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 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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 4/13/2026 has been entered.
Response to Arguments
Applicant's arguments filed 4/13/2026 have been fully considered but they are not persuasive.
Applicant introduced new 112 rejections with amendments, however the old 112 rejections are withdrawn.
Applicant argues, “the rejection improperly abstracts the claims…” Remarks 13. The rejection has been reworked to quote the claimed abstract idea. The argument is moot.
Applicant argues, the “claims are expressly directed to transmission and efficient neural network execution…” Remarks 15. The “efficient… execution” is an alleged improvement. There are many implementations inside of the claim scope that would make the neural network run slower, or with less accuracy. Nothing in the claims would necessarily make the neural network more efficient.
Applicant argues the claims must “be evaluated… under Ex part Desjardins… [and] the claims as a whole recite a specific technological improvement in the encoding … of neural network representations.” Remarks 16. DesJradins allowed a claimed improvement where “the specification identified improvements as to how the machine learning model itself operates…”1 The unclaimed technological improvement is alluded to in the specification paragraph 12, “It would be advantageous to have a concept at hand which renders transmission/updates of machine learning predictors… more efficient such as more efficient in terms of conservation of inference quality with reducing… a coded size of NN representations… or which enables a more frequent transmission/update of a NN than currently or which even improves the inference quality for a certain task at hand and/or for a certain local input data statistic.” The problem with the claimed invention is that the compression is “lossy”. Spec. 302. A lossy compression of a neural network is not an improvement to a neural network. As far as improvements to methods of sending neural networks, the improvement in the specification is more of a design choice – sacrificing accuracy for size/speed of transmission. Therefore, the improvement is not claimed in a way that a person would know what the inventor defines as their invention, the specification’s alleged improvement is not an improvement to the machine learning model, and the specifications improvement to model transmission is not an improvement.
Applicant argues, “claims recite a technological improvement, not a mere field-of-use limitation…” Remarks 16. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible “simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." MPEP 2016.05(h). As to the claimed “efficient execution”, this is not a technological improvement to machine learning models for the reasons stated above. With regards to “picture and/or video processing”, the claims don’t pick a field to definitively link, instead claiming “and/or”. Much more integrated examples have been found by courts to be “merely indicating a field of use…”2 Therefore, the mention of a picture or video processing is merely indicating a field of use.
Applicant states the claims require a “specific apparatus to generate a specially structured data stream…” Remarks 17. A specific apparatus is not needed to generate applicant’s data stream.
Applicant argues the claims “require a particular solution to a problem…” Remarks 18. This is not an exception to the rules for patent eligibility.
Applicant argues that the “claims reflect an improvement in computer functionality and machine learning technology…” Remarks 18. The alleged improvement to transmitting data is a design choice sacrificing accuracy for smaller transmissions.
Applicant argues, “examiner should not dismiss additional element as mere ‘generic computer components’ without considering whether those elements confer a technological improvement…” Remarks 19. The additional elements in this application are not a technological improvement, see above.
Applicant argues that the claims “require a particular stream structure, particular relationships among the stream portions and pointers, portion-specific quantization and reconstruction, and a particular arithmetic encoding behavior at portion boundaries…. That improves how machines handle encoded neural network representations.” Remarks 20. The claims trade lossy compression for large network representations, this is not an improvement. Further, the recited steps in the argument are part of the abstract idea.
Applicant’s arguments with respect to 103 rejections of claims 2-5, 21, 22, 24 and 26 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.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 2-5 and 21, 22, 24 and 26 rejected under 35 U.S.C. 101 because the claimed invention is directed to the abstract idea of a mental concept and mathematical relationship for encoding/decoding a representation of a neural network; encoding quantization indices of neural network parameters; quantizing the parameters differently for each layer of the neural network, where the different quantization is based on internal probability or arithmetic of the encoder; and structuring the data stream into separate portions. This judicial exception is not integrated into a practical application because Applicant providing and receiving a data stream is “mere data gathering”, which is insignificant extra solution activity. MPEP 2106.05(g). The claimed “efficient execution” is not realistic. If anything, encoding a neural network makes the execution of the neural network less efficient because the network has to be decoded before execution – and the decoded network will be lossy– therefore, this claim element does not integrate the abstract idea into a practical application. The additional element of picture and video processing merely link the abstract idea to picture and video processing. The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the computer and computer readable media are generic computer parts.
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.
Claims 2-5, 21, 22, 24 and 26 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as failing to set forth 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.
The term “efficient execution” in claims 2, 3, 21 and 22 is a relative term which renders the claim indefinite. The term “efficient execution” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. There is no definition in the claims or the rest of the specification, that describes how the algorithm is efficient or what the efficiency is.
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 2-5, 21, 22, 24 and 26 are rejected under 35 U.S.C. 103 as being unpatentable over US20180082181A1 to Brothers et al and US20020102026A1 to Rijavec et al.
