DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
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 allowance or after an Office action under Ex Parte Quayle, 25 USPQ 74, 453 O.G. 213 (Comm'r Pat. 1935). 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, prosecution in this application has been reopened pursuant to 37 CFR 1.114. Applicant's submission filed on 03/04/2026 has been entered.
Notice to Applicant
This application has been reopened in light of the Information Disclosure Statement filed by applicant on 03/04/2026.
Limitations appearing inside of {} are intended to indicate the limitations not taught by said prior art(s)/combinations.
Claims 1 -2,5-12 and 15-20 are pending.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1, 2, 7-12, 17-20 are rejected under 35 U.S.C. 103 as being unpatentable over “Yao” (Shuochao Yao, Jinyang Li, Dongxin Liu, Tianshi Wang, Shengzhong Liu, Huajie Shao, and Tarek Abdelzaher. 2020. Deep compressive offloading: speeding up neural network inference by trading edge computation for network latency. In Proceedings of the 18th Conference on Embedded Networked Sensor Systems (SenSys '20). Association for Computing Machinery, New York, NY, USA, 476–488. https://doi.org/10.1145/3384419.3430898), as cited in the IDS (02/26/2026), in view of :Kothari”, (Kothari et al., US 20230038935 A1) .
Regarding claim 20, Yao teaches an offloading system comprising:
a terminal device; and a server device, wherein the terminal device comprises: a camera configured to obtain an image corresponding to original data (Yao, [§1, p477, col 2, ¶4]; We implement this system on Android mobile devices and a Linux edge server with GPUs. Android mobile devices include cameras that may obtain an image);
a first memory storing one or more instructions (Yao, [§3.2.1, p479, col 2, ¶2]; Once we have trained the encoder and decoder, we can deploy them on the local device and edge side respectively; and see §5.1 software); and at least one first processor configured to execute the one or more instructions stored in the first memory to (Yao, [§5, p483, col 1, ¶1]; The mobile client is implemented on Android OS and tested on two different Android phones. Google Pixel is equipped with a Quad-core (2x2.15 GHz & 2x1.6 GHz) Kryo CPU and Adreno 530 GPU; Nexus 6 is equipped with a 2.7 GHz quad-core Krait 450 CPU and Adreno 420 GPU.):
generate latent representation data by using an extractor model having received the original data as an input (Yao, [§3, p478, col 2, ¶1]; lightweight encoder on the mobile side to compress the data to transfer); and
offload the latent representation data onto the server device (Yao, [§3, p478, col 2, ¶1]; a decoder on the edge server side to reconstruct the transferred data.),
the server device (Yao, [§5, p483, col 1, ¶1]; edge server) comprises: a second memory storing one or more instructions (Yao, [§3.2.1, p479, col 2, ¶2]; Once we have trained the encoder and decoder, we can deploy them on the local device and edge side respectively; and see §5.1 software); and at least one second processor configured to execute the one or more instructions stored in the second memory to (Yao, [§5, p483, col 1, ¶1]; The edge server is a Linux desktop equipped with an Intel Core i7-5820K CPU and two types of GPUs, including Nvidia Titan V and Nvidia GeForce GTX Titan X):
receive offloaded data from the terminal device (Yao, [§, p, col, ¶]; See Fig 2, shown below, exhibits edge server receiving offloaded data from the terminal device.
PNG
media_image1.png
312
581
media_image1.png
Greyscale
);
{modify the offloaded data to fit a structure of an input to a deep neural network model, by using an input structure modifier }; and
output inferred data corresponding to the offloaded data by using the deep neural network model having received the modified offloaded data as an input (Yao, [§3.2.1, p479, col 2, ¶2]; we can reconstruct the data with a one-shot inference of the decoder, x̂ = Gθ (y), and see Figure 3, shown below, exhibits reconstructed image x̂ within the EDGE server side.
PNG
media_image2.png
318
1125
media_image2.png
Greyscale
),
the extractor model and the deep neural network model are jointly trained by using loss information of the deep neural network model (Yao, [§3.2.1, p479, col 2, ¶1]; we are training an encoder Eϕ and a decoder Gθ (·) that can jointly compress and reconstruct the data during the offloading),
a size of the latent representation data is predefined (Yao, [§3.2.3, p481, col1, ¶1]; To have a lightweight encoder on the mobile and embedded devices, we set the convolution strides to be equal to the kernel sizes (equals to 4×4 by default)), and
{the outputting of the inferred data comprises inputting the modified offloaded data into the deep neural network model without decoding}.
