DETAILED ACTION
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 .
Claim Status
2. Claims 1-20 are currently pending.
Response to Arguments
3. Applicant’s arguments with respect to the rejection(s) of claims 1-20, have been considered but are moot in view of the new ground(s) of rejection.
Applicant’s argument in chief alleges at Pg.87-8 of the Remarks of 08/06/2025 that Chen and Jung references fail to teach as cited;
“Viewing the rejection of dependent claims 3 and 4 of the Office Action, the rejection of claim 3 asserts that an output from one "layer" or alleged "one of the activation modules" is an obvious input to another layer, and even if that were so, and even if multiplication and/or addition were somewhere involved, like asserted by the rejection of claim 4, the references do not reasonably suggest” and further "at least one of multiplying and adding an input to the at least one of the activation modules to an output of the at least one of the LeakyReLu function and the convolution function" as claimed.”
Examiner rebuts by stating that the respective sequence of mathematical operations represent the known concept of a deep convolutional network DNN/CNN structure including the claimed layers of a convolution of the input video data at a specific stride including kernels of weighting values, are represented by the Activation function i.e., LeakyReLu layer(s) at which the summing and multiplication operations are performed at the the ReLu or LeakyReLu layer(s), which in fact may be connected in a sequence of deep learning models of network configuration. See Chen at Figs. 2, Par.[0076, 0090], Fig.3, Par.[0077-0078], depicting an Upsampling convolutional process which includes the convolution layers connected to the LeakyReLu in multiple stages and is detailed according to DNN functional steps 395-399, in Fig.3, or Fig.6 Par.[0091], etc..
As it is further obviated by the prior art, the Activation layer of the neural network (e.g., DNN/CNN) i.e., the ReLu or LeakyReLu function involves the claimed subject matter of multiplying and summing the convoluted data according to the NPL, namely “EXXACT” teachings.
An analysis of the amended claim 1, including now partial limitations from the claims 2-6 directly depending from claim 1, defining a sequential “AND” process among limiting steps, warrants a different interpretation and a new search.
The herein response to arguments is considered part of the issued rejection and applies mutatis mutandis.
Applicant’s representative is encouraged to contact the Examiner prior to filing a response to this Office Action in order to reduce the amount of further prosecution.
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 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.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application does not currently name joint inventors.
4. Claims 1-7, 11-17 and 20 are rejected under 35 U.S.C. 103 as being obvious over Hu Chen et al., (hereinafter Chen) (US 2023/0076920) Id. (WO2021/228512A1) and Ju Hyun Jung et al., (hereinafter Jung) (US 2020/0311870) in view of “Activation Functions and optimizers for Deep Learning Models” (hereinafter “EXXACT”) https://www.exxactcorp.com/blog/Deep-Learning/activation-functions-and-optimizers-for-deep-learning-models, Nov. 26, 2019 and further in view of Hongtao Wang et al., (hereinafter Wang) (US 2022/0103845).
