters as well as the stride are design parameters selected by the CNN designer. Convolving image 52 with each one of filters 54.sub.1 and 54.sub.2 produces a feature map which includes two feature images or matrices, feature image 56.sub.1 and feature image 56.sub.2 respective of filters 54.sub.1 and 54.sub.2 (i.e., a respective image is produced for each filter).” – teaches generating one or more feature maps for an image using two or more filters)
at least one operation to identify one or more features in [[an]] the image based, at least in part, on the one or more feature maps (Wolf, [0019] – “Each object detector at least includes a respective CNN which includes a plurality of convolution layers. Each convolution layer convolves the input thereto with a plurality of filters and the results of this convolution are processed by an activation function. Each object detector further includes an object classifier, coupled with the convolutional neural network, for classifying objects in the image according to the results from the convolutional neural network.” – teaches using one or more neural networks to identify one or more feature in an image based on a feature map (results from convolutional neural network, determines probability than an object is located at each of the windows associated with the feature maps provided), Wolf teaches the features based on a feature map in [0034] – “This classification vector includes values relating to the probability that an object or objects (i.e., which the CNN was trained to detect) is located at each of the image windows associated the features map provided thereto. Furthermore, the classification vector determined by each one of classifiers 205.sub.1, 205.sub.2, 205.sub.3, 207.sub.1, 207.sub.2 and 207.sub.3 include values relating to image window correction factors for each of the image windows associated the features map provided thereto. These image window correction factors include for example corrections to the width and the height of the image window. These image window correction factors may further include corrections to the location of the image window as well as the orientation of the image window.”).
Wolf fails to explicitly teach circuitry and wherein at least two filters of the two or more filters are to be adaptively adjusted to have different characteristics based, at least in part, on different features at different location of the image.
However, analogous to the field of the claimed invention, Matveev teaches:
circuitry to use one or more neural networks to perform one or more operations including (Matveev, Pg. 17, Col. 6, Lines 37-50 – “In other instances, well-known methods, procedures, and components, modules, units and/or circuits have not been described in detail so as not to obscure the invention… Embodiments of the invention can allow for CNNs to be realizably implemented on a CPU. Further, while CPU based machines are discussed, GPUs or other types of processors may be used.”, Pg. 18, Col. 7, Lines 54-62 – “Computing device 100 may include a controller or processor 105 that may be or include, for example, one or more central processing unit processor(s) (CPU), one or more Graphics Processing Unit(s) (GPU or GPGPU), a chip or any suitable computing or computational device, an operating system 115, a memory 120, a storage 130, input devices 135 and output devices 140.”, and Fig. 2):
at least one operation to generate one or more feature maps for an image using two or more filters, wherein at least two filters of the two or more filters are to be adaptively adjusted to have different characteristics based, at least in part, on different features at different locations of the image (Matveev, Pg. 18, Col. 7, Lines 15-19 – “As is also known in the art, the convolutional layer can convolve the input with one or more filters. Each filter can be represented as a vector, matrix and/or array. For example, a convolutional filter array can be defined by row (R), column (C) and channel (CH).”, Pg. 18, Col. 7, Lines 34-37 – “A convolutional layer's parameters may include a set of learnable filters (e.g., kernels, convolutional filter arrays), which have a small receptive field, but extend through the full depth of the input volume.” and in Pg. 18, Col. 7 Lines 38-41 – “During the forward pass, each filter may be convolved across the width and height of the input volume. As a result, the NN may learn filters that activate when they detect some specific type of feature at some spatial position in the input.” – teaches at least one operation to generate one or more feature maps for an image using two or more filters (convolve input with multiple filters), wherein at least two filters of the two or more filters are to be adaptively adjusted (learnable filters, such as kernels, that are updated - thus filters are adaptively adjusted) to have different characteristics based at least in part on different features at different locations of the image (learns filters that activate based at least in part on some specific feature at some spatial position. Thus, teaches multiple learnable filters that activate upon detection of different features at different locations, wherein each filter is adaptively adjusted to have different characteristics, such as learnable filter arrays of size R and C or learnable kernels at different locations, based at least in part on the detected feature at some spatial position));
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the circuits and adaptively adjusted filters of Matveev to the neural networks, two or more filters, feature maps, and feature identification of Wolf in order to use one or more circuits to use one or more neural networks to identify features in an image using at least two filters. Doing so would provide mechanisms for self-correcting of a convolutional neural network (Matveev, Pg. 15, Col. 1) and improve computation efficiency of convolutional neural networks (Matveev, Pg. 16, Col. 3).
