Detailed Action
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
The Office Action is in response to claims filed on 12/17/2024 where claims 1-12, 14-19, and 21-22 are pending and ready for examination.
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 (i.e., changing from AIA to pre-AIA ) 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.
Applicant's arguments filed 8/4/2025 have been reviewed in their entirety. The Examiner has withdrawn the 35 USC 112(b) rejection.
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, 4-6, 8-10, 12, 15-17, 19, and 21-22 are rejected under 35 USC 103 as being unpatentable over Lee (US 20210306560) in view of Zhu (US 20220360973)
Regarding claim 1, Lee discloses a method, comprising:
receiving model feedback information from an access network device, wherein the model feedback information indicates a download mode of a first artificial intelligence model that is configured for a terminal device, and the download mode of the first artificial intelligence model that is configured for the terminal device is supported by the terminal device (Lee; Lee teaches AI model formats, including ONNX, and dynamic distribution/loading of AI models to target nodes. Inv iew of that format-based distributed loading, one of ordinary skill in the a art would be readily able to facilitate the exchanged feedback control information as identify or indicating the download mode used of the AI model configured for the receiving terminal/node; Because the reference distributes AI models to target nodes at different layers and configures nodes based on feedback/control data, one of the ordinary skill in the art would provide for having the selected model format/download mode must be compatible with the receiving’s node’s supported runtime/platform/layer configuration. It would have been a predictable implementation to select and indicated a download more supported by the receiving terminal/node so that the AI model can be loaded and used at that node;
This exchange is a technical prerequisite to delivering a usable AI model in Lee’s distributed environment because the receiving node must receive a model representation compatible with its runtime/platform/layer/configuration. Accordingly, the feedback/control information used for configuration identifies the supported model format/download mode for the receiving terminal/node)
[0091] Some features of distributed AI include the following:
[0092] Integration of AI model formats (including ONNX) for distributing AI models to nodes at different layers
[0098] Layered and distributed AI with dynamic loading of AI models)
[0099] Using control data to configure a meshed network of processing by defining the feedback of the output of nodes to configure other nodes in the hierarchy
[0100] Dynamically loadable AI model at each layer
A person of ordinary skill in the art would have recognized the need to exchange model-format, model-version, and compatibility information between the network side distributing component and the receiving terminal/client component before model delivery, so that the receiving device can validate and synchronize the correct supported model representation.
Moreover, because Lee’s distributed AI system expressly integrates AI models formats for distributing AI models to target nodes, the selected model format is the operative download mode used to make the model loadable and executable at the receiving node. Thus, under BRI, Lees’ format-based distribution/loading teaches indicating a download mode for the AI model, rather than merely exchanging training feedback or model-error information.); and
downloading the first artificial intelligence model in the download mode of the first artificial intelligence model that is configured for the terminal device (Lee teaches downloading/loading the AI model in the indicated download mode because it expressly discloses dynamic distribution/loading of AI models among nodes and integration of AI model formats, including ONNX. Thus, the distributed/loading operation provides the AI model in the selected format/download mode for use by the receiving node/terminal;
[0091] Some features of distributed AI include the following:
[0092] Integration of AI model formats (including ONNX) for distributing AI models to nodes at different layers
[0098] Layered and distributed AI with dynamic loading of AI models)
[0099] Using control data to configure a meshed network of processing by defining the feedback of the output of nodes to configure other nodes in the hierarchy
[0100] Dynamically loadable AI model at each layer;
The loading/distribution step therefore completes the same format-selection sequence identified above; the model is not merely transmitted generically, but is provided in the selected AI model format that permits the receiving node/terminal to load and use the model.).