Yao does not explicitly disclose modify the offloaded data to fit a structure of an input to a deep neural network model, by using an input structure modifier; and the outputting of the inferred data comprises inputting the modified offloaded data into the deep neural network model without decoding.
However, Kothari, a similar field of endeavor of machine learning model (e.g., neural network) that compresses input data, teaches modify the offloaded data to fit a structure of an input to a deep neural network model, by using an input structure modifier (The latent space representation may be reformatted or sampled (¶[0054]) or additional attributes may be concatenated to the latent space representation and then passed to the degradation engine 209 which outputs the final inference (¶[0055]); ¶[0054] the sampling engine 212 may format the latent space representation for passing to the concatenation engine 207 and/or may take a sampling that represents the vector of means 203 and/or the vector of standard deviations 205. In some examples, the sampling engine 212 may perform sampling differently during training than during inferencing. ¶[0055] The concatenation engine 207 may concatenate the latent space representation with an attribute or attributes 206 to produce concatenated information. The concatenated information may be provided to the degradation engine 209. In some examples, the concatenation engine 207 may concatenate the latent space representation with the attribute(s) 206 as described in relation to FIG. 1. For instance, the concatenation engine 207 may concatenate the latent space representation with location (e.g., x, y, z coordinates in the build volume), initial stress, initial quality metric (e.g., initial b*), temperature, and/or time (e.g., time increment), etc.).
Kothari further teaches the outputting of the inferred data comprises inputting the modified offloaded data into the deep neural network model without decoding (Kothari, [0031] In some examples, the decoder and/or decoder output (of the variational autoencoder model, for instance) may not be utilized after training. For instance, after the variational autoencoder model (e.g., network) is trained, the decoder of the variational autoencoder model may be removed. The trained encoder may be utilized to extract the latent space representation at an inferencing stage and/or runtime.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include modifying the offloaded data as taught by Kothari to the invention of Yao. The motivation to do so would be to utilize a machine learning model (e.g., regression prediction model) that can make an inference based on the input, i.e., concatenated information, and for example to meet hardware requirements which may reduce latency and improve accuracy of inference.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include inference without using a decoder as taught by Kothari to the invention of Yao. The motivation to do so would be because the model has been trained using the decoder and may infer output without the decoder using latent representation data improving prediction accuracy.
In addition, Torfason (2018), previously cited, shows that inference can be done without the decoder, see Fig 2, shown below, however Torfason is not relied upon for the rejection of the instant claim.
PNG
media_image3.png
495
307
media_image3.png
Greyscale
.
Amended claims 1 and 11 are similarly analyzed as analogous claim 20. Yao further discloses wherein the offloaded data includes latent representation data generated by an extractor model having received original data as an input (Yao, [§1, p477, col 1, ¶4]; our system will offload not only well-studied data types but also intermediate neural network features, whose sparse properties are unknown.),
Regarding claim 2, the combination of Yao and Kothari discloses the method of claim 1. Yao further wherein the extractor model is implemented as an encoder model of an autoencoder model (Yao, [§6.2, p484, col 1, ¶5]; Offload-AE+: leverages the state-of-the-art deep learning data compression technique [35] to compress (including quantization and coding) and reconstruct the offloaded data based on the autoencoder structure, called Offload-AE.).
Regarding claim 7, the combination of Yao and Kothari teaches the method of claim 1. Yao further teaches wherein the extractor model is trained by using a knowledge distillation technique (Yao teaches distilling knowledge from the intermediate feature. Yao, [§3.3, p481, col 2, ¶1]; see excerpt shown below:
PNG
media_image4.png
427
705
media_image4.png
Greyscale
).
Regarding claim 8, the combination of Yao and Kothari discloses the method of claim 1. Yao further teaches wherein the original data includes at least one of an image, a video, an audio, a text, and a sensor value to be used in an application using the deep neural network model (Yao, [§1, p477, col 1, ¶4]; data types, such as images or voice).
