CTFR 18/697,860 CTFR 89118 DETAILED ACTION This communication is in response to the Amendments and Arguments filed on 11 February 2026. Claims 1-4, 8-10, 13-15, 19, 22-24, 27, 31-33, and 36-38 are pending and have been examined. The Applicants’ amendment and remarks have been carefully considered, but are not persuasive. Hence, this Action has been made FINAL. All previous objections and rejections directed to the Applicant’s disclosure and claims not discussed in this Office Action have been withdrawn by the Examiner. Compact Prosecution In the interest of compact prosecution, the examiner suggests the following: Incorporate the embodiment of applicant’s figure 8a into independent claims 1, 8, 13, 31 and 36. As currently written, claims 1, 8, 13, 31 and 36 most closely resemble applicant’s figure 6A. Figure 8a includes both the accuracy loss 606 and discriminative loss 804 so as to distinguish between the two elements. With respect to independent claim 31, the examiner additionally suggests incorporating description of the “indication” and how an “indication” how this would be achieved. Incorporate the embodiment of applicant’s figure 8a into independent claim 22. As currently written, claims 1, 8, 13, 31 and 36 most closely resemble applicant’s figure 7. Figure 8a includes both the accuracy loss 606 and discriminative loss 804 so as to distinguish between the two elements. Response to Amendments and Arguments With respect to the 102 rejections, the applicant submits that Babaheidarian does not disclose each and every feature of claim 22 as amended. However, the examiner does not see any change substantive change to the interpretation of claim 22. The amendments do not provide any further detail of the applicant’ invention and merely shift the original claim wording around. With respect to the 103 rejections, the examiner notes that, although amended, there is no substantive change to the interpretation of independent claim 1. The applicant has also amended claim 8 and states that it recites features not recited by claim 1. However, the only significant difference the examiner observes between claim 1 and claim 8 is the use of “accuracy loss value” in claim 1 versus “discriminative loss value” in claim 8. However, as the claims are written, these two terms are interpreted the same – i.e., evaluating how well the model distinguishes between real versus fake data and then adjusting parameters accordingly. The applicant has also amended claims 13 and 36 in a manner similar to that of claim 1 and involve no substantive change to their interpretation. With respect to claim 31, the applicant submits that the examiner has applied conclusory statements and impermissible hindsight. The examiner disputes this, noting that the claim language “an indication of whether the encoded media file was generated by the generative model” is very broad. There is no explanation as what this indication is or how it operates. A user could use as a message to be hidden, data enabling to assess this determination. (i.e., the user could provide the indication of whether the encoded media file was generated by the generative model). Therefore, the 103 rejections are maintained and shown below. Claim Rejections - 35 USC § 102 07-07-aia AIA 07-07 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 – 07-08-aia AIA (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. 07-15-aia AIA Claim(s) 22 and 24 is/are rejected under 35 U.S.C. 102 (a)(1) as being anticipated by US 20210150769, hereinafter referred to as Babaheidarian . Regarding claim 22 (currently amended) , Babaheidarian discloses a computer-implemented training method (“ In some examples, the steganography encoder and the steganalysis decoder can be train ed together to ensure that their final output has a reduced entropy ,” Babaheidarian, para [0039].), comprising: generating, using one or more processors (Babaheidarian, para [0007].), based at least in part on first dat a of a processing system , a synthetically generated media file, the first data being source data for the synthetically generated media file (“ Moreover, the generator 136 can implement a generativ e adversarial network (GAN), which can take as input the lower entropy target image from the deep preprocessing network 120 and generate the higher entropy target image as its output ,” Babaheidarian, para [0067].) based at least in part on first data using a generative model ; processing, using an interpretive model implemented by the one or more processors, content of the synthetically generated media file using an interpretive model to generate first interpreted data, the first interpreted data representing content of the synthetically generated media file (“FIG. 5 illustrates an example configuration of a neural network 500 that can be implemented by one or more components of the deep preprocessing network 120 and/or the deep postprocessing network 130, such as the steganography encoder 122, the steganalysis decoder 124, the compression encoder 126, the decompression decoder 132, the generator 136, and/or the discriminator 138. For example, the neural network 500 can be implemented by the steganography encoder 122 to generate a steganography image including a cover image with a target image embedded into the cover image, the steganalysis decoder 124 to recover the target image from the steganography image and generate a lower entropy target having minimized entropy information (min H(T)), the generator 136 to learn style information (with