Claims 9, 17, and 25 incorporate substantively all the limitations of claim 1 in a non-transitory machine-readable medium, one or more processors, and a method, and are rejected on similar grounds as above.
Regarding claim 2, the combination of Wolf and Matveev teaches the one or more processors of claim 1, to identify the one or more features in the image by at least: calculating a first value of a first filter for at least one location of the different locations; calculating a second value of a second filter for at least one location of the different locations; and calculating a third value for at least one of the feature maps based at least in part on the first value and the second value (Wolf, [0003] – “Reference is now made to FIG. 1, which is a schematic illustration of a CNN, generally referenced 10, which is known in the art. CNN 10 is employed for detecting features in an image such as image 16. Neural network 10 includes a plurality of layers, such as layer 12.sub.1 (FIG. 1). CNN 10 includes a plurality of layers 12.sub.1, 12.sub.2, . . . , 12.sub.N and a classifier 14. An input image 16 is supplied to layer 12.sub.1. Layer 12.sub.1 at least convolves image 16 with the respective filters thereof and multiplies each of the outputs of the filters by an activation function. Layer 12.sub.1 provides the output thereof to layer 12.sub.2 which performs the respective operations thereof with the respective filters. This process repeats until the output of layer 12.sub.N is provided to classifier 14. The output of Layer 12.sub.N is a map of features corresponding to the filters employed in CNN 10.” – teaches calculating a first value of a first filter for the location (layer 12.sub.1 convolves image and multiplies outputs of filters by an activation function), a second value of a second filter for the location (layer 12.sub.1 performs respective operations thereof with respective filters), and a third value for a feature map based at least in part on the first and second values (layer 12.sub.N is a map of features corresponding to the filters employed). In addition to the previously cited passages, Wolf further teaches in [0006] – “Accordingly, each one of filters 54.sub.1 and 54.sub.2 is shifted over selected positions in the image. At each selected position, the pixel values overlapping with filter are multiplied by the respective weights of the filter and the result of this multiplication is summed (i.e., a multiply and sum operation). Generally, the selected positions are defined by shifting the filter over the image by a predetermined step size referred to as ‘stride’…Convolving image 52 with each one of filters 54.sub.1 and 54.sub.2 produces a feature map which includes two feature images or matrices, feature image 56.sub.1 and feature image 56.sub.2 respective of filters 54.sub.1 and 54.sub.2 (i.e., a respective image is produced for each filter). Each pixel or entry in the feature image corresponds to the result of one multiplication and sum operation.” – teaches calculating the first and second values of a first and second filter for at least one location of the different locations (pixel values at selected locations are multiplied and summed with the filter, selected position shifts based on the stride, thus the values of the filters are calculated at a location of the different locations)).
Wolf fails to explicitly teach wherein the circuitry is to perform the at least one operation.
However, analogous to the field of the claimed invention, Matveev teaches:
wherein the circuitry is to perform the at least one operation (Matveev, Pg. 17, Col. 6, Lines 37-50 – “In other instances, well-known methods, procedures, and components, modules, units and/or circuits have not been described in detail so as not to obscure the invention… Embodiments of the invention can allow for CNNs to be realizably implemented on a CPU. Further, while CPU based machines are discussed, GPUs or other types of processors may be used.”, Pg. 18, Col. 7, Lines 54-62 – “Computing device 100 may include a controller or processor 105 that may be or include, for example, one or more central processing unit processor(s) (CPU), one or more Graphics Processing Unit(s) (GPU or GPGPU), a chip or any suitable computing or computational device, an operating system 115, a memory 120, a storage 130, input devices 135 and output devices 140.”, and Fig. 2)
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the circuits of Matveev to the feature identification of Wolf in order to use circuits to identify one or more features in an image. Doing so would enable outputting probabilities or weights that a feature exists within an image based on the filters (Wolf, [0035]) and allow for CNNs to be implemented on computing components and devices (Matveev, Pg. 17, Col. 6).