Re Claim 1. (currently amended): Chen discloses, a method for video decoding performed by at least one processor (a decoding method implemented at decoder 30 and processor 32 in Fig.12, 13, or Fig.15, Par.[0019, 0164]..), the method comprising:
receiving a video bitstream comprising a current block in a current picture (receiving at video decoder 30, encoded picture data 21, Par.[0164] as non-overlapping video blocks 203, Par.[0181-0186] Fig.14); and
reconstructing the current block by transforming the current block by a neural network comprising a plurality of upsample modules (reconstructing the image at a neural network by applying a plurality of upsampling and a plurality of convolutional layers, per process 298 Fig.2 or Fig.6, Par.[0069-0070 ) and activation modules (the neural network including activation function layers, Par.[0069-0071] e.g., ReLu in Fig.3),
at least one of the upsample modules comprises of a convolution layer (per Fig.6 and Fig.2, each upsampling stage comprises a convolution layer, Par.[0076]) and
Chen does not expressly teach about applying a pixel shuffle layer to the neural network, but in an analogous art,
Jung, teaches about, a pixel shuffle layer (applying a pixel shuffle layer to the neural network, Par.[0012, 0019, 0069]), and
at least one of the activation modules comprises a LeakyReLu function and a convolution function (the activation function may be a leaky rectified linear unit LReLU, Par.[0013, 0020, 0069], being recognized by the skilled in the art that the activation layer may include leaky rectified linear units, LReLUs functions, Par.0013 which are typically connected to the output of the convolution layers as part of the neural network filter, Abstract and Par.[0007-0011, 0013-0018, 0020, 0069] where the activation parameter corresponding to an input channel is included in the next layer, as well as is an output of one channel, Par.[0060]and further comprising a convolution layer in order to enhance the resolution, Par.[0070-0071]), and
The ordinary skilled in the art would have found obvious before the effective filing date of invention to consider the coding apparatus and method using neural networks, taught by Chen where according suggestions indicating the application of high resolution solutions at a decoding process during the up-convolution and concatenation to Fig.2, and Par.[0090] and find the incentive to examine other methods of increasing the resolution at decoder as described in Jung, by adding a pixel-shuffle layer to the convolution layer in order to reduce the computational complexity, as indicated at Par.[0069], thus representing a predictable combination in lieu of the claimed matter.
However, as necessitated by the amendment, the art to “EXXACT” is found to teach the mathematical function of the Activation layer LeakyReLu used in neural networks representing the intermediary process of the current block reconstruction as further detailed,
reconstructing the current block comprises at least one of multiplying and adding an input to the at least one of the activation modules to an output of the at least one of the LeakyReLu function and the convolution function (where the Deep Learning Models using neural networks, DNN based on CNN, inherently use non-linear models based on LeakyReLu Activation functions, applied over the output data of a particular layer of neurons before being connected to the next layer, Pg.3/23, e.g., by having the neurons x1…. Xn, are multiplied by the Weight vectors W1….Wn, of a specific kernel being applied at the input of the Activation – LeakyReLu function where are summed according to the function
PNG
media_image1.png
200
400
media_image1.png
Greyscale
and depicted in the Image Source at Pg.4/23 shown below
PNG
media_image2.png
200
400
media_image2.png
Greyscale
. It is to be noted that the term b, in the equation represents an added bias to enhance the error response, and it is optional to the model).
;wherein at least one of the upsample modules comprises of a convolution layer and a pixel shuffle layer, and wherein at least one of the activation modules comprises a LeakyReLu function and a convolution function.
Furthermore, Wang defines the DNN configuration claiming,
at least one of the upsample modules comprises of a convolution layer and a pixel shuffle layer, and at least one of the activation modules comprises a LeakyReLu function and a convolution function, and reconstructing the current block comprises at least one of multiplying and adding an input to the at least one of the activation modules to an output of the at least one of the LeakyReLu function and the convolution function (according to Fig.7 and the description given for multiple Activation functions design, e.g., Leaky ReLu at Par.[0005, 0018] and the connectivity to the input and to convolution layers at Par.[0099-0104]).
The one of ordinary skill in the art would have found obvious before the effective filing date of invention to consider the combined apparatus and method using neural networks, taught by Chen and Jung, and to seek further operational explanation regarding to the activation layer connectivity to various neuron layers generally accepted as convolution output neurons in a single hidden layer or in subsequent refinement layers as detailed by “EXXACT” in order to obtain eliminate the “dead neuron” phenomena by eliminating the negative inputs thus avoiding the “vanishing gradient” or saturation problem, taught at Pgs.10-13 and the respective figures, along with the connectivity of ReLu to Convolution layers in a sequential input/output description found in Wang, hence deeming the combination predictable in lieu of the claimed matter.
Re Claim 2. (original): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 1,
Jung teaches the, wherein the at least one of the upsample modules consists of the convolution layer and the pixel shuffle layer (the deep neural network being comprised of multiple convolution layers includes pixel shuffle layers(s) Par.[0012, 0019, 0069]).