or without interacting with the steganography encoder 122) and transfer learned style information to the lower entropy target image from the deep preprocessing network 120 to generate a higher entropy target image, and/or the discriminator 138 to classify or label the higher entropy target image from the generator 136 as real or fake ,” Babaheidarian, para [0107]. Here, the neural network serves to encode and decode (i.e., interpret) the hidden/source data. Here, the higher entropy target image is the output /first interpreted data.) ; comparing, using the one or more processors, the first interpreted data to the first data (Babaheidarian, para [0072]. The loss function acts as a comparator – i.e., comparing the first interpreted data to the first data.) ; generating, using the one or more processors based at least in part on comparing the first interpreted data to the first data, a first accuracy loss value based at least in part on the first data and the first interpreted data (“ In some cases, the discriminator 138 can downsample (e.g., by average pooling or any other mechanism) the generated image and extract features from the downsampled image. However, in other cases, the discriminator 138 can extract the features from the generated image without downsampling the generated image. In some examples, the discriminator 138 can apply a loss function to the generated image and/or a feature map associated with the generated image and output a result of the loss function. In some examples, the loss function can be a least squares loss function ,” Babaheidarian, para [0072]. Note that block 138 depends on decoder 132. See fig. 3B. Here, first data is the source data (i.e., data to be hidden), and the first interpreted data is the decoded hidden/source data.); and modifying, using the one or more processors based at least in part on the first accuracy loss value , one or more parameters of the a generative model used to generate the synthetically generated media file (“ In some cases, the neural network 500 can adjust the weights of the nodes using a training process such as backpropagation. Backpropagation can include a forward pass , a loss function , a backward pass, and a weight update . The forward pass, loss function, backward pass, and parameter update is performed for one training iteration. The process can be repeated for a certain number of iterations for each set of training data (e.g., image data) until the weights of the layers 502, 504, 506 in the neural network 500 are accurately tuned ,” Babaheidarian, para [0115]. And, “The neural network 500 can include any other deep network, such as an autoencoder (e.g., a variable autoencoder, etc.), a deep belief nets (DBNs), a recurrent neural networks (RNNs), a residual network (ResNet), a GAN, a steganography encoder network, a steganalysis decoder network , among others,” Babaheidarian, para [0119].) based at least in part on the first accuracy loss value . Claims 25-26 cancelled . Claim Rejections - 35 USC § 103 07-20-aia AIA 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. 07-21-aia AIA Claim (s) 1-3, 8, 13-14, 31-33, and 36-37 is/are rejected under 35 U.S.C. 103 as being unpatentable over “Hide and Speak: Deep Neural Networks for Speech Steganography”, hereinafter referred to as Kreuk et al., in view of “Time is on my side: Steganography in filesystem metadata”, hereinafter referred to as Neuner et al . Regarding claim 1 (currently amended) , Kreuk et al. discloses a computer-implemented training method (Kreuk et al., fig. 1 caption – During training the reconstruction loss is applied between c and m to c^ and m^, respectively.), comprising: generating, using one or more processors (“ The Encoder Network E, gets as input a carrier c ∈ X*, and a message m ∈ X*, and outputs a joint latent representation of both carrier and message, h = E (c, m). There are numerous ways for generating such representation ,” Kreuk et al., section 4, before section 4.1. See also, Kreuk et al., fig. 1.) based at least in part on first data of a processing system , a synthetically generated media file, the first data being source data for the synthetically generated media file (Kreuk et al., E c (c).) ; based at least in part on first data using a generative model; encoding, using a steganography encoder implemented by the one or more processors (Kreuk et al., D c. ), second data (Kreuk et al., m.) into the synthetically generated media file using a steganography encoder to generate an encoded media file, the second data being based at least in part on including at least a portion of the first data; processing, using a steganography decoder implemented by the one or more processors, the encoded media file using a steganography decoder to generate decoded data (Kreuk et al., retrieve ^m using D m .); comparing, using the one or more processors, the decoded data to the second data (Kreuk et al., eqn. (1). This loss function is based on a comparison between the decoded data and the second data.) ; generating, using the one or more processors based at least in part on comparing the decoded data to the second data, an accuracy loss value (Kreuk et al., eqn. (1).) based at least in part on the second data and the decoded data ; and modifying, using the one or more processors based at least in part on the accuracy loss value, at least one of one or more parameters of the steganography encoder or one or more parameters of the steganography decoder (“ Each of the above components is a neural network , where