As evidence of the rationale above Zhu discloses:
receiving model feedback information from an access network device, wherein the model feedback information indicates a download mode of a first artificial intelligence model that is configured for a terminal device, and the download mode of the first artificial intelligence model that is configured for the terminal device is supported by the terminal device (Zhu; Zhu teaches UE capability signaling between the UE and the BS/CN in Fig. 6, and [0075] teaches that the UE capability information includes supported model formats, including ONNX and TensorFlow, indicated via a supported format list, as well as supported models based on a UE-tested and cached model list. Under BRI, the supported model format/list is the claimed download mode because it identifies the format/mode in which the AI/model is supported for terminal-device use. Zhu further teaches that the download mode is supported by the terminal device because the format/model information is expressly UE capability information. Zhu therefore confirms that, in the terminal-device context, model-format support is treated as UE capability information, and that the supported format list identifies the download mode in which the AI/ML model can be provided to that UE. One of ordinary skill in the art wuld be able to discern that that the access network side to use that reported capability information when selecting and indicating the compatible model format for delivery to the terminal device.
[0075] fifth capability parameter may correspond to supported model formats. For instance, supported model formats may include open neural network exchange (ONNX) and/or tensor flow (TF). The supported model formats may be indicated via a supported format list. ML models may also be compressed. Thus, whether the supported model formats include model compression may be further indicated (e.g., Yes/No) via the capability parameter. A sixth capability parameter may correspond to supported models (e.g., based on a UE tested and cached model list). While an ML model may be indicative of a function F(X), a same function F may be supported by different UEs using different models. Thus, the capability parameter may indicate the ML models that are supported by the different UEs.
PNG
media_image1.png
509
799
media_image1.png
Greyscale
PNG
media_image2.png
572
765
media_image2.png
Greyscale
)
downloading the first artificial intelligence model in the download mode of the first artificial intelligence model that is configured for the terminal device (Zhu teaches downloading/providing the first AI model in the indicated download mode because Zhu’s UE capability information identifies the model formats supported by the UE, including ONNX and TensorFlow, and identifies supported models based on a UE-tested/ached model list. Since the capability information is exchanged for the UE’s AI/ML operation, the model provided to the UE is necessarily provided in one of the supported model formats indicated by that capability information. Thus, under BRI, Zhu teaches downloading the AI model in the download mode configured for and supported by the terminal device. Because Zhu’s capability exchange identifies both supported model formats and supported models for the UE, providing the model to the UE necessarily occurs using one of the supported formats identified by that capability information. Thus the model delivery is tied to the UE-supported download mode, not to an unrelated training-feedback process)
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices. The combination applies Lee’s known format-based AI model distribution/loading to Zhu’s UE/access-network capability signaling environment so that the access network device selects and provides an AI model in a format supported by the terminal device. This yields the predictable benefit of avoiding delivery of an incompatible AI model representation and improving successful provisioning of AI models to heterogeneous devices.
Regarding claim 2, Lee in view of Zhu disclose the method according to claim 1, wherein the method further comprises:
sending model capability information to the access network device, wherein the model capability information indicates a download mode of an artificial intelligence model that is supported by the terminal device (The combined solution per Zhu (see e.g. Fig. 6)).
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices.
Regarding claim 4. Lee inv view of Zhu disclose the method according to claim 2, wherein the download mode indicated by the model capability information comprises at least one of the following:
a download mode 1 of downloading an artificial intelligence model in a specified format (The combined solution with respect to claim1, Lee teaches downloading the first AI model in a download mode corresponding to a model format, including ONNX. Therefore, the claimed “download mode 1 of downloading an AI intelligence model in a specified format” is taught because the specified format is the format model used for the download mode already identified in claim 1.
Zhu further confirms the same format -based mapping because Zhu ([0075]) teaches supported model formats including ONNX and Tensor Flow indicated via a supported format list).;
a download mode 2 of downloading computer code that describes an artificial intelligence model; or
a download mode 3 of downloading an artificial intelligence model in an intermediate format, wherein the artificial intelligence model in the intermediate format can be converted into an artificial intelligence model in a format supported by the terminal device.
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices.