Re Claim 3. (currently amended): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 1,
Jung teaches the, wherein the at least one of the activation modules comprises a plurality of the LeakyReLu functions, including the LeakyReLu function and a second LeakyReLu function,
such that an output of the second LeakyRelu function is output to the convolution function and an output of the convolution function is output to a second the LeakyReLu function of the plurality of LeakyReLu functions, and
an output of the LeakyRelu function is the output of the at least one of the LeakyReLu function and the convolution function (being recognized by the skilled in the art that the activation layer may include multiple leaky rectified linear units, LReLUs functions, alternately connected, Par.[0011, 0013] which are typically connected to the output of the convolution layers as part of the neural network filter, Abstract and Par.[0007-0011, 0013-0018, 0020, 0069] where the activation parameter corresponding to an input channel is included in the next layer, as well as is an output of one channel, Par.[0060]).
Re Claim 4. (currently amended): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 1,
Jung teaches, wherein an output of the LeakyRelu function is output to the convolution function and
the output of the at least one of the LeakyReLu function and the convolution function is an output of the convolution function (multiple leaky rectified linear units, LReLUs functions, alternately connected, Par.[0011, 0013] which are typically connected to the output of the convolution layers as part of the neural network filter, Abstract and Par.[0007-0011, 0013-0018, 0020, 0069] where the activation parameter corresponding to an input channel is included in the next layer, as well as is an output of one channel, Par.[0060]) and
“EXXACT” teaches that,
is output to a multiplication function, the multiplication function multiplying the input to the at least one of the activation modules to the output of the convolution function (using non-linear models based on LeakyReLu Activation functions, applied over the output data of a particular layer of neurons before being connected to the next layer, Pg.3/23, e.g., by having the neurons x1…. Xn, are multiplied by the Weight vectors W1….Wn, of a specific kernel being applied at the input of the Activation – LeakyReLu function where are summed according to the function
PNG
media_image1.png
200
400
media_image1.png
Greyscale
and depicted in the Image Source at Pg.4/23 shown below
PNG
media_image2.png
200
400
media_image2.png
Greyscale
alternately connected per Jung at claim 3).
Re Claim 5. (original): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 4,
Chen teaches the, wherein an output of the multiplication function is output to an addition function (the reconstruction unit adder (214), sums the residual (213), sample by sample, to prediction block (265), Par.[0177]).
“EXXACT” teaches this matter at, (Pg.3/23, e.g., by having the neurons x1…. Xn, are multiplied by the Weight vectors W1….Wn, of a specific kernel being applied at the input of the Activation – LeakyReLu function where are summed according to the function
PNG
media_image1.png
200
400
media_image1.png
Greyscale
and depicted in the Image Source at Pg.4/23 etc.)
Re Claim 6. (currently amended): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 1,
Jung teaches the, wherein an output of the LeakyRelu function is output to the convolution function (the alternate connectivity of convolution and activation per Par.[0011, 0013, 0020] according to Chen at path 298 Par.[0069-0071]) and
the output of the at least one of the LeakyReLu function and the convolution function is an output of the convolution function and is output to an addition function, the addition function adding the input to the at least one of the activation modules to the output of the convolution function (Par.[0011]).
Re Claim 7. (original): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 1,
Chen teaches the, wherein the convolution function comprises a 1x1 filter (the 1x1 final layer convolution 395, is used to map the 64-component feature vector, Par.[0078]).
Re Claim 11. (currently amended): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 1, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 12. (original): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 2, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 13. (currently amended): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 3, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 14. (currently amended): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 4, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 15. (original): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 5, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 16. (currently amended): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 6, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 17. (original): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 7, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 20. (currently amended): This claim represents the non-transitory computer readable medium storing a program taught in (Chen: non-transitory medium at Par.[0030]), comprising the program performing each and every limitation of the method of claim 1, hence it is rejected on the same mapped evidence mutatis mutandis.
5. Claims 8-10 and 18-19 are rejected under 35 U.S.C. 103 as being obvious over Chen, Jung and “EXXACT” along with Wang, in view of Zhaobin Zhang et al., (hereinafter Zhang) (WO 2024/186738A1) in lieu of Prov. No. 63/449,861.
