Notice of Pre-AIA or AIA Status
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
2. Applicant's arguments filed 05/14/2026 have been fully considered but they are not persuasive.
On pages 5-6 of the amendment, Applicant argued that “trained quantization” in Besenbruch is limited to a mathematical proxy used internally during the training phase to facilitate gradient descent.
However, the examiner respectfully disagrees, Besenbruch clearly teaches on paragraph 0259 that a training quantisation function is used in the forward pass steps (iii) and (iv), wherein the forward pass step (iii) discloses quantizing the latent representation to produce a quantized latent ŷ (see paragraph 0243; see also FIG. 2); and step (iv) discloses using the decoder neural network to produce an output image {circumflex over (x)} from the quantized latent, wherein the output image is an approximation of the input image (see paragraph 0244; see also FIG. 2). Therefore, Besenbruch’s training quantisation function is part of an NN encoding process applied on input image data, that in the same field of endeavor of the NN encoding process of CRICRÌ.
Claim Objections
3. Claim 7 objected to because of the following informalities: “updating the selected subset based the dequantized…” should be corrected to “updating the selected subset based on the dequantized…”. Appropriate correction is required.
Claim Rejections - 35 USC § 103
4. 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.
5. The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
6. This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
7. Claim(s) 1, 5-6, 9, 13-14 and 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over CRICRÌ et al. (US 2023/0269387) published as WO2021255605, hereinafter “CRICRÌ” in view of BESENBRUCH et al. (US 2022/0272345) hereinafter “BESENBRUCH”.
As per claim 1, CRICRÌ discloses a method for encoding an input image, the method comprising:
determining, by an encoding device, using a deep neural network based on a first model an embedding representative of the input image (paragraph 0212, Input the image to the encoding pipeline, obtaining a latent tensor; see also Fig. 8 or Fig. 10);
selecting a subset of parameters to be updated based on the input image (paragraph 0171, a set of Overfitting Parameters (OPs) are selected; see also paragraphs 0212-0216, which teach that OPs are updated based on the input image);
determining a parameter update for the selected subset of parameters (paragraphs 0212-0216 teach that OPs are updated) for fine-tuning a second neural network model based on the first model (paragraph 0152, Various embodiments described herein, propose to initially train the encoder and decoder NNs to be optimal for performing further learning or finetuning at test phase. For example, when the encoder-side device overfits some of parameters to the content that needs to be encoded), wherein the fine-tuning is based on the input image (see Fig. 8) and a decoded version of the embedding as decoded using a deep neural network based on the second neural network model (paragraphs 0152-0153, In other words, during the training phase or during a fine-tuning phase, the NNs need to ‘learn to learn to compress and reconstruct’ a content that is provided at inference phase, by leveraging the meta-learning paradigm… During meta-learning, first one or more sets of parameters to be overfitted to a content being encoded are selected. For example, the latent tensor output by the encoder and the decoder NN’s bias terms);
quantizing the parameter update based on a…quantization with one or more quantization parameters (paragraph 0182, a quantization operation, which are applied on the updates to the overfitted Decoder NN’s parameters…In case of quantization, a differentiable approximation of a quantizer may be used, such as a quantizer based on Straight-Through Estimator); and
generating encoded data comprising at least an encoded quantized embedding information representative of the selected subset of parameters and an encoded quantized parameter update (paragraph 0197 and 0199, In case the OPs were some of the Decoder NN’s parameters, the encoded latent tensor and the overfitted decoder’s parameters would be both part of the encoded data to be sent to the decoder-side, where the overfitted decoder’s parameters may comprise one or more of the following possible options: the overfitted decoder’s parameters, a compressed version of the overfitted decoder’s parameters, an update to the decoder’s parameters, a compressed version of an update to the decoder’s parameters, a cluster label, a compressed version of cluster label. Compression of the decoder’s parameters or of an update to the decoder’s parameters or of cluster labels may comprise one or more of sparsification, quantization and entropy encoding… the decoder Overfitting Parameters was compressed, such as quantized and entropy encoded; see also FIG. 8 or FIG. 10), and information representative of the one or more quantization parameters (paragraph 0111).
However, CRICRÌ does not explicitly disclose quantizing the parameters update based on a trained quantization.
In the same field of endeavor, BESENBRUCH discloses quantizing the parameters update based on a trained quantization (paragraphs 0259-0260, the quantized latent is calculated using a training quantisation function).
Therefore, it would have been obvious for one having skill in the art before the effective filing date of the claimed invention to modify the teachings of CRICRÌ in view of BESENBRUCH, by using a trained quantization, thus converging to a target rate on a training set of images reasonably guarantees that we will have the same constraint satisfied on a test set of images (BESENBRUCH, paragraphs 0259-0260)
As per claim 5, CRICRÌ discloses wherein the fine-tuning is based on a loss function to minimize a measure of a distortion between the input image and an image reconstructed using a deep neural network based on the second neural network model updated with one or more updated parameters (paragraph 0149, The optimization may consist of using the output of the encoder and decoder for computing a loss function similar to the loss function used during training, and then differentiating it with respect to the parameters to be optimized, for example, one or more of the three above mentioned options. This optimization may be referred to also as finetuning or overfitting, in various embodiments).