Regarding claim 5. Lee in view of Zhu disclose the method according to claim 4, wherein the specified format is comprised in M formats, and M is an integer greater than or equal to 1(The combined solution as Lee teaches that distributed AI includes integration of AI model formats including ONNX, for distributing AI models to nodes at different layers and dynamically loading AI models at each Layer. Thus, Lee teaches that the download mode indicated by the model capability information corresponds to a format-based mode for providing the AI model to the terminal/node. Lee further renders obvious that the model capability information may identify a plurality of such format based download mode to an Mth download mode, because a distributed AI system that loads AI models at different nodes/layers must identify which model format is supported by the receiving terminal/node before providing the model. Each indexed download mode therefore indicates that the terminal device supports the same AI model in a corresponding ith format, where i ranges from 1 to M).
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices.
Regarding claim 6, Lee in view of Zhu disclose The method according to claim 4, wherein the model capability information further indicates parameter information corresponding to the download mode 1, wherein the parameter information corresponding to the download mode 1 comprises at least one of model framework format information, or operator package information; the model framework format information indicates at least one of a name of a model framework corresponding to the artificial intelligence model, a version number of the model framework, or a model file format of the artificial intelligence model; and the operator package information indicates an operator unit set of the artificial intelligence model (The combined solution as the parameter information corresponding to download mode 1 because the supported model format, including ONNX/TensorFlow, is a model file format of the artificial intelligence model)
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices.
Regarding claim 8, Lee in view of Zhu disclose The method according to claim 4, wherein the model capability information further indicates parameter information corresponding to the download mode 3, wherein the parameter information corresponding to the download mode 3 comprises at least one of a format name of the intermediate format or a version number of the intermediate format (The combined invention provides for ONNX per independent claim 1 which by definition is an intermediate exchange format)
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices.
Regarding claim 9. Lee in view of Zhu disclose the method according to claim 1, wherein the method further comprises: sending model requirement information to the access network device, wherein the model requirement information is used to trigger downloading of the first artificial intelligence model (The combined solution provides sending model requirement information to the access network device because Zhu teaches the UE sending AI/ML capability information, including supported model format/model information, to the network device. The model requirement information is used to trigger downloading of the first AI model because, once the network receives the UE -supported format/model indication, the network provides/downloads the corresponding AI model in the qualified supported download modes, consistent with Lee’s distributed AI model loading).
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices.
Regarding claim 10, The method according to claim 9, wherein the model requirement information further indicates at least one of the following information: a function type of the first artificial intelligence model; application scenario information of the first artificial intelligence model; information about time at which the first artificial intelligence model is requested to be used; validity time information of the model requirement information; or a stored artificial intelligence mode (The combined solution as Zhu ([0075]) teaches supported models based on a UE tested and cached model list. A cached model is stored model under BRI, thereby satisfying the claimed stored artificial intelligence model alternative)
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices.
Regarding claim 12, claim 12 comprises the same and/or similar subject matter as claim2 and is considered an obvious variation; therefore it is rejected under the same rationale.
Regarding claim 15, claim 15 comprises the same and/or similar subject matter as claim 4 and is considered an obvious variation; therefore it is rejected under the same rationale.
Regarding claim 16, Lee in view of Zhu discloses The method according to claim 15, wherein the specified format is comprised in M formats, and M is an integer greater than or equal to 1(The combined solution as Lee teaches that distributed AI includes integration of AI model formats including ONNX, for distributing AI models to nodes at different layers and dynamically loading AI models at each Layer. Thus, Lee teaches that the download mode indicated by the model capability information corresponds to a format-based mode for providing the AI model to the terminal/node. Lee further renders obvious that the model capability information may identify a plurality of such format based download mode to an Mth download mode, because a distributed AI system that loads AI models at different nodes/layers must identify which model format is supported by the receiving terminal/node before providing the model. Each indexed download mode therefore indicates that the terminal device supports the same AI model in a corresponding ith format, where i ranges from 1 to M).
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices. The combination applies Lee’s known format-based AI model distribution/loading to Zhu’s UE/access-network capability signaling environment so that the access network device selects and provides an AI model in a format supported by the terminal device. This yields the predictable benefit of avoiding delivery of an incompatible AI model representation and improving successful provisioning of AI models to heterogeneous devices.
