DETAILED ACTION
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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 April 19, 2026, has been entered.
Response to Amendment
Claims 1-5, 7-15, and 17-20 were previously pending. Applicant’s amendment filed April 19, 2026, has been entered in full. Claims 1, 9, 11 and 19 are amended. No claims are added or cancelled. Accordingly, claims 1-5, 7-15, and 17-20 remain pending.
Response to Arguments
Applicant requests acceptance of the amended drawings (Remarks filed April 19, 2026, hereinafter Remarks: Page 6). Examiner agrees that the amended drawing is adequately supported and the amended drawings are accepted.
Applicant traverses the previous rejections under 35 U.S.C. 103 (Remarks: Pages 6-9). Examiner respectfully disagrees.
Applicant notes that Qin determines a GPU is overloaded based on a rendering command with a high resolution and argues that this is prediction, rather than detection (Remarks: Page 7). Examiner respectfully disagrees. Qiu recognizes that a GPU is overloaded when tasked with rendering a high-resolution frame (e.g., [0073]). Qiu detects this overloaded state by, for example, determining whether a rendering command calls for the GPU to render a frame with resolution higher than a threshold (e.g., [0084], [0086]). This detection process is continued while the GPU renders graphics in a video (e.g., Fig. 9, left rendering commands are delivered and intercepted for each frame; e.g., Fig. 9, right, the detection is performed during a process of rendering graphics using the GPU).
To interpret “detecting” as precluding any sort of inference or prediction of the overloading state based on received or intercepted indications would require reading the claim apart from the specification. For example, par. [0019] (as-published) describes a Power-and-Performance (PaP) manager that “may estimate the power consumption of processors 110 based on the current processor loading from load monitor 140 and the expected computation load based on the complexity of the graphics scenes” (emphasis added). The PaP manager then “compares the estimated power consumption with a system power budget.” ([0019]). This clearly indicates that the claimed “detection” of overloading may be based on an expected load that is itself estimated based on complexity of the graphics scenes. Note that one measure of graphic scene complexity is resolution (e.g., Qin: [0073]). The specification further describes inferring overloading based on other “quality indicators” including GPU frequency, utilization rate, etc. (e.g., Fig. 2, step 210). Again, the scope of “detecting” must include inferring whether a GPU is overloaded based on observed indications, such as inferring that the GPU is overloaded based on a high-resolution (i.e., high-scene-complexity) rendering instruction. Otherwise, it would read apart from the specification.
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) 1-3, 8, 10-13, 18, and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over ‘Qin’ (US 2024/0311948 A1) in view of ‘Yeo’ (“Neural Adaptive Content-aware Internet Video Delivery,” 2018).
Regarding claim 1, Examiner notes that the claim recites a method that is substantially the same as the method performed by the system of claim 11. Qin in view of Yeo teaches the system of claim 11 (see below). Accordingly, claim 1 is also rejected under 35 U.S.C. 103 as being unpatentable over Qin in view of Yeo for substantially the same reasons as claim 11.
Regarding claim 2, Examiner notes that the claim recites a method that is substantially the same as the method performed by the system of claim 12. Qin in view of Yeo teaches the system of claim 12 (see below). Accordingly, claim 2 is also rejected under 35 U.S.C. 103 as being unpatentable over Qin in view of Yeo for substantially the same reasons as claim 12.
Regarding claim 3, Examiner notes that the claim recites a method that is substantially the same as the method performed by the system of claim 13. Qin in view of Yeo teaches the system of claim 13 (see below). Accordingly, claim 3 is also rejected under 35 U.S.C. 103 as being unpatentable over Qin in view of Yeo for substantially the same reasons as claim 13.
Regarding claim 8, Examiner notes that the claim recites a method that is substantially the same as the method performed by the system of claim 18. Qin in view of Yeo teaches the system of claim 18 (see below). Accordingly, claim 8 is also rejected under 35 U.S.C. 103 as being unpatentable over Qin in view of Yeo for substantially the same reasons as claim 18.
Regarding claim 10, Examiner notes that the claim recites a method that is substantially the same as the method performed by the system of claim 20. Qin in view of Yeo teaches the system of claim 20 (see below). Accordingly, claim 10 is also rejected under 35 U.S.C. 103 as being unpatentable over Qin in view of Yeo for substantially the same reasons as claim 20.