the parameters are found by minimizing the mean squared error […] between the original message and the reconstructed message,” Kreuk et al., section 4.1.) . one or both of: one or more parameters of the steganography encoder, based at least in part on the accuracy loss value; or one or more parameters of the steganography decoder, based at least in part on the accuracy loss value. Kreuk et al., though, does not disclose what the hidden message may be. Neuner et al., though, notes that audio files are good candidates to embed steganography information due to the fact that it can’t be detected from filesystem analysis (" Format containers for multimedia content (e.g., audio or video) are transparent to and independent of the underlying filesystem that hosts the multimedia file ,” fourth paragraph page 578. Neuner et al. also makes clear that the media file may be synthetically generated. See Neuner et al., sec. “Synthetic data”.) Therefore, it would be obvious for one skilled in the art to combine the teachings of Kreuk et al. with those of Neuner et al. to improve the speech steganography of Kreuk et al. by using metadata of the original file as both a problem to be solved and solution. As to claim 8 , method claim 8 is rejected on the same grounds as claim 1. As to claim 13 , method claim 13 is rejected on the same grounds as claim 1. As to claim 36 , method claim 36 is rejected on the same grounds as claim 1. Regarding claim 2 (currently amended) , Kreuk et al., as modified by Neuner et al., discloses the method of claim 1, further comprising: generating, using a discriminative model implemented by the one or more processors, based at least in part on processing the encoded media file, a discriminative loss value (Kreuk et al., first part of equation (1).) based at least in part on processing the encoded media file using a discriminative model ; and modifying, using the one or more processors based at least in part on the discriminative loss value, the one or more parameters of the steganography encoder (“ Each of the above components is a neural network , where the parameters are found by minimizing the mean squared error […] between the original message and the reconstructed message,” Kreuk et al., section 4.1.) based at least in part on the discriminative loss value . As to claim 14 , method claim 14 is rejected on the same grounds as claim 2. Regarding claim 3 (currently amended) , Kreuk et al., as modified by Neuner et al., method of claim 1, wherein the steganography encoder is a part of the a generative model used to generate the synthetically generated media file (Kreuk et al. discloses the encoder as part of the generative model used to generate the synthetically generated media file .) . As to claim 37 , method claim 14 is rejected on the same grounds as claim 3. Regarding claim 31 (currently amended) , Kreuk et al. discloses the steps of claim 31. For the present opinion, it is considered that the encoded metadata is meant to achieve determine an indication of whether the encoded media file was generated by the generative model. In order to achieve this result, the skilled person would use as a message to be hidden, data enabling to assess this determination. Regarding claim 32 , Kreuk et al. discloses outputting the indication of whether the encoded media file was generated by a generative model is performed in response to receiving an input from a user, noting that the input is either directly or indirectly received from a user. Regarding claim 33 , Kreuk et al. discloses the method of claim 31, further comprising outputting, using the one or more processors, the decoded data (as for claim 1). Claims 34-35 cancelled . 07-21-aia AIA Claim (s) 4 and 9 is/are rejected under 35 U.S.C. 103 as being unpatentable over “Hide and Speak: Deep Neural Networks for Speech Steganography”, hereinafter referred to as Kreuk et al., in view of “Time is on my side: Steganography in filesystem metadata”, hereinafter referred to as Neuner et al., and further in view of “Distribution-Preserving Steganography Based on Text-to-Speech Generative Models”, hereinafter referred to as Chen et al . Regarding claim 4 (amended) , Kreuk et al., as modified by Neuner et al., discloses the method of claim 1 3 , but not wherein : the first data is a text sequence, the generative model is a text-to- speech model. Chen et al. is cited to disclose wherein the first data is a text sequence, the generative model is a text-to-speech model (“ In this paper, we propose two distribution-preserving steganographic methods based on text-to-speech generative models, WaveGlow [40] and WaveRNN [41], ” Chen et al., p. 3, highlighted section.), and the synthetically generated media file is an audio file including synthesized speech generated by the text-to-speech model based at least in part on the text sequence (Chen et al. “ To solve this, we design stegosystems based on text-to-speech autoregressive models, WaveRNN, which can generate audio with abundant semantics ,” Chen et al., p. 6, highlighted section.). Chen et al. benefits Kreuk et al. by incorporating sampler-based distribution-preserving steganography with a high capacity and efficiency of message extraction, thereby providing an explicit distribution of the generative media. Therefore, it would be obvious to combine the teachings of Kreuk et al. with those of Chen et al. to improve the steganography quality of Kreuk et al. As to claim 9 , method claim 9 is rejected on the same grounds as claim 4. Claims 5-7 cancelled . 