Re Claim 8. (original): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 1,
Zhang teaches the limitation, wherein each of the plurality of upsample modules comprises respectively different numbers of kilo multiply and accumulate (KMAC) operations (using the KMAC/pixel as the decoding time reducing metric, Sec.4.1).
It would have been obvious to the ordinary skilled in the art before the effective filing date of invention to seek methods of reduced decoding operations over complex neural networks taught by Chen and Jung and consequentially associate with other efficient neural network prediction methods found in Zhang’s, teachings, by reducing the number of channels, the number of total convolution layers or kernels, (Sec.3.1, Tables 1 and 2) to simplify the decoder prediction and decoding time, (Sec.4.1), hence deeming the combination predictable.
Re Claim 9. (original): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 8,
Zhang teaches, wherein one of the plurality of upsample modules comprises more than double a number of KMAC operations than does any of the other of the plurality of upsample modules (examples for the number of channels of luma and chroma are given for different number of convolution layers, input size and kernel size as being a multiple of operations for decoder scaling, for Hyper Scale Decoder, at Pg.28-29 Table 2).
Re Claim 10. (original): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 8,
Chen teaches this limitation, wherein the plurality of upsample modules consists of four upsample modules (upsampling the feature map by a (one) 2x2 convolution of the cropped, two 3x3 convolutions and a final layer of 1x1 convolution are used to map each of the 64-component feature vector, Par.[0078]).
Re Claim 18. (original): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 8, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 19. (original): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 9, hence it is rejected on the same evidentiary probe mutatis mutandis.
Conclusion
6. THIS ACTION IS MADE FINAL. 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 mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DRAMOS KALAPODAS whose telephone number is (571)272-4622. The examiner can normally be reached on Monday-Friday 8am-5pm.
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, David Czekaj can be reached on 571-272-7327. 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.
DRAMOS . KALAPODAS
Primary Examiner
Art Unit 2487
/DRAMOS KALAPODDETAILED ACTION
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 .
Claim Status
2. Claims 1-20 are currently pending.
Response to Arguments
3. Applicant’s arguments with respect to the rejection(s) of claims 1-20, have been considered but are moot in view of the new ground(s) of rejection.
Applicant’s argument in chief alleges at Pg.87-8 of the Remarks of 08/06/2025 that Chen and Jung references fail to teach as cited;
“Viewing the rejection of dependent claims 3 and 4 of the Office Action, the rejection of claim 3 asserts that an output from one "layer" or alleged "one of the activation modules" is an obvious input to another layer, and even if that were so, and even if multiplication and/or addition were somewhere involved, like asserted by the rejection of claim 4, the references do not reasonably suggest” and further "at least one of multiplying and adding an input to the at least one of the activation modules to an output of the at least one of the LeakyReLu function and the convolution function" as claimed.”
Examiner rebuts by stating that the respective sequence of mathematical operations represent the known concept of a deep convolutional network DNN/CNN structure including the claimed layers of a convolution of the input video data at a specific stride including kernels of weighting values, are represented by the Activation function i.e., LeakyReLu layer(s) at which the summing and multiplication operations are performed at the the ReLu or LeakyReLu layer(s), which in fact may be connected in a sequence of deep learning models of network configuration. See Chen at Figs. 2, Par.[0076, 0090], Fig.3, Par.[0077-0078], depicting an Upsampling convolutional process which includes the convolution layers connected to the LeakyReLu in multiple stages and is detailed according to DNN functional steps 395-399, in Fig.3, or Fig.6 Par.[0091], etc..
As it is further obviated by the prior art, the Activation layer of the neural network (e.g., DNN/CNN) i.e., the ReLu or LeakyReLu function involves the claimed subject matter of multiplying and summing the convoluted data according to the NPL, namely “EXXACT” teachings.
An analysis of the amended claim 1, including now partial limitations from the claims 2-6 directly depending from claim 1, defining a sequential “AND” process among limiting steps, warrants a different interpretation and a new search.
The herein response to arguments is considered part of the issued rejection and applies mutatis mutandis.