Regarding claim 11, Qin teaches a computing system operative to perform artificial-intelligence (AI) super-resolution (SR) (e.g., Fig. 4), comprising:
a plurality of processors including a graphics processing unit (GPU) (e.g., Fig. 4, GPU) and an AI processing unit (APU) (e.g., Fig. 4, NPU); and
a memory (e.g., Fig. 4, memory) to store a plurality of AI models (The memory of Fig. 4 is capable of performing the intended use of storing a plurality of AI models; also, e.g., [0126], model is loaded into a memory; also see Note Regarding Models below), wherein the processors are operative to:
detect that loading of the GPU exceeds a threshold (e.g., [0084], [0086]-[0087], if the required resolution exceeds a threshold – which indicates that loading of the GPU will be too high – then AI SR processing is enabled; This is a detection of an overloading state of the GPU at least because it is based on an understanding that the GPU is overloaded if instructed to render a scene with too-high a resolution – e.g., [0073]-[0074]; Examiner notes that image resolution is a measure of scene complexity at least because a scene with higher resolution has more pixels, and thus will be more complex to render – see, e.g., [0073]) when the GPU renders graphics in a video (e.g., Fig. 9 illustrates a process of using a GPU to render graphics in a video, which is performed for each frame – see especially the left and right blocks);
reduce resolution of the video output from the GPU in response to the detecting that the loading of the GPU exceeds the threshold (e.g., [0084], after learning that the rendering instruction requires a high resolution – and thus detecting the overloading – the resolution of the video output by the GPU is lowered);
after detecting that the loading of the GPU exceeds the threshold (e.g., [0121]-[0126], Fig. 10, model selection and loading at S104 occurs after detection of overloading based on resolution at S103) select an AI model (e.g., [0125]-[0126], AI SR model is selected and loaded for use in upscaling) among the plurality of AI models based on graphics scenes in the video and respective power consumption estimates of the AI models (see Note Regarding Models below), wherein the AI model is selected based on a power consumption estimate of the AI model and a power budget surplus of the computing system (see Note Regarding Models below); and
perform, by the APU, AI SR operations on the video using the selected AI model to restore the resolution of the video for display (e.g., [0084], “The NPU performs super-resolution rendering processing on the image data with the low resolution, to obtain image data with a high resolution” for display).
Note Regarding Models. Qin does not explicitly teach storing multiple AI models, that the selection of an AI model is based on graphics scenes in the video and respective power consumption estimates of the AI models, or that the AI model is selected based on a power consumption estimate of the AI model and a power budget surplus of the computing system.
However, Yeo does teach techniques for AI super-resolution that include storing a plurality of AI models (e.g., Table 2 summarizes the different AI SR models; e.g., page 651, left column, Training content-aware DNNs, a specific version of every model summarized in Table 2 is trained for each specific video; also see Sections 4.1-4.2) and selecting an AI model among the plurality of AI models based on graphics scenes in the video (e.g., Sec. 4.1, selection of AI models is limited to AI models trained for specific content – i.e., graphics scenes in the video) and respective power consumption estimates of the AI models (e.g., Sec. 5.2, Choosing a DNN from multiple options (client-side), computational power consumption is estimated for each of the AI models and one is selected for providing highest quality while fitting within available computational power budget), wherein the AI model is selected based on a power consumption estimate of the AI model (e.g., Sec. 5.2, Choosing a DNN from multiple options (client-side), computational power consumption is estimated for each of the AI models – i.e., information about each DNN is used to test their computational power consumption, with larger models requiring more power and thus taking longer to execute) and a power budget surplus of the computing system (e.g., Sec. 5.2, Choosing a DNN from multiple options (client-side), “the clients select the largest (highest-quality) DNN that runs in real-time”; I.e., the DNN with the largest power consumption estimation that still falls within a power budget surplus of the computing system is selected; As discussed at, e.g., Sec. 2, Under-utilization of client’s computation, this exploitation of surplus computational power budget is fundamental to Yeo’s technique).
Yeo recognizes that “[t]he available capacity of computing changes across time and space because of heterogeneity of client devices, changes in workloads, and multiplexing” (Sec. 4.2, 1st paragraph), but a single AI scaling model requiring a fixed amount of computing power cannot adapt to this time-varying capacity (Sec. 4.2, 1st par.). This means that using a single DNN of fixed complexity either under-utilizes the available computing capacity (compromising output image quality) or over-loads the available computing capacity, both of which are undesirable (e.g., Sec. 4.2, 1st par.). Yeo recognizes that this problem can be solved by offering multiple AI SR models with a range of options differing in computational requirements (and quality) and selecting one that best fits the available computational resources (e.g., Sec. 4.2, 2nd par.). Yeo also recognizes that selecting an AI SR model based on particular graphics scenes in video (i.e., based on content) is advantageous because it can provide better performance than more-generalized AI SR models (e.g., Sec. 4.1).