Regarding claim 9, the combination of Yao and Kothari discloses the method of claim 1. Yao further teaches wherein the deep neural network model is a model that performs image classification, image segmentation, image captioning, object detection, depth estimation, localization, or pose estimation, based on the original data (Yao, [§6.1, p483, col 2, ¶1]; object detection; [¶2]; The image recognition service classifies an image into one of 1000 object categories).
Regarding claim 10, the combination of Yao and Kothari discloses the method of claim 1. Yao further discloses further comprising transmitting the inferred data to the terminal device (Yao, [§5.2, p483, col 2, ¶3]; The inference result is transferred back to the mobile device via the same link).
Claim 12 is similarly analyzed as analogous claim 2.
Claim 17 is similarly analyzed as analogous claim 7.
Claim 18 is similarly analyzed as analogous claim 8.
Claim 19 is similarly analyzed as analogous claim 9.
Claims 5 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Yao in view of Kothari, and further in view of Zhang, et. al., US 20220292654 A1, hereinafter Zhang.
Regarding claim 5, the combination of Yao and Kothari discloses the method of claim 4. The combination does not explicitly disclose wherein the decoder model includes a single upsampling layer and a single convolutional layer.
However, Zhang, a similar field of endeavor of generating images that depict original images using neural networks, discloses wherein the decoder model includes a single upsampling layer and a single convolutional layer (Zhang, ¶[0107]; decoder 606 or neural network decoder 422 includes a number of layers, including ConvBlock layers, Upsampling layers, and a single convolutional layer).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include a single upsampling layer and a single convolutional layer in the decoder as taught by Zhang to the combined invention of Yao and Kothari. The motivation to do so would be to generate decoded images with the same dimensions as the original image, which may then be used for generating inferred images with the same dimensions as the original, for the purpose of generating lossless images as perceived by human vision.
Claim 15 is similarly analyzed as analogous claim 5.
Claims 6 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Yao in view of Kothari, and further in view of Aytekin, et. al., WO 2019185987 A1, hereinafter Aytekin.
Regarding claim 6, the combination of Yao and Kothari teaches the method of claim 1. Wherein the deep neural network model is a first deep neural network model, the inferred data is first inferred data (Yao, [§3.2.1, p479, col 2, ¶1]; we are training an encoder Eϕ and a decoder Gθ (·) that can jointly compress and reconstruct the data during the offloading),
{the method further comprises outputting second inferred data corresponding to the offloaded data by using a second deep neural network model having received the modified offloaded data as an input, and
the extractor model is jointly trained by using loss information of the first deep neural network model and loss information of the second deep neural network model.}
Yao does not explicitly disclose the method further comprises outputting second inferred data corresponding to the offloaded data by using a second deep neural network model having received the modified offloaded data as an input, and
the extractor model is jointly trained by using loss information of the first deep neural network model and loss information of the second deep neural network model.
However, Aytekin, a similar field of endeavor of use of neural networks such as autoencoders for compressing and de-compressing data, teaches the method further comprises outputting second inferred data corresponding to the offloaded data by using a second deep neural network model having received the modified offloaded data as an input (Aytekin, Fig 5B, shown below, and ¶[0062]; neural networks 700A, 700B, and 700C, used in an inference phase), and
the extractor model is jointly trained by using loss information of the first deep neural network model and loss information of the second deep neural network model (Aytekin, ¶[0035], and See Fig 1, shown below, The final loss 165 is used in a training operation 170 (of the training method 101), the outputs 171, 172 of which are respectively routed back to the neural encoder 120 and the neural decoder 130.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include multiple neural networks for inferred data, and using the loss as taught by Aytekin to the combined invention of Yao and Kothari. The motivation to do so would be to have multiple neural networks using various techniques, that are “expert” neural networks that may then be used at inference time.
Claim 16 is similarly analyzed as analogous claim 6.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. See PTO-892 Notice of References Cited.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to CHANDHANA PEDAPATI whose telephone number is (571)272-5325. The examiner can normally be reached M-F 8:30am-6pm (ET).
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, Chan Park can be reached on 571-272-7409. 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.
/CHANDHANA PEDAPATI/Examiner, Art Unit 2669
/JOHN B STREGE/Primary Examiner, Art Unit 2669
Prosecution Insights