Regarding claim 17, claim 17 comprises the same and/or similar subject matter as claim 6 and is considered an obvious variation; therefore it is rejected under the same rationale.
Regarding claim 19, claim 19 comprises the same and/or similar subject matter as claim 8 and is considered an obvious variation; therefore it is rejected under the same rationale.
Regarding claim 21, claim 21 comprises the same and/or similar subject matter as claim 2 and is considered an obvious variation; therefore it is rejected under the same rationale.
Regarding claim 22, claim 22 comprises the same and/or similar subject matter as claim 9 and is considered an obvious variation; therefore it is rejected under the same rationale.
Claims 3 and 14 are rejected under 35 USC 103 as being unpatentable over Lee in view of Zhu and in further view of Acharya (US 11,494,171)
Regarding claim 3, Lee in view of Zhu disclose the method according to claim 2, wherein the download mode indicated by the model capability information comprises at least one of a first download mode to an Mth download mode, wherein an ith download mode indicates that the terminal device supports an artificial intelligence model in an ith format, a value of i ranges from 1 to M, and M is an integer greater than or equal to 1 (The combined solution as Lee teaches that distributed AI includes integration of AI model formats including ONNX, for distributing AI models to nodes at different layers and dynamically loading AI models at each Layer. Thus, Lee teaches that the download mode indicated by the model capability information corresponds to a format-based mode for providing the AI model to the terminal/node. Lee further renders obvious that the model capability information may identify a plurality of such format based download mode to an Mth download mode, because a distributed AI system that loads AI models at different nodes/layers must identify which model format is supported by the receiving terminal/node before providing the model. Each indexed download mode therefore indicates that the terminal device supports the same AI model in a corresponding ith format, where i ranges from 1 to M)
As evidence of the rationale above Acharya discloses:
AI model formats (Acharya; Acharya teaches that every uploaded AI model undergoes auto testing and validation, including converting the received AI model into optimized compressed, mobile-friendly, and web-friendly formats that can run on destination platforms such as computer devices, cloud servers, web browsers, and smartphones, and validating each converted format for the destination platform while discarding incompatible models. This confirms that a single AI model may be represented in multiple destination-specific formats and that each format corresponds to a supported download mode for a particular terminal device platform;
See e.g. Column 12, Lines 40 - 50 “According to the embodiments of the present subject matter, every AI model being uploaded to the network undergoes the auto-testing and validation process. This includes configuring a runtime compatibility testing module 516 for converting the received at least one AI model into an optimized, compressed, mobile friendly, web friendly format that can run on any computer device including cloud-servers, web-browsers, smartphones etcetera. The converted format is then loaded in a custom simulator that validates the model for each of these destination platforms. Incompatible models are discarded.”);
Therefore it would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Acharya’ scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning devices.
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices. The combination applies Lee’s known format-based AI model distribution/loading to Zhu’s UE/access-network capability signaling environment so that the access network device selects and provides an AI model in a format supported by the terminal device. This yields the predictable benefit of avoiding delivery of an incompatible AI model representation and improving successful provisioning of AI models to heterogeneous devices.
Regarding claim 14, claim 14 comprises the same and/or similar subject matter as claim 3 and is considered an obvious variation; therefore it connected under the same rationale.
Claim 11 is rejected under 35 USC as being unpatentable over Lee in view of Zhu and in further view of Buttner (US 20220067588)
Regarding claim 11, Lee in view of Zhu disclose the method according to claim 1, Lee does not expressly disclose wherein the model feedback information further indicates at least one of: effective time information of the first artificial intelligence model; a model download license for downloading the first artificial intelligence model; a function type of the first artificial intelligence model; or application scenario information of the first artificial intelligence model.
However in analogous art Buttner discloses:
application scenario information of the first artificial intelligence model (Buttner;
see e.g. [0092] In an embodiment, the system enables lay users to transform their pre-trained AI model into a trustworthy AI model within a detachable step, including the option to use the system anytime, for example flexibly after a retraining phase of the trust with the calibrated AI model, which has for example be necessary due to changed applications or scenarios the AI model is used for.).