Brothers teaches claim 2. (Currently Amended) An apparatus for encoding a representation of a neural network into a data stream for a transmission to an apparatus for neural network processing, the apparatus comprising: (Brothers fig. 4)
a processor; and
a memory coupled to the processor and storing instructions that, when executed by the processor, cause the apparatus to: (Brothers fig. 3)
encode a representation of a neural network into the data stream, (Brothers fig. 4 compress weights 407, store compressed weights 409.)
provide the data stream with, (Brother fig. 4 store compressed weights 409.)
encode quantization indices of neural network parameters, which represent a neural network, into a data stream; (Brothers fig. 4 reorder weights 405, compress the reordered weights 408, store compressed weights. Brothers para 46 “single shared Huffman table may be used for weight decoding. For a set of weight indices for a sequence of output nodes (e.g., output node 0, 1 . . . 7). There is an even distribution of weight index usage in which low indices are more common than high indices. A single Huffman table is used to exploit the higher frequency of low indices throughout the whole set of weights.”)
quantize the neural network parameters differently for different portions of the neural network based on a predetermined criterion; and (Brothers para 40 “In one embodiment, the manner in which zero values are handled depends in part on the layer type (e.g., convolutional layer vs. fully connected layer).”)
generate the data stream including, for each of the neural network portions, a reconstruction rule for dequantizing neural network parameters corresponding to the respective neural network portion and by encoding the representation of the neural network into the data stream (Brothers fig. 4 store compressed weights 409. Brothers para 40 “ the manner in which zero values are handled depends in part on the layer type (e.g., convolutional layer vs. fully connected layer).”)
wherein the structuring of the data stream(Brothers teaches that weights “will also typically be arbitrarily organized in memory… [which] impacts… execution efficiency.” Brothers solves this with reordering, fig. 4 step 405, before streaming the reordered weights to the memory in step 409. Brothers para 3 “Example applications of NNs include image processing, speech recognition…”)
Brothers doesn’t teach individually accessible portions of the data stream.
However, Rijavec teaches so that the data stream is structured into one or more individually accessible portions, each portion representing a corresponding… (Rijavec abs “a compressed data set, at least one pointer to a location in the compressed data stream…”)
for each of one or more predetermined individually accessible portions, a pointer pointing to a beginning of the respective predetermined individually accessible portion (Rijavec abs “a compressed data set, at least one pointer to a location in the compressed data stream…”)
using arithmetic encoding with newly initializing the arithmetic encoding at the beginning of each of the one or more predetermined individually accessible portions, wherein newly initializing refers to re-initializing an internal encoder state including context probability states and/or internal arithmetic encoding variables for the arithmetic encoding; (Rijavec para 10 “ at least one pointer to a location in the compressed data stream whose decoded output comprises a location on a line of data, and decoding information for each received pointer that enables decoding from a point within the compressed data stream addressed by the pointer.”)
Rijavec, the claims, and Brothers all encode compressed date onto a data stream. It would have been obvious to a person having ordinary skill in the art, at the time of filing, to use Rijavec’s portions to allow a user to decode “only a section… that a user wants access[ to].” Rijavec para 10.
Brothers teaches claims 3, 21 and 22. (Currently Amended) An apparatus for decoding a representation of a neural network from a data stream for a transmission to an apparatus for neural network processing, the apparatus comprising: (Brothers fig. 5)
a processor; and
a memory coupled to the processor and storing instructions that, when executed by the processor, cause the apparatus to: (Brothers fig. 3)
receive a data stream containing a representation of a neural network and quantization indices of neural network parameters; (Brother fig. 5 Read compressed weights 505, decompress and reorder weights 510 515.)
decode the representation of the neural network from the data stream, (Brothers fig. 5 decompress weights 510.)
decode from the data stream, (Brothers fig. 5 decompress weights 510.)
wherein the representation of the neural network is decoded from the data stream (Brothers fig. 5 decompress weights 510.)
decode the neural network parameters from the data stream, wherein the neural network parameters corresponding to different portions of the neural network have been quantized differently based on a predefined quantization scheme; (Brothers para 40 “In one embodiment, the manner in which zero values are handled depends in part on the layer type (e.g., convolutional layer vs. fully connected layer).”)
extract the data stream, for each of the neural network portions, a corresponding reconstruction rule for dequantizing neural network parameters; and (Brothers para 40 “In one embodiment, the manner in which zero values are handled depends in part on the layer type (e.g., convolutional layer vs. fully connected layer).” The rule is the different handling of different layer types.)
apply the extracted reconstruction rule to reconstruct the neural network parameters for each respective neural network portion; (Brothers para 40 “In one embodiment, the manner in which zero values are handled depends in part on the layer type (e.g., convolutional layer vs. fully connected layer).” And Brothers fig. 5 execute model 515.)
wherein the structuring of the data stream (Brothers teaches that weights “will also typically be arbitrarily organized in memory… [which] impacts… execution efficiency.” Brothers solves this with reordering, fig. 4 step 405, before streaming the reordered weights to the memory in step 409. Brothers para 3 “Example applications of NNs include image processing, speech recognition…”)
Brothers doesn’t teach individually accessible portions of the data stream.