07-21-aia AIA Claim (s) 10 and 38 are rejected under 35 U.S.C. 103 as being unpatentable over “Hide and Speak: Deep Neural Networks for Speech Steganography”, hereinafter referred to as Kreuk et al., in view of “Time is on my side: Steganography in filesystem metadata”, hereinafter referred to as Neuner et al., and further in view of “Distribution-Preserving Steganography Based on Text-to-Speech Generative Models”, hereinafter referred to as Chen et al., and further in view of EP 3534283, hereinafter referred to as Elkind et al . Regarding claim 10 (currently amended) , Kreuk et al., as modified by Neuner et al. and Chen et al., discloses the method of claim 9, wherein encoding the second data into the synthetically generated media file using the steganography encoder to generate the encoded media file comprises includes at least one of : encoding the text sequence into the encoded synthetically generated media file (“FIG. 10 depicts an example network that can be used for classifying unordered discrete inputs. Unordered discrete inputs include tokenized text strings. Example tokenized string text include command line text and natural language text. Input in other examples can include ordered discrete inputs or other types of input. In FIG. 10, the tokenized inputs 1005 may be represented to computational layers as real-valued vectors or initial "embeddings" representing the input. These initial embeddings can be created using word2vec, one-hot encoding, feature hashing, or latent semantic analysis, or other techniques,” Elkind et al., para [0122].); encoding a tokenized version of the text sequence into the encoded synthetically generated media file (Elkind et al., para [0122].); or encoding a vector embedding based on the text sequence into the encoded synthetically generated media file (Elkind et al., para [0122].) . Elkind et al. benefits Kreuk et al. by tokenizing the input text into smaller, discrete units that can then be mapped to unique numerical identifiers or embeddings, thereby allowing the model to process and understand the textual input. Therefore, it would be obvious to combine the teachings of Kreuk et al. with those of Elkind et al. as an important step of performing text encoding in Kreuk et al. Claims 11-12 cancelled. Regarding claim 38 (currently amended) , Kreuk et al., as modified by Neuner et al., discloses the method of claim 36 37, wherein: the first data is a text sequence, the generative model is a text-to-speech model, a nd the synthetically generated media file is an audio file including synthesized speech generated by the text-to-speech model based at least in part on the text sequence , ; and encoding the second data into the synthetically generated media file using the steganography encoder to generate the encoded media file comprises includes at least one of: encoding one or more words of the text sequence into the encoded media file; encoding a tokenized version of one or more words of the text sequence into the synthetically generated encoded media file; encoding a vector embedding based on one or more words of the text sequence into the synthetically generated encoded media file; or encoding a vector into the synthetically generated encoded media file, the vector representing a classification generated by an interpretive model based on the text sequence. Chen et al. is cited to disclose that the first data is a text sequence, the generative model is a text-to-speech model (“ In this paper, we propose two distribution-preserving steganographic methods based on text-to-speech generative models, WaveGlow [40] and WaveRNN [41], ” Chen et al., p. 3, highlighted section.), and the synthetically generated media file is an audio file including synthesized speech generated by the text-to-speech model based at least in part on the text sequence (“ To solve this, we design stegosystems based on text-to-speech autoregressive models, WaveRNN, which can generate audio with abundant semantics ,” Chen et al., p. 6, highlighted section.). Chen et al. benefits Kreuk et al. by incorporating sampler-based distribution-preserving steganography with a high capacity and efficiency of message extraction, thereby providing an explicit distribution of the generative media. Therefore, it would be obvious to combine the teachings of Kreuk et al. with those of Chen et al. to improve the steganography quality of Kreuk et al. And, Elkind et al. is cited to disclose encoding the second data into the synthetically generated media file using the steganography encoder to generate the encoded media file comprises includes at least one of : encoding one or more words of the text sequence into the encoded media file (“FIG. 10 depicts an example network that can be used for classifying unordered discrete inputs. Unordered discrete inputs include tokenized text strings. Example tokenized string text include command line text and natural language text. Input in other examples can include ordered discrete inputs or other types of input. In FIG. 10, the tokenized inputs 1005 may be represented to computational layers as real-valued vectors or initial "embeddings" representing the input. These initial embeddings can be created using word2vec, one-hot encoding, feature hashing, or latent semantic analysis, or other techniques,” Elkind et al., para [0122].) ; encoding a tokenized version of one or more words of the text sequence into the synthetically generated encoded media file (Elkind et al., para [0122].) ; encoding a vector embedding based on one or more words of the text sequence into the synthetically generated encoded media file (Elkind et al., para [0122].). or encoding a vector into the synthetically generated encoded media file, the vector representing a classification generated by an interpretive model based on the text sequence. Elkind et al. benefits Kreuk et al. by tokenizing the input text into smaller, discrete units that can then be mapped to unique numerical identifiers or embeddings, thereby allowing the model to process and understand the textual input. Therefore, it would be obvious to combine the teachings of Kreuk et al. with those of Elkind et al. as an important step of performing text encoding in Kreuk et al. Claims 39-43 cancelled . 