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 .
Claims 1-3, 5-6, 8, 10-18, 20-23, and 25-28 are pending in this application. Claims 1, 5, 6, 8, 10, 12, 18, 20, and 25-26 are amended, claims 4, 7, 9, 19, and 24 are canceled, and claims 27-28 are new by applicant’s amendment filed 16 July 2026.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 10-17 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding Claim 10, it recites “The method of claim 1, wherein a layout of the plurality of partitions that are repeated within the layer is predefined.” However, as amended, claim 1 recites “two or more partitions,” not “a plurality of partitions.” Claim 1 also does not recite partitions that are repeated. This limitation is introduced in claim 8, but claim 10, as amended, depends on claim 1, not on claim 8. So, the term “the plurality of partitions that are repeated” is indefinite and unclear. For the purposes of examination under prior art, the examiner will interpret the terms of the present claim in the same manner as similar terms in claim 8.
Regarding Claim 12, it recites “the plurality of partitions,” which is indefinite in the same manner as in claim 10.
Regarding Claims 11 and 13-17, they are rejected as being dependent on rejected base claims.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1, 8, 10-11, 18, 20, and 25-27 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Xiong et al. (U.S. 2024/0169463, hereinafter “Xiong”).
Regarding Claim 1, Xiong teaches a method (fig. 10A; ¶ [0066]), comprising:
at a device (fig. 1; Abstract and ¶ [0044]):
processing an input, through a plurality of layers of a neural network, to predict a data value for the input (fig. 5; ¶ [0057]—an input is processed through a plurality of layers including first MoE {Mixture of Experts} layer and second MoE layer),
wherein a layer of the plurality of layers of the neural network includes a set of parameters that are partitioned into two or more partitions each comprised of a different subset of parameters in the set of parameters, and wherein a partition of the two or more partitions is dynamically selected according to the input for processing the input through the layer (fig. 5; ¶ [0057]—each MoE layer is partitioned into multiple Expert Sub-Models 40, each comprised of a different subset of parameters of the layer. A number k {one or more} of the partitions are dynamically selected according to the input for processing the input through the layer. Fig. 2 and ¶ [0049] – [0050] further describe the partitioning of parameter values 41 across the Expert Sub-Models 40 and the gating function that dynamically selects the partitions according to the input); and
outputting the data value (figs. 2 and 5; ¶ [0050] and [0057]—an output tensor is output to a next layer or as an output of the neural network).
Regarding Claim 8, Xiong teaches wherein, the two or more partitions include a plurality of partitions that are repeated within the layer at a defined frequency (figs. 2 and 5; ¶ [0049] and [0057]—the partitions include a plurality of partitions {Expert Sub-Models 40}. The number of partitions is defined, so it can be considered as being repeated at a defined frequency).
Regarding Claim 10, Xiong teaches wherein a layout of plurality of partitions that are repeated within the layer is predefined (fig. 2; ¶ [0049]—the configuration and ordering of the partitions and their connection to the gating function {i.e. the layout} is predefined).
Regarding Claim 11, Xiong teaches wherein the layout includes a random order (fig. 2; ¶ [0049]—the partitions are ordered in an arbitrary manner. They may be arranged in any order, so the layout can be considered to include a random order).
Regarding Claim 18, Xiong teaches wherein each of the two or more partitions is configured to process a corresponding range of inputs (fig. 2; ¶ [0048] – [0050]—each partition is configured to receive and process a corresponding range of input tokens).
Regarding Claim 20, Xiong teaches wherein, for the layer of the neural network, only the selected partition is active for processing the input (¶ [0049]—only the MoEs {partitions} selected by the gating function are selected. Unselected expert sub-models do not receive input tokens, so they are not active).
Regarding Claim 25, Xiong teaches a system (fig. 1; ¶ [0044]), comprising:
a non-transitory memory storage comprising instructions (¶ [0177] – [0178]); and
one or more processors in communication with the memory, wherein the one or more processors execute the instructions (¶ [0177] – [0178]) to:
process an input, through a plurality of layers of a neural network, to predict a data value for the input (fig. 5; ¶ [0057]—an input is processed through a plurality of layers including first MoE {Mixture of Experts} layer and second MoE layer),
wherein a layer of the plurality of layers of the neural network includes a set of parameters that are partitioned into two or more partitions each comprised of a different subset of parameters in the set of parameters, and wherein a partition of the two or more partitions is dynamically selected according to the input for processing the input through the layer (fig. 5; ¶ [0057]—each MoE layer is partitioned into multiple Expert Sub-Models 40, each comprised of a different subset of parameters of the layer. A number k {one or more} of the partitions are dynamically selected according to the input for processing the input through the layer. Fig. 2 and ¶ [0049] – [0050] further describe the partitioning of parameter values 41 across the Expert Sub-Models 40 and the gating function that dynamically selects the partitions according to the input); and
output the data value (figs. 2 and 5; ¶ [0050] and [0057]—an output tensor is output to a next layer or as an output of the neural network).