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the system of Qin with the AI model selection of Yeo in order to improve the system with the reasonable expectation that this would result in a system that better utilized available computing capacity and achieved better performance than using a single, more-generalized AI SR model. This technique for improving the system of Qin was within the ordinary ability of one of ordinary skill in the art based on the teachings of Yeo.
Therefore, it would have been obvious to one of ordinary skill in the art to combine the teachings of Qin and Yeo to obtain the invention as specified in claim 11.
Regarding claim 12, Qin in view of Yeo teaches the computing system of claim 11, and Qin further teaches that the AI model is selected such that increased system power consumption caused by the selected AI model is estimated to be less than reduced system power consumption caused by the reduced resolution (e.g., [0073]-[0074], Qin recognizes that rendering may not be able to be completed real-time a computational power budget for high-resolution images (i.e., “frame freezing during running”) and solves this problem by reducing resolution and using an NPU/APU to maintain real-time processing within an available computational power budget; Yeo selects an AI model so that its estimated power consumption is low enough to maintain real-time processing – e.g., Sec. 5.2, Choosing a DNN from multiple options (client-side); Accordingly, for a system of Qin in view of Yeo as applied above, the AI model is selected such that increased system power consumption caused by the AI model is estimated to be less than reduced system power consumption caused by the reduced resolution at least because the selected AI model can operate in real-time, while the power consumption was too great for real-time operation prior to the resolution reduction).
Regarding claim 13, Qin in view of Yeo teaches the computing system of claim 11, and Yeo further teaches that each power consumption estimate is based on a total count of nodes in a neural network represented by the AI model (e.g., Sec. 5.2, a specification of layers, channels, etc. for each AI model are used to randomly initialize a total count of all nodes in an AI model and count a total time required to execute all the nodes as a measure of computational power consumption; Also, in general, the higher the total count of nodes, the higher the inference time and therefore the higher the power consumption estimate).
Regarding claim 18, Qin in view of Yeo teaches the computing system of claim 11, and Qin further teaches that an increase in loading of the GPU is detected from an increase in graphics scene complexity in the video (e.g., [0086], GPU loading is detected as having increased, thereby triggering use of AI SR, based on an increase past a threshold of resolution in the video; e.g., [0073], higher-resolution video is more complex for a GPU to render, therefore increasing its loading).
Regarding claim 20, Qin in view of Yeo teaches the computing system of claim 11, and Qin further teaches that the APU is operative to perform the AI SR operations according to a whitelist that specifies a configuration of a plurality of functions used in rendering a plurality of graphics scenes in the video (e.g., [0083], [0128]).
Claim(s) 4 and 14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Qin in view of Yeo as applied above, and further in view of ‘Maghazeh’ (“Perception-Aware Power Management for Mobile Games via Dynamic Resolution Scaling,” 2015).
Regarding claim 4, Examiner notes that the claim recites a method that is substantially the same as the method performed by the system of claim 14. Qin in view of Yeo and Maghazeh teaches the system of claim 14 (see below). Accordingly, claim 4 is also rejected under 35 U.S.C. 103 as being unpatentable over Qin in view of Yeo and Maghazeh for substantially the same reasons as claim 14.
Regarding claim 14, Qin in view of Yeo teaches the computing system of claim 11.
Qin does not discuss frame rate (i.e., FPS).
Yeo selects specific AI SR models in order to maintain the frame rate (i.e., FPS) of the video (e.g., Table 6, Sec. 7.4, Heterogeneous clients, video FPS is maintained at 30, with AI SR models being selected to provide highest quality while maintaining at least that FPS), thereby providing power saving (see various discussion in Qin and Yeo mapped above).
Yeo does not explicitly teach increasing the FPS of the video without exceeding a power budget of the computing system when performance is prioritized over power saving.
However, Maghazeh does teach that, when system performance is prioritized over power saving, the FPS of video may be increased (Sec. IV.B, Frame rate subsection at pages 615-616, GPU can run faster, thereby increasing frame rate/FPS) without exceeding a power budget of the computing system (Sec. IV.B, Frame rate subsection at pages 615-616, GPU can deliver higher frame rates up to the power budget/limit of the system; In the described example, power use is increased to 4.7 watts at 90 FPS, which is within the power budget/limit of the system).