However, Rijavec teaches so that the data stream is structured into one or more individually accessible portions, each portion representing a corresponding… (Rijavec abs “a compressed data set, at least one pointer to a location in the compressed data stream…”)
for each of one or more predetermined individually accessible portions, a pointer pointing to a beginning of the respective predetermined individually accessible portion (Rijavec abs “a compressed data set, at least one pointer to a location in the compressed data stream…”)
decoded from the data stream using arithmetic decoding with newly initializing the arithmetic decoding at the beginning of each of the one or more predetermined individually accessible portions, wherein newly initializing refers to reinitializing an internal decoder state including context probability states and/or internal arithmetic decoding variables for the arithmetic decoding; (Rijavec para 10 “ at least one pointer to a location in the compressed data stream whose decoded output comprises a location on a line of data, and decoding information for each received pointer that enables decoding from a point within the compressed data stream addressed by the pointer.”)
Rijavec, the claims, and Brothers all encode compressed date onto a data stream. It would have been obvious to a person having ordinary skill in the art, at the time of filing, to use Rijavec’s portions to allow a user to decode “only a section… that a user wants access[ to].” Rijavec para 10.
Brothers teaches claim 4. (Original) Apparatus of claim 3, wherein the neural network portions comprise neural network layers of the neural network and/or layer portions into which a predetermined neural network layer of the neural network is subdivided. (Brothers para 40 “In one embodiment, the manner in which zero values are handled depends in part on the layer type (e.g., convolutional layer vs. fully connected layer).”)
Brothers teaches claim 5. (Original) Apparatus of claim 3, wherein the apparatus is configured to decode, from the data stream, a first reconstruction rule for dequantizing neural network parameters relating to a first neural network portion, in a manner delta-decoded relative to a second reconstruction rule for dequantizing neural network parameters relating to a second neural network portion. (The reconstruction rule is different for different types of layers/portions. Brother teaches delta decoding in para 35-36 “The deltas are computed versus prediction 635. For example, the differences between adjacent columns and/or rows in a cluster may be computed. Other transformation may be applied to a “base” column or row used to make predictions for the other columns and rows. …[0036] An optional adjustment 645 of the deltas may be performed to improve compressibility and then retraining performed to mitigate accuracy loss. For example, a delta value might be adjusted up or down a small amount in order to improve compressibility.” Brothers goes on to show that the rules are delta-decoded for different layers, rows, and/or clusters, Brothers para 34 “reordering may include reordering corresponding to switching around feature maps or feature map nodes in fully-connected layers. However, the reordering may also include reordering to improve compression. The reordering may include reordering into clusters and reordering based on column and row attributes. … Additionally, rows may be reordered within a group of columns to effectively compress iteratively in the row dimension. For example, row 1 elements are predicted to be the same as row 0, plus some small positive delta and the deltas are compressed. …”)
Brothers teaches claims 24 and 26. (Currently Amended) The apparatus according to claim 2, wherein the apparatus is: an image processing device configured to execute the neural network for the image processing; or a video processing device configured to execute the neural network for the video processing; or an image and video processing device configured to execute the neural network for the image and video processing. (Brothers para 3 “Example applications of NNs include image processing, speech recognition…”)
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Austin Hicks whose telephone number is (571)270-3377. The examiner can normally be reached Monday - Thursday 8-4 PST.
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, Mariela Reyes can be reached at (571) 270-1006. 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.
/AUSTIN HICKS/Primary Examiner, Art Unit 2142
1 https://www.uspto.gov/sites/default/files/documents/memo-desjardins.pdf
2 “viii. Language specifying that the abstract idea of budgeting was to be implemented using a "communication medium" that broadly included the Internet and telephone networks, because this limitation merely limited the use of the exception to a particular technological environment, Intellectual Ventures I v. Capital One Bank, 792 F.3d 1363, 1367, 115 USPQ2d 1636, 1640 (Fed. Cir. 2015);
ix. Specifying that the abstract idea of using advertising as currency is used on the Internet, because this narrowing limitation is merely an attempt to limit the use of the abstract idea to a particular technological environment, Ultramercial, Inc. v. Hulu, LLC, 772 F.3d 709, 716, 112 USPQ2d 1750, 1755 (Fed. Cir. 2014); and
x. Requiring that the abstract idea of creating a contractual relationship that guarantees performance of a transaction (a) be performed using a computer that receives and sends information over a network, or (b) be limited to guaranteeing online transactions, because these limitations simply attempted to limit the use of the abstract idea to computer environments, buySAFE Inc. v. Google, Inc., 765 F.3d 1350, 1354, 112 USPQ2d 1093, 1095-96 (Fed. Cir. 2014).”
MPEP 2106.05(h)
Prosecution Insights