Applicant’s representative is encouraged to contact the Examiner prior to filing a response to this Office Action in order to reduce the amount of further prosecution.
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 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.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application does not currently name joint inventors.
4. Claims 1-7, 11-17 and 20 are rejected under 35 U.S.C. 103 as being obvious over Hu Chen et al., (hereinafter Chen) (US 2023/0076920) Id. (WO2021/228512A1) and Ju Hyun Jung et al., (hereinafter Jung) (US 2020/0311870) in view of “Activation Functions and optimizers for Deep Learning Models” (hereinafter “EXXACT”) https://www.exxactcorp.com/blog/Deep-Learning/activation-functions-and-optimizers-for-deep-learning-models, Nov. 26, 2019 and further in view of Hongtao Wang et al., (hereinafter Wang) (US 2022/0103845).
Re Claim 1. (currently amended): Chen discloses, a method for video decoding performed by at least one processor (a decoding method implemented at decoder 30 and processor 32 in Fig.12, 13, or Fig.15, Par.[0019, 0164]..), the method comprising:
receiving a video bitstream comprising a current block in a current picture (receiving at video decoder 30, encoded picture data 21, Par.[0164] as non-overlapping video blocks 203, Par.[0181-0186] Fig.14); and
reconstructing the current block by transforming the current block by a neural network comprising a plurality of upsample modules (reconstructing the image at a neural network by applying a plurality of upsampling and a plurality of convolutional layers, per process 298 Fig.2 or Fig.6, Par.[0069-0070 ) and activation modules (the neural network including activation function layers, Par.[0069-0071] e.g., ReLu in Fig.3),
at least one of the upsample modules comprises of a convolution layer (per Fig.6 and Fig.2, each upsampling stage comprises a convolution layer, Par.[0076]) and
Chen does not expressly teach about applying a pixel shuffle layer to the neural network, but in an analogous art,
Jung, teaches about, a pixel shuffle layer (applying a pixel shuffle layer to the neural network, Par.[0012, 0019, 0069]), and
at least one of the activation modules comprises a LeakyReLu function and a convolution function (the activation function may be a leaky rectified linear unit LReLU, Par.[0013, 0020, 0069], being recognized by the skilled in the art that the activation layer may include leaky rectified linear units, LReLUs functions, Par.0013 which are typically connected to the output of the convolution layers as part of the neural network filter, Abstract and Par.[0007-0011, 0013-0018, 0020, 0069] where the activation parameter corresponding to an input channel is included in the next layer, as well as is an output of one channel, Par.[0060]and further comprising a convolution layer in order to enhance the resolution, Par.[0070-0071]), and
The ordinary skilled in the art would have found obvious before the effective filing date of invention to consider the coding apparatus and method using neural networks, taught by Chen where according suggestions indicating the application of high resolution solutions at a decoding process during the up-convolution and concatenation to Fig.2, and Par.[0090] and find the incentive to examine other methods of increasing the resolution at decoder as described in Jung, by adding a pixel-shuffle layer to the convolution layer in order to reduce the computational complexity, as indicated at Par.[0069], thus representing a predictable combination in lieu of the claimed matter.
However, as necessitated by the amendment, the art to “EXXACT” is found to teach the mathematical function of the Activation layer LeakyReLu used in neural networks representing the intermediary process of the current block reconstruction as further detailed,
reconstructing the current block comprises at least one of multiplying and adding an input to the at least one of the activation modules to an output of the at least one of the LeakyReLu function and the convolution function (where the Deep Learning Models using neural networks, DNN based on CNN, inherently use non-linear models based on LeakyReLu Activation functions, applied over the output data of a particular layer of neurons before being connected to the next layer, Pg.3/23, e.g., by having the neurons x1…. Xn, are multiplied by the Weight vectors W1….Wn, of a specific kernel being applied at the input of the Activation – LeakyReLu function where are summed according to the function
PNG
media_image1.png
200
400
media_image1.png
Greyscale
and depicted in the Image Source at Pg.4/23 shown below
PNG
media_image2.png
200
400
media_image2.png
Greyscale
. It is to be noted that the term b, in the equation represents an added bias to enhance the error response, and it is optional to the model).