Claims 10 and 18 are similar to claim 2, hence similarly rejected.
Regarding claim 3, the combination of Wolf and Matveev teaches the one or more processors of claim 2, wherein the third value is calculated at least in part by aggregating the first value and the second value (Wolf, [0006] – “Convolving image 52 with each one of filters 54.sub.1 and 54.sub.2 produces a feature map which includes two feature images or matrices, feature image 56.sub.1 and feature image 56.sub.2 respective of filters 54.sub.1 and 54.sub.2 (i.e., a respective image is produced for each filter). Each pixel or entry in the feature image corresponds to the result of one multiplication and sum operation. Thus, each one of matrices 56.sub.1 and 56.sub.2 is associated with a respective image feature corresponding to the respective one of filters 54.sub.1 and 54.sub.2.” – teaches wherein the third value is calculated at least in part by aggregative the first and second values (each pixel or entry in the feature image corresponds to the result of one multiplication and sum (aggregation) operation, thus each matrix is associated with a respective image feature)).
Claim 11 is similar to claim 3, hence similarly rejected.
Regarding claim 4, the combination of Wolf and Matveev teaches the one or more processors of claim 2, wherein the first value and second value indicate a probability that the first filter or the second filter detected the one or more features (Wolf, [0003] – “Layer 12.sub.1 provides the output thereof to layer 12.sub.2 which performs the respective operations thereof with the respective filters. This process repeats until the output of layer 12.sub.N is provided to classifier 14. The output of Layer 12.sub.N is a map of features corresponding to the filters employed in CNN 10. This feature map relates to the probability that a feature is present in input image 16 within respective image windows associated with the feature map. The features map at the output of layer 12.sub.N can be embodied as a plurality of matrices, each corresponding to a feature, where the value of entry in each matrix represents the probability that input image 16 includes the feature associated with that matrix, in a specific image window (i.e., a bounding box) associated with the entry location in the matrix (i.e., the indices of the entry)” – teaches wherein the first and second values (values of layers 12.sub.1 - 12.sub.N) indicate a probability that the first filter or second filter detected one or more features for the location (probability that input image 16 includes the feature associated with that matrix (filter))).
Claims 20 are similar to claim 4, hence similarly rejected.
Regarding claim 5, the combination of Wolf and Matveev teaches the one or more processors of claim 1, wherein the one or more neural networks comprise convolutional neural networks (Wolf, [0019] – “Each object detector at least includes a respective CNN which includes a plurality of convolution layers. Each convolution layer convolves the input thereto with a plurality of filters and the results of this convolution are processed by an activation function. Each object detector further includes an object classifier, coupled with the convolutional neural network, for classifying objects in the image according to the results from the convolutional neural network.” – teaches wherein the one or more neural networks comprise convolutional neural networks).
Claims 21 and 29 are similar to claim 5, hence similarly rejected.
Regarding claim 6, the combination of Wolf and Matveev teaches the one or more processors of claim 1, wherein the at least one operation to generate the one or more feature maps for the image using the two or more filters comprises combining values from the two or more filters differently at the different locations (Wolf, [0026] – “Each layer of the respective CNN in each one of object detectors 108.sub.1, 108.sub.2, . . . , 108.sub.L convolves the image provided thereto with corresponding filters. The output of each CNN is a map of features corresponding to the filters employed by the CNN. The feature map includes entries of values. Each value of each entry in the feature map represents the feature intensity of the features associated various filters, within an image window associated with the entry.” – teaches generating one or more feature maps using two or more filters (corresponding various filters) wherein the feature map comprises combining values from the two or more filters differently at different locations (image window associated with the entry), Wolf teaches combining filters in [0006] – “ In first layer 51.sub.1, image 52 is convolved with each one of filters 54.sub.1 and 54.sub.2. Filters 54.sub.1 and 54.sub.2 are also referred to as convolution kernels or just kernels. Accordingly, each one of filters 54.sub.1 and 54.sub.2 is shifted over selected positions in the image. At each selected position, the pixel values overlapping with filter are multiplied by the respective weights of the filter and the result of this multiplication is summed (i.e., a multiply and sum operation). Generally, the selected positions are defined by shifting the filter over the image by a predetermined step size referred to as ‘stride’. Each one of filters 54.sub.1 and 54.sub.2 corresponds to a feature to be identified in the image.” – teaches generating the one or more feature maps for the image using the two or more filters including combining values from the two or more filters differently at the different locations (at each selected position, both filters are multiplied and the results are summed to produce a feature to be identified in the image, thus generating a feature map using the two or more filters by combining the filters differently at different locations)).