Regarding Claim 26, Xiong teaches a non-transitory computer-readable media storing computer instructions which when executed by one or more processors of a device (¶ [0177] – [0178]). Xiong teaches the instructions cause the device to perform the operations of the present claim in the same manner as for claim 25, above.
Regarding Claim 27, Xiong teaches wherein the two or more partitions include at least:
a first partition comprised of a first subset of parameters in the set of parameters (fig. 5; ¶ [0057]—First MoE Layer 32A comprised of a first subset of parameters in Expert Sub-Models 40. Fig. 2 and ¶ [0049] – [0050] further describe the partitioning of parameter values 41 across the Expert Sub-Models 4), and
a second partition comprised of a second subset of parameters in the set of parameters (fig. 5; ¶ [0057]—Second MoE Layer 32B comprised of a second subset of parameters in Expert Sub-Models 40. Fig. 2 and ¶ [0049] – [0050] further describe the partitioning of parameter values 41 across the Expert Sub-Models 4),
wherein at least a portion of the first subset of parameters differs from at least a portion of the second subset of parameters (¶ [0042]—different MoE models process different inputs with different sets of parameters).
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 2-3, 13-17, and 21-23 are rejected under 35 U.S.C. 103 as being unpatentable over Xiong, as applied to claims 1 and 10, above, in view of Odena et al. (U.S. 2019/0236438, hereinafter “Odena”).
Regarding Claim 2, Xiong does not specifically teach wherein the input includes a coordinate position. However, Odena teaches an input that includes a coordinate position (¶ [0022]—an input may be an image, which necessarily includes pixels arranged at coordinate positions of the image. ¶ [0047] further describes spatial positions within an image).
All of the claimed elements were known in Xiong and Odena and could have been combined by known methods with no change in their respective functions. It therefore would have been obvious to a person of ordinary skill in the art at the time of filing of the applicant’s invention to combine the coordinate positions of Odena with the input of Xiong to yield the predictable result of wherein the input includes a coordinate position. One would be motivated to make this combination for the purpose of improving the usefulness of the neural network by applying it to any kind of data input (Odena, ¶ [0021]).
Regarding Claim 3, Xiong/Odena teaches wherein the input further includes at least one additional parameter value (Odena, fig. 1, Usage input 152; ¶ [0031] – [0035]—the input includes a Usage Input {additional parameter value} in addition to the Network Input 102. The Usage input is used to select subnetworks within each partition {layer of the present claims}, as described in ¶ [0049]).
Regarding Claim 13, Xiong/Odena teaches wherein the layout is predefined for a task to be performed using the neural network (Odena, ¶ [0020] and [0043]—the neural network is a task neural network that is configured in a layout predefined for a task to be performed).
Regarding Claim 14, Xiong/Odena teaches applying the neural network to a variety of different tasks using a variety of different input types (Odena, ¶ [0021] – [0028]). Although it does not explicitly state wherein the task is an image generation task, image generation is an obvious task that is well-known in the art for neural networks.
Regarding Claim 15, Xiong/Odena teaches applying the neural network to a variety of different tasks using a variety of different input types (Odena, ¶ [0021] – [0028]). Although it does not explicitly state wherein the task is a novel-view synthesis task, view synthesis is an obvious task that is well-known in the art for neural networks.
Regarding Claim 16, Xiong/Odena teaches applying the neural network to a variety of different tasks using a variety of different input types (Odena, ¶ [0021] – [0028]). Although it does not explicitly state wherein the task is an image fitting task, image fitting is an obvious task that is well-known in the art for neural networks.
Regarding Claim 17, Xiong/Odena teaches applying the neural network to a variety of different tasks using a variety of different input types (Odena, ¶ [0021] – [0028]). Although it does not explicitly state wherein the task is a video fitting task, video fitting is an obvious task that is well-known in the art for neural networks.