07-21-aia AIA Claim (s) 15 is/are rejected under 35 U.S.C. 103 as being unpatentable over “Hide and Speak: Deep Neural Networks for Speech Steganography”, hereinafter referred to as Kreuk et al., in view of “Time is on my side: Steganography in filesystem metadata”, hereinafter referred to as Neuner et al., further in view of US 20150074814, and further in view of US 20190189135, hereinafter referred to as Ananthabhotla et al . Regarding claim 15 (original) , Kreuk et al., as modified by Neuner et al., discloses the method of claim 13, but not wherein the media file is a synthetically generated media file generated by a generative model. Ananthabhotla et al. is cited to disclose wherein the media file is a synthetically generated media file generated by a generative model (“ FIG. 6 is a simplified block diagram of an example embodiment of a system for sending hidden data within cover audio. Shown in FIG. 6 are input data to be hidden 605, input cover audio 610, codebook waveform selection application 615, hidden data sequence generator 620, cover audio with superimposed hidden data sequence signal generator 625, transmitter 640, receiver 650, hidden data recovery application 660, output cover audio 680 and recovered hidden data 685 ,” Ananthabhotla et al., para [0045].). Ananthabhotla et al. benefits Kreuk et al. by providing hiding techniques that permit embedded data to survive the linear prediction-based speech coding protocols that are widely used to transmit audio. Therefore, it would be obvious to combine the teachings of Kreuk et al. with those of Ananthabhotla et al. to improve the steganography quality of Kreuk et al. Claims 16-18 cancelled . 07-21-aia AIA Claim (s) 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over “Hide and Speak: Deep Neural Networks for Speech Steganography”, hereinafter referred to as Kreuk et al., in view of “Time is on my side: Steganography in filesystem metadata”, hereinafter referred to as Neuner et al., in view of US 20150074814, hereinafter referred to as Joshi et al., and further in view of US 20150074814, and further in view of EP 3534283, hereinafter referred to as Elkind et al . Regarding claim 19 (currently amended) , Kreuk et al., as modified by Neuner et al., discloses the method of claim 13, but not wherein : the media file contains speech, and encoding the at least the portion of the source data the first data into the media file using the steganography encoder to generate the encoded media file comprises includes at least one of: encoding a text sequence into the encoded media file, the text sequence including a transcript or a translation of at least a portion of the speech; encoding a tokenized version of a the text sequence into the encoded media file , the text sequence including a transcript or a translation of at least a portion of the speech ; or encoding a vector embedding based on a text sequence into the encoded media file , the text sequence including a transcript or a translation of at least a portion of the speech . Joshi et al. is cited to disclose encoding a text sequence into the encoded media file, the text sequence including a transcript or a translation of at least a portion of the speech (“ The encoding and decoding are applicable to various text documents such as transcripts ,” Joshi et al., para [0073].) . Joshi et al. benefits Kreuk et al. by extending the applicability of Kreuk to text data. Therefore, it would be obvious to combine the teachings of Kreuk et al. with those of Joshi et al. to improve the steganography capabilities of Kreuk et al. Neither Kreuk et al. nor Neuner et al. nor Joshi et al. specifically disclose encoding a tokenized version of a the text sequence into the encoded media file , the text sequence including a transcript or a translation of at least a portion of the speech ; encoding a tokenized version of a the text sequence into the encoded media file , the text sequence including a transcript or a translation of at least a portion of the speech . Elkind et al. is cited to disclose encoding a tokenized version of a the text sequence into the encoded media file , the text sequence including a transcript or a translation of at least a portion of the speech (Elkind et al., para [0122].); and encoding a vector embedding based on a text sequence into the encoded media file , the text sequence including a transcript or a translation of at least a portion of the speech (Elkind et al., para [0122].). Elkind et al. benefits Kreuk et al. by tokenizing the input text into smaller, discrete units that can then be mapped to unique numerical identifiers or embeddings, thereby allowing the model to process and understand the textual input. Therefore, it would be obvious to combine the teachings of Kreuk et al. with those of Elkind et al. as an important step of performing text encoding in Kreuk et al. Claims 20-21 cancelled . 