Claims 15 and 22 are similar to claim 6, hence similarly rejected.
Regarding claim 7, the combination of Wolf and Matveev teaches the one or more processors of claim 1, wherein the one or more neural networks identify the one or more features in a convolutional layer of a convolutional neural network of the one or more neural networks (Wolf, [0027] – “According to the disclosed technique, object detectors in a group of object detectors are associated with common convolution layers (i.e., since the input image to these object detectors is the same). As such, these common convolutional layers need to be computed only once for each group of object detectors.” – teaches the one or more neural networks identify features in a convolutional layer of a convolutional neural network of the one or more neural networks (convolutional layers in object detectors comprised of convolutional neural networks)).
Claim 23 is similar to claim 7, hence similarly rejected.
Regarding claim 12, the combination of Wolf and Matveev teaches the non-transitory machine-readable medium of claim 10, wherein the first value and second value indicate a weight that the first filter or the second filter detected the one or more features (Wolf, [0003] – “Layer 12.sub.1 provides the output thereof to layer 12.sub.2 which performs the respective operations thereof with the respective filters. This process repeats until the output of layer 12.sub.N is provided to classifier 14. The output of Layer 12.sub.N is a map of features corresponding to the filters employed in CNN 10. This feature map relates to the probability that a feature is present in input image 16 within respective image windows associated with the feature map. The features map at the output of layer 12.sub.N can be embodied as a plurality of matrices, each corresponding to a feature, where the value of entry in each matrix represents the probability that input image 16 includes the feature associated with that matrix, in a specific image window (i.e., a bounding box) associated with the entry location in the matrix (i.e., the indices of the entry)” – teaches wherein the first and second values (values of layers 12.sub.1 - 12.sub.N) indicate a weight that the first filter or second filter detected one or more features (probability that input image 16 includes the feature associated with that matrix (filter))).
Regarding claim 13, the combination of Wolf and Matveev teaches the non-transitory machine-readable medium of claim 12, wherein the instructions, when performed, cause the one or more processors to calculate the weight using a softmax function (Matveev, Pg. 17, Col. 6, Line 66 – Pg. 18, Col. 7, Line 3 – “ NN 1100 may in one example have layers 1130 (convolution), 1132 (pooling), 1134 (convolution), 1136 (pooling), and one or more output layers 1138, which may include for example an FC layer 1138A and a softmax layer 1138B.” – teaches calculating the weight (output) using a softmax function (softmax layer)).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the use of the softmax function of Matveev to the weight calculation of Wolf in order to calculate the weights using a softmax function. Doing so would provide mechanisms for self-correcting of a convolutional neural network (Matveev, Pg. 15, Col. 1)
Regarding claim 14, the combination of Wolf and Matveev teaches the non-transitory machine-readable medium of claim 9, wherein the one or more neural networks are convolutional neural networks (Wolf, [0019] – “Each object detector at least includes a respective CNN which includes a plurality of convolution layers. Each convolution layer convolves the input thereto with a plurality of filters and the results of this convolution are processed by an activation function. Each object detector further includes an object classifier, coupled with the convolutional neural network, for classifying objects in the image according to the results from the convolutional neural network.” – teaches wherein the one or more neural networks are convolutional neural networks).
Regarding claim 19, the combination of Wolf and Matveev teaches the one or more processors of claim 18, wherein the third value is calculated at least in part by applying a function to the first value and the second value and aggregating results of applying the function to the first value and the second value (Wolf, [0003] – “Layer 12.sub.1 at least convolves image 16 with the respective filters thereof and multiplies each of the outputs of the filters by an activation function. Layer 12.sub.1 provides the output thereof to layer 12.sub.2 which performs the respective operations thereof with the respective filters. This process repeats until the output of layer 12.sub.N is provided to classifier 14. The output of Layer 12.sub.N is a map of features corresponding to the filters employed in CNN 10.” – teaches wherein the third value is calculated by applying a function (activation function) to the first and second value (first and second layer’s convolved image with respective filter output prior to being multiplied by activation function) and aggregating the results of applying the function to the first and second value (layer output is fed into proceeding layer until output as a feature map, aggregating results of preceding layers)).