Therefore it would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Buttner’s app/scenario scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies of provisioning AI models)
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices. The combination applies Lee’s known format-based AI model distribution/loading to Zhu’s UE/access-network capability signaling environment so that the access network device selects and provides an AI model in a format supported by the terminal device. This yields the predictable benefit of avoiding delivery of an incompatible AI model representation and improving successful provisioning of AI models to heterogeneous devices.
Claims 7 and 18 is rejected under 35 USC 103 as being unpatentable over Lee in view of Zhu and in further view of Hammond (US 20170213155)
Regarding claim 18, Lee in view of Zhu disclose the method according to claim 15, wherein the model capability information further indicates parameter information corresponding to the download mode 2, wherein the parameter information corresponding to the download mode 2 comprises at least one of a name of a programming language of the computer code that describes the artificial intelligence model or a version number of the programming language.
Hammond discloses:
a name of a programming language of the computer code that describes the artificial intelligence model (Hammond;
see e.g. Abstract
see e.g. [0022] “... 1) scripted software code written in a pedagogical software programming language, such as Inkling™ ...”
see e.g. [0035] “... . Additional searchable criteria relative to the AI object include metadata about the specific algorithms and parameters used to train that AI object, data input type fed into the AI object, the type(s) of algorithm chosen to train the AI object, the architecture of the network of processing nodes in the AI object, ...”)
Therefore it would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Hammond’s programming language and metadata describing an AI model. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies of provisioning AI models.
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices. The combination applies Lee’s known format-based AI model distribution/loading to Zhu’s UE/access-network capability signaling environment so that the access network device selects and provides an AI model in a format supported by the terminal device. This yields the predictable benefit of avoiding delivery of an incompatible AI model representation and improving successful provisioning of AI models to heterogeneous devices.
Regarding claim 7, claim 7 comprises the same and/or similar subject matter as claim 18 and is considered an obvious variation; therefore it connected under the same rationale.
Claims 7 and 18 is rejected under 35 USC 103 as being unpatentable over Lee in view of Zhu and in further view of Acar (US 20150066467)
Regarding claim 18, Lee in view of Zhu disclose the method according to claim 15, wherein the model capability information further indicates parameter information corresponding to the download mode 2, wherein the parameter information corresponding to the download mode 2 comprises at least one of a name of a programming language of the computer code that describes the artificial intelligence model or a version number of the programming language.
Acar discloses:
a version number of the programming language. (Acar;
[0059] “In one embodiment, the software toolkit is in a Tool Command Language (e.g., TCL/TK version 8.6/8.5, 2013) programming language having SVM-Model capabilities (SVM-train and SVM-predict ...” )
Therefore it would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Acar’s software metadata The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies of provisioning AI models.
Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zhu’s scheme. The motivation being the combined solution provides for implementing a known technique resulting in increased efficiencies in provisioning AI model formats at terminal devices. The combination applies Lee’s known format-based AI model distribution/loading to Zhu’s UE/access-network capability signaling environment so that the access network device selects and provides an AI model in a format supported by the terminal device. This yields the predictable benefit of avoiding delivery of an incompatible AI model representation and improving successful provisioning of AI models to heterogeneous devices.
Regarding claim 7, claim 7 comprises the same and/or similar subject matter as claim 18 and is considered an obvious variation; therefore it connected under the same rationale.
Any inquiry concerning this communication or earlier communications from the Examiner should be directed to TODD L. BARKER whose telephone number is (571) 270 0257. The Examiner can normally be reached on Monday through Friday, 7:30am to 5:00pm.
If attempts to reach the Examiner by telephone are unsuccessful, the Examiner's supervisor Vivek Srivastava can be reached on (571) 272 7304.
/TODD L BARKER/Primary Examiner, Art Unit 2449
Prosecution Insights