Furthermore, Wang defines the DNN configuration claiming,
at least one of the upsample modules comprises of a convolution layer and a pixel shuffle layer, and at least one of the activation modules comprises a LeakyReLu function and a convolution function, and reconstructing the current block comprises at least one of multiplying and adding an input to the at least one of the activation modules to an output of the at least one of the LeakyReLu function and the convolution function (according to Fig.7 and the description given for multiple Activation functions design, e.g., Leaky ReLu at Par.[0005, 0018] and the connectivity to the input and to convolution layers at Par.[0099-0104]).
The one of ordinary skill in the art would have found obvious before the effective filing date of invention to consider the combined apparatus and method using neural networks, taught by Chen and Jung, and to seek further operational explanation regarding to the activation layer connectivity to various neuron layers generally accepted as convolution output neurons in a single hidden layer or in subsequent refinement layers as detailed by “EXXACT” in order to obtain eliminate the “dead neuron” phenomena by eliminating the negative inputs thus avoiding the “vanishing gradient” or saturation problem, taught at Pgs.10-13 and the respective figures, along with the connectivity of ReLu to Convolution layers in a sequential input/output description found in Wang, hence deeming the combination predictable in lieu of the claimed matter.
Re Claim 2. (original): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 1,
Jung teaches the, wherein the at least one of the upsample modules consists of the convolution layer and the pixel shuffle layer (the deep neural network being comprised of multiple convolution layers includes pixel shuffle layers(s) Par.[0012, 0019, 0069]).
Re Claim 3. (currently amended): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 1,
Jung teaches the, wherein the at least one of the activation modules comprises a plurality of the LeakyReLu functions, including the LeakyReLu function and a second LeakyReLu function,
such that an output of the second LeakyRelu function is output to the convolution function and an output of the convolution function is output to the LeakyReLu function of the plurality of LeakyReLu functions, and
an output of the LeakyRelu function is the output of the at least one of the LeakyReLu function and the convolution function (being recognized by the skilled in the art that the activation layer may include multiple leaky rectified linear units, LReLUs functions, alternately connected, Par.[0011, 0013] which are typically connected to the output of the convolution layers as part of the neural network filter, Abstract and Par.[0007-0011, 0013-0018, 0020, 0069] where the activation parameter corresponding to an input channel is included in the next layer, as well as is an output of one channel, Par.[0060]).
Re Claim 4. (currently amended): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 1,
Jung teaches, wherein an output of the LeakyRelu function is output to the convolution function and
the output of the at least one of the LeakyReLu function and the convolution function is an output of the convolution function (multiple leaky rectified linear units, LReLUs functions, alternately connected, Par.[0011, 0013] which are typically connected to the output of the convolution layers as part of the neural network filter, Abstract and Par.[0007-0011, 0013-0018, 0020, 0069] where the activation parameter corresponding to an input channel is included in the next layer, as well as is an output of one channel, Par.[0060]) and
“EXXACT” teaches that,
is output to a multiplication function, the multiplication function multiplying the input to the at least one of the activation modules to the output of the convolution function (using non-linear models based on LeakyReLu Activation functions, applied over the output data of a particular layer of neurons before being connected to the next layer, Pg.3/23, e.g., by having the neurons x1…. Xn, are multiplied by the Weight vectors W1….Wn, of a specific kernel being applied at the input of the Activation – LeakyReLu function where are summed according to the function
PNG
media_image1.png
200
400
media_image1.png
Greyscale
and depicted in the Image Source at Pg.4/23 shown below
PNG
media_image2.png
200
400
media_image2.png
Greyscale
alternately connected per Jung at claim 3).
Re Claim 5. (original): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 4,
Chen teaches the, wherein an output of the multiplication function is output to an addition function (the reconstruction unit adder (214), sums the residual (213), sample by sample, to prediction block (265), Par.[0177]).