Wolf fails to explicitly teach applying a softmax function.
However, analogous to the field of convolutional neural networks and identifying features, Matveev teaches:
applying a softmax function (Matveev, Pg. 17, Col. 6, Line 66 – Pg. 18, Col. 7, Line 3 – “NN 1100 may in one example have layers 1130 (convolution), 1132 (pooling), 1134 (convolution), 1136 (pooling), and one or more output layers 1138, which may include for example an FC layer 1138A and a softmax layer 1138B.” – teaches calculating the weight (output) using a softmax function (softmax layer))
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the softmax function of Matveev to the third value calculation of Wolf in order to calculate the third value using a softmax function. Doing so would provide mechanisms for self-correcting of a convolutional neural network (Matveev, Pg. 15, Col. 1)
Regarding claim 26, the combination of Wolf and Matveev teaches the method of claim 25, wherein the one or more neural networks include at least one convolutional layer (Wolf, [0003] – “Neural network 10 includes a plurality of layers, such as layer 12.sub.1 (FIG. 1). CNN 10 includes a plurality of layers 12.sub.1, 12.sub.2, . . . , 12.sub.N and a classifier 14.” – teaches one or more neural networks including at least one convolutional layer), the convolutional layer:
applying two or more filters to the image (Wolf, [0003] – “An input image 16 is supplied to layer 12.sub.1. Layer 12.sub.1 at least convolves image 16 with the respective filters thereof and multiplies each of the outputs of the filters by an activation function.” – teaches applying two or more filters to the image. In addition to the previously cited passages, Wolf further teaches in [0006] – “In first layer 51.sub.1, image 52 is convolved with each one of filters 54.sub.1 and 54.sub.2. Filters 54.sub.1 and 54.sub.2 are also referred to as convolution kernels or just kernels.” – applying two or more filters to the image);
calculating weights associated with two or more outputs of the two or more filters (Wolf, [0003] – “An input image 16 is supplied to layer 12.sub.1. Layer 12.sub.1 at least convolves image 16 with the respective filters thereof and multiplies each of the outputs of the filters by an activation function.” – teaches applying an activation function on each of the outputs of the filters to determine a weight. In addition to the previously cited passages, Wolf further teaches in [0006] – “In first layer 51.sub.1, image 52 is convolved with each one of filters 54.sub.1 and 54.sub.2. Filters 54.sub.1 and 54.sub.2 are also referred to as convolution kernels or just kernels… Convolving image 52 with each one of filters 54.sub.1 and 54.sub.2 produces a feature map which includes two feature images or matrices, feature image 56.sub.1 and feature image 56.sub.2 respective of filters 54.sub.1 and 54.sub.2 (i.e., a respective image is produced for each filter)” and in [0007] – “The error (e.g., the squared sum of differences, log loss, softmaxlog loss) between the classification vectors of the CNN and the pre-determined classification vectors is determined. This error is than employed to update the weights and parameters of the CNN in a backpropagation process which may include one or more iterations.” – calculating weights associated with two or more outputs of the two or more filters (the two or more filters produce two or more outputs which are used to update the weights of the two or more filters)); and
aggregating the weights into the one or more feature maps for the image (Wolf, [0003] – “This process repeats until the output of layer 12.sub.N is provided to classifier 14. The output of Layer 12.sub.N is a map of features corresponding to the filters employed in CNN 10.” – teaches aggregating the weights into one or more features maps (the layer 12.sub.N outputs a feature map corresponding to the filters employed in the preceding layers, thus aggregating the weights to produce a feature map)).