There is a fundamental tradeoff between resource (e.g., power) usage and performance that is recognized by each of Qin, Yeo, and Maghazeh. Maghazeh recognizes that “frame rate is an important parameter for user experience” so it may be preferable to prioritize performance as indicated by FPS (i.e., frame rate) over power consumption (Sec. IV.B, Frame rate subsection at pages 615-616, 2nd par.) in order to improve user experience.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the system of Qin in view of Yeo as applied above with the FPS increase taught by Maghazeh in order to improve the system with the reasonable expectation that this would result in a system that advantageously improved user experience. This technique for improving the system of Qin in view of Yeo was within the ordinary ability of one of ordinary skill in the art based on the teachings of Maghazeh.
Therefore, it would have been obvious to one of ordinary skill in the art to combine the teachings of Qin, Yeo, and Maghazeh to obtain the invention as specified in claim 14.
Claim(s) 5, 7, 15, and 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Qin in view of Yeo as applied above, and further in view of ‘Park’ (US 2015/0181117 A1).
Regarding claim 5, Examiner notes that the claim recites a method that is substantially the same as the method performed by the system of claim 15. Qin in view of Yeo and Park teaches the system of claim 15 (see below). Accordingly, claim 5 is also rejected under 35 U.S.C. 103 as being unpatentable over Qin in view of Yeo and Park for substantially the same reasons as claim 15.
Regarding claim 7, Examiner notes that the claim recites a method that is substantially the same as the method performed by the system of claim 17. Qin in view of Yeo and Park teaches the system of claim 17 (see below). Accordingly, claim 7 is also rejected under 35 U.S.C. 103 as being unpatentable over Qin in view of Yeo and Park for substantially the same reasons as claim 17.
Regarding claim 15, Qin in view of Yeo teaches the computing system of claim 11.
Qin is focused on avoiding high temperatures and power consumption (e.g., [0073]-[0074]), but does not explicitly teach measuring temperature. As discussed above with respect to claim 11, Qin also does not explicitly teach multiple AI SR models.
Yeo detects computing power consumption of the processors in the computing system (e.g., Sec. 5.2, last par., power consumption in terms of computing time is measured) and replaces the selected AI model with a different one of the AI models for the AI SR operations such that the power consumption stays within a power budget (e.g., Sec. 7.4, Temporal variation, Fig. 16, different models (each with a different number of optional blocks) are selected such that power consumption stays within a power budget available within a device, such as shown by the Ideal line in Fig. 16).
Nevertheless, Yeo does not teach detecting a temperature of processors or considering the temperature in AI model selection.
However, Park does teach detecting not only power consumption (e.g., Fig. 7, step 142, P) but also temperature (e.g., Fig. 7, step 142, T) of processors in a computing system. Park further teaches selecting an image processing parameter (e.g., Fig. 7, steps 146 and 152, parameter S) such that the power consumption stays within a power budget at the detected temperature (e.g., Fig. 7, the parameter S is being set to keep power P within its budget, including at the detected temperature). Park teaches that one such parameter that may be adjusted is image resolution (e.g., Figs. 5A-B, [0098]-[0102]), with lower image resolution being associated with both lower power consumption and lower quality (e.g., Figs. 5A-B, [0098]-[0102]).
It is important to note that Yeo associates different AI SR models with different image resolutions (e.g., Table 2). Accordingly, in a system of Park in view of Yeo and further modified with the temperature-based resolution adjustment of Park, temperature-based changes to the image resolution will result in changes to the selected AI model.
Qin recognizes that it is desirable to avoid excessive computing power and temperature (e.g., [0073]-[0074]). The temperature-based resolution adjustment of Park helps avoid excessive computing power and temperature, while advantageously maximizing user experience (e.g., [0111]-[0112]).
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the system of Qin in view of Yeo as applied above with the temperature-based resolution adjustment of Park in order to improve the system with the reasonable expectation that this would result in a system that advantageously avoided excessive power consumption or temperature while maximizing user experience. This technique for improving the system of Qin in view of Yeo was within the ordinary ability of one of ordinary skill in the art based on the teachings of Park.
Therefore, it would have been obvious to one of ordinary skill in the art to combine the teachings of Qin, Yeo, and Park to obtain the invention as specified in claim 15.
Regarding claim 17, Qin in view of Yeo teaches the computing system of claim 11.