Regarding Claim 21, Xiong/Odena teaches wherein the input is a pixel position, and wherein the data value is a color at the pixel position (Odena, ¶ [0022] and [0047]—the input may be an image or a feature map derived from an image. It is well-known in the art that images are made up of pixels at spatial positions, and that pixel values are colors {e.g. RGB values} at the pixel position).
Regarding Claim 22, Xiong/Odena teaches wherein the input is processed along with a conditional input, through the plurality of layers of the neural network, to predict the data value for the input (Odena, ¶ [0049]—the input is processed along with the Usage input, which is a conditional input that controls how the network processes the input).
Regarding Claim 23, Xiong/Odena teaches wherein the conditional input is a vector derived from at least one of a text or an image (Odena, ¶ [0031] – [0034]—the Usage input specifies a set of one or more usage factors, so it can be considered a vector. The factors such as quality of network output, computational complexity, and risk sensitivity may be derived from text that specifies service requirements).
Claims 5-6 and 28 are rejected under 35 U.S.C. 103 as being unpatentable over Xiong, as applied to claims 1 and 27, above, in view of Darvish Rouhani et al. (U.S. 2023/0316042, hereinafter “Darvish”).
Regarding Claim 5, Xiong does not specifically teach wherein the parameters are weights. However, Darvish teaches parameters of a Mixture of Experts neural network that are weights (fig. 1; ¶ [0030] and [0034]).
All of the claimed elements were known in Xiong and Darvish and could have been combined by known methods with no change in their respective functions. It therefore would have been obvious to a person of ordinary skill in the art at the time of filing of the applicant’s invention to combine the weights of Darvish with the parameters of Xiong to yield the predictable result of wherein the parameters are weights. One would be motivated to make this combination for the purpose of enabling MoE models to operate more efficiently and with a smaller footprint (Darvish, ¶ [0021]).
Regarding Claim 6, Xiong/Darvish teaches wherein each subset of weights is arranged as a matrix (Darvish, ¶ [0034] and [0048]).
Regarding Claim 28, Xiong/Darvish teaches wherein parameters in the set of parameters are weights of the neural network (Darvish, fig. 1; ¶ [0030] and [0034]).
Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over Xiong in view of Odena, as applied to claim 10, above, and further in view of Li et al. (U.S. Patent 11,922,287, hereinafter “Li”).
Regarding Claim 12, Xiong/Odena does not specifically teach wherein the layout includes a smooth interpolation across the plurality of partitions within the partitioned layer. However, Li teaches a layout that includes a smooth interpolation across a plurality of partitions within a partitioned layer (fig. 1; col. 6, lines 13-34—outputs of experts {partitions} in a mixture of experts model are combined using a weighted sum, which is a smooth interpolation across the partitions).
All of the claimed elements were known in Xiong/Odena and Li and could have been combined by known methods with no change in their respective functions. It therefore would have been obvious to a person of ordinary skill in the art at the time of filing of the applicant’s invention to combine the smooth interpolation of outputs of a mixture of experts of Li with the partitions of Xiong/Odena to yield the predictable result of wherein the layout includes a smooth interpolation across the plurality of partitions within the partitioned layer. One would be motivated to make this combination for the purpose of optimizing different or conflicting objectives to improve prediction results (Li, col. 1, lines 30-51).
Response to Arguments
The amendments to the claims filed 16 July 2026 are accepted as overcoming the previous rejections under 35 U.S.C. 112(b). Note, however, the new rejections under 35 U.S.C. 112(b) of claims 10-17, which are necessitated by these amendments.
Applicant’s arguments with respect to claims 1-3, 5-6, 8, 10-18, 20-23, and 25-28 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. Although Han et al. (NPL "Dynamic neural networks: A survey") does not teach all of the limitations of independent claims 1, 25, and 26, as amended, new prior art reference Xiong teaches these limitations, as detailed above. New prior art references Odena, Darvish, and Li are relied upon to teach some of the dependent claims, several of which were previously too indefinite to examine with respect to prior art.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 nonprovisional extension fee (37 CFR 1.17(a)) 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 HAL W SCHNEE whose telephone number is (571) 270-1918. The examiner can normally be reached M-F 7:30 a.m. - 6:00 p.m.
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, Michael Huntley can be reached at 303-297-4307. 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.
/HAL SCHNEE/Primary Examiner, Art Unit 2129
Prosecution Insights