“EXXACT” teaches this matter at, (Pg.3/23, e.g., by having the neurons x1…. Xn, are multiplied by the Weight vectors W1….Wn, of a specific kernel being applied at the input of the Activation – LeakyReLu function where are summed according to the function
PNG
media_image1.png
200
400
media_image1.png
Greyscale
and depicted in the Image Source at Pg.4/23 etc.)
Re Claim 6. (currently amended): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 1,
Jung teaches the, wherein an output of the LeakyRelu function is output to the convolution function (the alternate connectivity of convolution and activation per Par.[0011, 0013, 0020] according to Chen at path 298 Par.[0069-0071]) and
the output of the at least one of the LeakyReLu function and the convolution function is an output of the convolution function and is output to an addition function, the addition function adding the input to the at least one of the activation modules to the output of the convolution function (Par.[0011]).
Re Claim 7. (original): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 1,
Chen teaches the, wherein the convolution function comprises a 1x1 filter (the 1x1 final layer convolution 395, is used to map the 64-component feature vector, Par.[0078]).
Re Claim 11. (currently amended): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 1, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 12. (original): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 2, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 13. (currently amended): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 3, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 14. (currently amended): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 4, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 15. (original): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 5, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 16. (currently amended): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 6, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 17. (original): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 7, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 20. (currently amended): This claim represents the non-transitory computer readable medium storing a program taught in (Chen: non-transitory medium at Par.[0030]), comprising the program performing each and every limitation of the method of claim 1, hence it is rejected on the same mapped evidence mutatis mutandis.
5. Claims 8-10 and 18-19 are rejected under 35 U.S.C. 103 as being obvious over Chen, Jung and “EXXACT” along with Wang, in view of Zhaobin Zhang et al., (hereinafter Zhang) (WO 2024/186738A1) in lieu of Prov. No. 63/449,861.
Re Claim 8. (original): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 1,
Zhang teaches the limitation, wherein each of the plurality of upsample modules comprises respectively different numbers of kilo multiply and accumulate (KMAC) operations (using the KMAC/pixel as the decoding time reducing metric, Sec.4.1).
It would have been obvious to the ordinary skilled in the art before the effective filing date of invention to seek methods of reduced decoding operations over complex neural networks taught by Chen and Jung and consequentially associate with other efficient neural network prediction methods found in Zhang’s, teachings, by reducing the number of channels, the number of total convolution layers or kernels, (Sec.3.1, Tables 1 and 2) to simplify the decoder prediction and decoding time, (Sec.4.1), hence deeming the combination predictable.
Re Claim 9. (original): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 8,
Zhang teaches, wherein one of the plurality of upsample modules comprises more than double a number of KMAC operations than does any of the other of the plurality of upsample modules (examples for the number of channels of luma and chroma are given for different number of convolution layers, input size and kernel size as being a multiple of operations for decoder scaling, for Hyper Scale Decoder, at Pg.28-29 Table 2).
Re Claim 10. (original): Chen, Jung and “EXXACT” along with Wang disclose, the method according to claim 8,
Chen teaches this limitation, wherein the plurality of upsample modules consists of four upsample modules (upsampling the feature map by a (one) 2x2 convolution of the cropped, two 3x3 convolutions and a final layer of 1x1 convolution are used to map each of the 64-component feature vector, Par.[0078]).
Re Claim 18. (original): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 8, hence it is rejected on the same evidentiary probe mutatis mutandis.
Re Claim 19. (original): This claim represents the decoding apparatus comprising at least one memory (Chen: memory at Par.[0020, 0027]) and at least one processor (Chen: processors at Par.[0024, 0030]), implementing each and every limitation of the method claim 9, hence it is rejected on the same evidentiary probe mutatis mutandis.
Conclusion
6. THIS ACTION IS MADE FINAL. 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 mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DRAMOS KALAPODAS whose telephone number is (571)272-4622. The examiner can normally be reached on Monday-Friday 8am-5pm.
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, David Czekaj can be reached on 571-272-7327. 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.
DRAMOS . KALAPODAS
Primary Examiner
Art Unit 2487
/DRAMOS KALAPODAS/
AS/
Prosecution Insights