Qin teaches different scenarios where AI SR operations on the video are deactivated (e.g., [0125], deactivation due to application resolution being low or lack of NPU; e.g., [0083], deactivation due to absence from whitelist). Accordingly, one of the system parameters of Qin is whether or not AI SR operations are activated or deactivated. Qin further teaches that AI-based SR operations may be selected depending on resolution (e.g., [0077], NPU is used for specific resolutions).
Nevertheless, Qin does not teach detecting temperature and power consumption of processors in the computing system, with the AI SR deactivation occurring when the power consumption reaches or exceeds a power budget at the detected temperature. Yeo also does not teach these features.
However, Park does teach temperature sensors to detect a temperature (e.g., Fig. 7, step 142, temperature T) and power consumption (e.g., Fig. 7, step 142, power consumption P) of in processors the computing system, wherein the AI SR operations on the video are deactivated (e.g., Fig. 7, steps 146-148, parameter S is adjusted; As explained above, one of the parameters in Qin in view of Yeo is whether or not AI SR is activated/deactivated; Additionally, Park adjusts resolution – e.g., Figs. 5A-B and [0098]-[0102] – as one of its parameters, which affects whether AI SR is performed on Qin’s NPU as explained above) when the power consumption reaches or exceeds a power budget at the detected temperature (e.g., Fig. 7, YES path from step 144).
Qin recognizes that it is desirable to avoid excessive computing power and temperature (e.g., [0073]-[0074]). The temperature-based resolution adjustment of Park helps avoid excessive computing power and temperature, while advantageously maximizing user experience (e.g., [0111]-[0112]).
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the system of Qin in view of Yeo as applied above with the temperature-based processing adjustment of Park in order to improve the system with the reasonable expectation that this would result in a system that advantageously avoided excessive power consumption or temperature while maximizing user experience. This technique for improving the system of Qin in view of Yeo was within the ordinary ability of one of ordinary skill in the art based on the teachings of Park.
Therefore, it would have been obvious to one of ordinary skill in the art to combine the teachings of Qin, Yeo, and Park to obtain the invention as specified in claim 17.
Claim(s) 9 and 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Qin in view of Yeo as applied above, and further in view of ‘Wang’ (US 2017/0116951 A1).
Regarding claim 9, Examiner notes that the claim recites a method that is substantially the same as the method performed by the system of claim 19. Qin in view of Yeo and Wang teaches the system of claim 19 (see below). Accordingly, claim 9 is also rejected under 35 U.S.C. 103 as being unpatentable over Qin in view of Yeo and Wang for substantially the same reasons as claim 19.
Regarding claim 19, Qin in view of Yeo teaches the computing system of claim 11.
Qin recognizes that higher resolutions will cause over-loading of a GPU (e.g., [0073]), and so it uses resolution as an indication of the loading of the GPU (e.g., [0086]).
Qin does not explicitly teach that the loading of the GPU is detected from one or more of: an operating frequency of the GPU, a utilization rate of the GPU, and unstable frame per second (FPS) of the video.
Yeo also does not teach this feature.
However, Wang does teach that an indication of the loading of the GPU is detected from one or more of: an operating frequency of the GPU, a utilization rate of the GPU, and unstable frame per second (FPS) of the video (e.g., [0003], [0037], unstable frame rate – i.e., FPS – of video is indicative of high GPU loading).
Resolution is not the only factor that can affect GPU loading. As recognized by Wang, complexity of a rendered signal (e.g., a game) can also cause a GPU to become overloaded beyond its capacity, resulting in an unstable frame rate (i.e., FPS) (e.g., [0003]). It is a goal of Qin to avoid such overloading (e.g., [0073]-[0074]).
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the system of Qin in view of Yeo as applied above with the frame-rate-based GPU loading detection of Wang in order to improve the system with the reasonable expectation that this would result in a system that could advantageously detect and mitigate high GPU loading caused by factors other than resolution, such as complexity. This technique for improving the system of Qin in view of Yeo was within the ordinary ability of one of ordinary skill in the art based on the teachings of Wang.
Therefore, it would have been obvious to one of ordinary skill in the art to combine the teachings of Qin, Yeo, and Wang to obtain the invention as specified in claim 19.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to GEOFFREY E SUMMERS whose telephone number is (571)272-9915. The examiner can normally be reached Monday-Friday, 7:00 AM to 3:30 PM 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 at (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.
/GEOFFREY E SUMMERS/Examiner, Art Unit 2669
Prosecution Insights