Regarding claim 27, the combination of Wolf and Matveev teaches the method of claim 26, wherein the weights are calculated at the different locations in the image (Wolf, [0034] – “This classification vector includes values relating to the probability that an object or objects (i.e., which the CNN was trained to detect) is located at each of the image windows associated the features map provided thereto. Furthermore, the classification vector determined by each one of classifiers 205.sub.1, 205.sub.2, 205.sub.3, 207.sub.1, 207.sub.2 and 207.sub.3 include values relating to image window correction factors for each of the image windows associated the features map provided thereto. These image window correction factors include for example corrections to the width and the height of the image window. These image window correction factors may further include corrections to the location of the image window as well as the orientation of the image window.” – teaches wherein the weights (probabilities) are calculated at different locations (windows) of the image).
Regarding claim 28, the combination of Wolf and Matveev teaches the method of claim 26, wherein each of the two or more filters identify each of the one or more features in the image (Wolf, [0019] – “Each object detector at least includes a respective CNN which includes a plurality of convolution layers. Each convolution layer convolves the input thereto with a plurality of filters and the results of this convolution are processed by an activation function. Each object detector further includes an object classifier, coupled with the convolutional neural network, for classifying objects in the image according to the results from the convolutional neural network.” – teaches convolution filters to identify features (objects) in the image. In addition to the previous cited passages, Wolf further teaches in [0006] – “Each one of filters 54.sub.1 and 54.sub.2 corresponds to a feature to be identified in the image.” – teaches wherein each of the two or more filters identify each of the one or more features in the image (each filters corresponds to a feature to be identified)).
Regarding claim 30, the combination of Wolf and Matveev teaches the method of claim 25, wherein the one or more feature maps contain values corresponding to the one or more features (Wolf, [0026] – “Each layer of the respective CNN in each one of object detectors 108.sub.1, 108.sub.2, . . . , 108.sub.L convolves the image provided thereto with corresponding filters. The output of each CNN is a map of features corresponding to the filters employed by the CNN. The feature map includes entries of values. Each value of each entry in the feature map represents the feature intensity of the features associated various filters, within an image window associated with the entry.” – teaches feature maps containing values corresponding to one or more features (feature maps of values representing feature intensity of feature associated with filters)).
Claim(s) 8, 16, 24, and 31 is/are rejected under 35 U.S.C. 103 as being unpatentable over Wolf and Matveev as applied to claims 1 and 17 above, and further in view of Shulz-Trieglaff et al. (US Pub. No. 2019/0220704, filed Jan. 2019, hereinafter “Schulz-Trieglaff”).
Regarding claim 8, the combination of Wolf and Matveev teaches the one or more processors of claim 7.
The combination of Wolf and Matveev fails to explicitly teach wherein the convolutional layer contains a depthwise convolution and a pointwise convolution.
However, analogous to the field of convolutional neural networks and identifying features, Schulz-Trieglaff teaches:
wherein the convolutional layer contains a depthwise convolution and a pointwise convolution (Schulz-Trieglaff, [0175] – “Specifically, the convolutional neural network architecture illustrated in FIG. 3D uses depthwise separable convolutions. In contrast to a standard convolution, a depthwise separable convolution performs a separate convolution of each channel of the input data and then performs a pointwise convolution to mix the channels.” – teaches wherein the convolutional layer contains a depthwise convolution and a pointwise convolution).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the depthwise and pointwise convolutions in Schulz-Trieglaff’s convolutional layer to the convolutional layer in object detectors of Wolf and Matveev in order to create a convolutional layer applying depthwise and pointwise convolution. Doing so would enable performing separate convolution of each channel of the input data (Schulz-Trieglaff, [0175]).
Claim 24 and 31 are similar to claim 8, hence similarly rejected.
Regarding claim 16, the combination of Wolf and Matveev teaches the non-transitory machine-readable medium of claim 9, wherein the instructions, when performed, cause the one or more processors to identify the one or more features in a convolutional layer of a convolutional neural network of the one or more neural networks (Wolf, [0027] – “According to the disclosed technique, object detectors in a group of object detectors are associated with common convolution layers (i.e., since the input image to these object detectors is the same). As such, these common convolutional layers need to be computed only once for each group of object detectors.” – teaches the one or more neural networks identify features in a convolutional layer of a convolutional neural network of the one or more neural networks (convolutional layers in object detectors comprised of convolutional neural networks)),
The combination of Wolf and Matveev fails to explicitly teach the convolutional layer including a depthwise convolution and pointwise convolution.