As per claim 6, CRICRÌ discloses wherein the selected subset of parameters is selected from among a set comprising a bias, a weight, one or more parameters of a non-linear function of a model, a subset of layers of the model, a specific layer of the model, a bias of a specific layer of the model, and a subset of neurons of the model (paragraphs 0171-0174; see also paragraphs 0145-0148).
As per claims 9 and 13-14, arguments analogous to those applied for claims 1 and 5-6 are applicable for claims 9 and 13-14; in addition, CRICRÌ discloses using an encoder for encoding an input image (see FIG. 8 or FIG. 10).
As per claim 18, arguments analogous to those applied for claim 1 are applicable for claim 18; in addition, CRICRÌ discloses using a non-transitory computer readable medium comprising program code instructions for implementing the encoding and decoding methods (paragraph 0337).
Claim Rejections - 35 USC § 102
8. 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.
9. Claim(s) 7-8, 15-16 and 19 is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by CRICRÌ et al. (US 2023/0269387).
As per claim 7, CRICRÌ discloses a method for decoding an image represented by encoded data, the method comprising:
Obtaining, by a decoding device, a decoded embedding, information representative of a selected subset of parameters of a model of a deep neural network (paragraph 0199, At decoder side, in case the OPs were some of the Decoder NN’s parameters, the decoder device would use the set or version of Overfitting Parameters which was signaled by the encoder device……If the signaling comprises a cluster label, the decoder device selects the set or version of overfitted Overfitting Parameters that corresponds to the signaled cluster label, for example by using a look-up table that maps cluster labels to sets or versions of overfitted Overfitting Parameters…If the signaling information about the decoder Overfitting Parameters was compressed, such as quantized and entropy encoded, the decoder device may need to first decompress the information, such as entropy decode and dequantize the received data), a decoded parameters update from the encoded data (paragraph 0199, At decoder side, in case the OPs were some of the Decoder NN’s parameters, the decoder device would use the set or version of Overfitting Parameters which was signaled by the encoder device. If the signaling comprises the overfitted decoder’s parameters, then these parameters are used within the Decoder NN. If the signaling comprises an update to the decoder’s parameters, then this update is applied to the Decoder NN’s parameters…If the signaling information about the decoder Overfitting Parameters was compressed, such as quantized and entropy encoded, the decoder device may need to first decompress the information, such as entropy decode and dequantize the received data), and information representative of the one or more quantization parameters (FIGs. 8 and 10; paragraphs 0111 and 0199);
selecting the selected subset of parameters based on the information (paragraph 0199, the decoder device selects the set or version of overfitted Overfitting Parameters that corresponds to the signaled cluster label, for example by using a look-up table that maps cluster labels to sets or versions of overfitted Overfitting Parameters);
dequantizing the decoded parameters update based on the information representative of the one or more quantization parameters (paragraph 0199);
updating the selected subset based the dequantized parameters update (paragraphs 0200-0204, During development or training phase, after having selected the Overfitting Parameters, the neural networks of the codec are trained by using the proposed meta-learning procedure as part of the training procedure. The proposed training procedure may be used for training a randomly initialized neural network, or for training a neural network which has been previously trained by using any suitable training method, such as a more conventional training method that does not utilize meta-learning In an embodiment, the training procedure may comprise following two nested loops:…To this end, first perform a forward operation on the neural networks, compute a loss, which can be a sum of multiple losses, as already described, then compute derivatives of the loss with respect to the selected Overfitting Parameters, and then update the OPs by using the computed derivatives. As a result of the overfitting stage, a model with updated parameters, due to the overfitting of the Overfitting Parameters, is obtained; see also paragraph 0199 regarding based the dequantized parameters); and
determining, using the deep neural network with the updated parameters (paragraph 0200-0204, During development or training phase…and then update the OPs by using the computed derivatives. As a result of the overfitting stage, a model with updated parameters, due to the overfitting of the Overfitting Parameters, is obtained), a decoded image based on the obtained decoded embedding (paragraph 0144, The phase when a codec or neural network are deployed and used is referred to as inference phase or test phase in various embodiments. Normally, at inference phase (for example, once the model is trained), the input image is passed through the encoding stages, for example, an encoder NN, a quantizer, and an arithmetic encoder, to obtain a bitstream and send it to the decoder side for decoding).
As per claim 8, arguments analogous to those applied for claim 6 are applicable for claim 8.
As per claims 15-16, arguments analogous to those applied for claims 7-8 are applicable for claims 15-16; in addition, CRICRÌ discloses using a decoder for decoding an image (see FIG. 8 or FIG. 10).
As per claim 19, arguments analogous to those applied for claim 7 are applicable for claim 19; in addition, CRICRÌ discloses using a non-transitory computer readable medium comprising program code instructions for implementing the encoding and decoding methods (paragraph 0337).
Conclusion
THIS ACTION IS MADE FINAL. 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 MOHAMMED JEBARI whose telephone number is (571)270-7945. The examiner can normally be reached Mon-Fri: 09:00am-06:00pm.
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, Chris Kelley can be reached at 571-272-7331. 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.
/MOHAMMED JEBARI/Primary Examiner, Art Unit 2482
Prosecution Insights