However, analogous to the field of convolutional neural networks and identifying features, Schulz-Trieglaff teaches
the convolutional layer including a depthwise convolution and a pointwise convolution (Schulz-Trieglaff, [0175] – “Specifically, the convolutional neural network architecture illustrated in FIG. 3D uses depthwise separable convolutions. In contrast to a standard convolution, a depthwise separable convolution performs a separate convolution of each channel of the input data and then performs a pointwise convolution to mix the channels.” – teaches wherein the convolutional layer contains a depthwise convolution and a pointwise convolution).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the depthwise and pointwise convolutions in Schulz-Trieglaff’s convolutional layer to the convolutional layer in object detectors of Wolf and Matveev in order to create a convolutional layer applying depthwise and pointwise convolution. Doing so would enable performing separate convolution of each channel of the input data (Shulz-Trieglaff, [0175]).
Response to Arguments
Applicant’s arguments, see pp. 1-4 of Remarks, filed 6 January 2026, have been fully considered but they are not persuasive. Applicant argues, on pp. 2 of Remarks, that the limitation of claim 1 regarding “at least one operation to identify one or more features in the image based, at least in part, on the one or more feature maps” cannot be practically performed in the human mind. Examiner respectfully disagrees. The limitation, as currently drafted, amounts to observing a feature representation, or feature map, and evaluating the representation to identify features of a feature map of an image. A human could observe the feature map, and use evaluation and judgement to identify features of the feature map. The plain meaning of “identifying” encompasses mental observations or evaluations (consistent with the analysis of the “detecting one or more anomalies…” limitations under Step 2A, Prong One of Example 47 claims 2 and 3).
Applicant further argues on pp. 2-3 of Remarks that claim 1 recites additional elements that integrate any alleged abstract idea into a practical application. Examiner respectfully disagrees. The claim does not recite any specific mechanisms for adaptively adjusting the filters or identifying features that achieves the purported improvement. Instead, the claims recite the adjustment of filters and identification of features at a high level of generality. The additional elements of the claim, “one or more processors…” and “circuitry to use one or more neural networks…” are mere instructions to apply the abstract idea on a generic computer (See MPEP 2106.05(f)). Even when viewed as a whole, the claims merely implement the abstract idea using generic computing components and thus fails to integrate the abstract idea into a practical application.
Applicant’s arguments, see pp. 4-6 of Remarks, filed 6 January 2026, with respect to the rejection(s) of claim(s) 1, 9, 17, and 25 under 35 U.S.C. 103 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made over Wolf in view of Matveev et al. (US Patent No. 10,902,318, filed Nov. 2018). Wolf teaches the limitations of claim 1 regarding “at least one operation to generate one or more feature maps…” and “at least one operation to identify one or more features…”. Matveev teaches the limitations of claim 1 regarding “circuitry to use one or more neural networks…” and “at least one operation to generate one or more feature maps for an image using two or more filters…”.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Su et al. (NPL: Pixel-Adaptive Convolutional Neural Networks, dated June 2019) teaches a pixel-adaptive convolution operation in which filter weights are multiplied with a spatially varying kernel that depends on the features located in different positions of a feature map.
Niklaus et al. (NPL: Video Frame Interpolation via Adaptive Separable Convolution, published Aug. 2017) teaches formulating frame interpolation as local separable convolution over input frames using small kernels, in which the separable convolution is adaptive.
Tabernik et al. (NPL: Spatially-Adaptive Filter Units for Deep Neural Networks, published Mar. 2018) teaches a displacement aggregation unit which are learned and enables filters to spatially-adapt their receptive field according to features located in certain positions. Analyzes spatial distributions of the displacement aggregation unit filters and the number of parameters required for spatial convergence within the filter.
Ren et al. (US Patent No. 11,354,577, filed Sept. 2018) teaches systems and methods for generating a convolutional neural networks with filters that may be trimmed and dilate convolutions. Trimming filters includes reducing the dimensions of filters at one or more intermediate convolutional layers.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LOUIS C NYE whose telephone number is 571-272-0636. The examiner can normally be reached Monday - Friday 9:00AM - 5:00PM.
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 at 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 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.
/LOUIS CHRISTOPHER NYE/Examiner, Art Unit 2141
/MATTHEW ELL/Supervisory Patent Examiner, Art Unit 2141
Prosecution Insights