07-21-aia AIA Claim (s) 23 is/are rejected under 35 U.S.C. 103 as being unpatentable over US 20210150769, hereinafter referred to as Babaheidarian, in view of “Distribution-Preserving Steganography Based on Text-to-Speech Generative Models”, hereinafter referred to as Chen et al . Regarding claim 23 (currently amended) , Babaheidarian discloses the method of claim 22, but not wherein : the first data is a first text sequence, the generative model is a text-to-speech model, the synthetically generated media file is an audio file including synthesized speech generated by the text-to-speech model based at least in part on the first text sequence, and the interpretive model is an automatic speech recognition (ASR) model , and the first interpreted data is a second text sequence representing an interpretation of the synthesized speech by the ASR model . Chen et al. is cited to disclose wherein the first data is a first text sequence (Chen et al., p. 3, highlighted section.), the generative model is a text-to-speech model (“ In this paper, we propose two distribution-preserving steganographic methods based on text-to-speech generative models, WaveGlow [40] and WaveRNN [41], ” Chen et al., p. 3, highlighted section.), the synthetically generated media file is an audio file including synthesized speech generated by the text-to-speech model based at least in part on the first text sequence (“ To solve this, we design stegosystems based on text-to-speech autoregressive models, WaveRNN, which can generate audio with abundant semantics ,” Chen et al., p. 6, highlighted section.), and the interpretive model is an automatic speech recognition (ASR) model (“ Moreover, the semantics of speech are robust, which means that the receiver can obtain the text of speech through speech recognition ,” Chen et al., p. 2, highlighted section.) , and the first interpreted data is a second text sequence representing an interpretation of the synthesized speech by the ASR model (Chen et al., p. 2, highlighted section.). Chen et al. benefits Kreuk et al. by incorporating sampler-based distribution-preserving steganography with a high capacity and efficiency of message extraction, thereby providing an explicit distribution of the generative media. Therefore, it would be obvious to combine the teachings of Kreuk et al. with those of Chen et al. to improve the steganography quality of Kreuk et al. Regarding claim 24 (currently amended) , Babaheidarian discloses the method of claim 22, further comprising wherein : comparing the first interpreted data to the first data includes identifying, using the one or more processors, a difference between the first data and the first interpreted data (Babaheidarian, para [0072].) , and ; the method further comprises: encoding, using a steganography encoder implemented by the one or more processors, second data into the synthetically generated media file using a steganography encoder to generate an encoded media file, the second data being based at least in part on the identified difference (“ The neural network 500 further includes an output layer 506 that provides an output resulting from the processing performed by the hidden layers 504. For example, the output layer 506 can provide an encod ed or decoded image (e.g., a steganography image, a lower entropy target image, a higher entropy target image, a compressed image, a decompressed image, etc.), a discrimination result (e.g., a classification or label), a feature extraction result, etc. ,” Babaheidarian, para [0110].), the second data being based at least in part on the identified difference (“ With the initial weights, the neural network 500 may be unable to detect or learn some features or details and thus may yield poor results for some features or details. A loss function can be used to analyze error in the output. Any suitable loss function definition can be used. One example of a loss function includes a mean squared error (MSE). The MSE is defined as E.sub.total=Σ½(target−output).sup.2, which calculates the sum of one-half times the actual answer minus the predicted (output) answer squared. The loss can be set to be equal to the value of E.sub.total ,” Babaheidarian, para [0072]. Here, the second data is the error (i.e., difference) signal. And, Babaheidarian, para [0039], notes that the encoder is trained, which involves the use of the error/difference signal.); processing, using the interpretive model implemented by the one or more processors, the encoded media file using the interpretive model to generate second interpreted data (Babaheidarian, para [0107], explains that the neural network (i.e., interpretive model) may be part of the encoder.); comparing, using the one or more processors, the second interpreted data to the first data (Babaheidarian, para [0072]. The loss function acts as a comparator – i.e., comparing the second interpreted data to the second data.) ; generating, using the one or more processors based at least in part on comparing the second interpreted data to the first data, a second accuracy loss value based at least in part on the first data and the second interpreted data (It is noted that the MSE described in Babaheidarian, para [0072], is calculated using multiple difference signals. Thus, the MSE may be interpreted as a second accuracy loss value (based at least in part on the first data and the second interpreted data).); and modifying, using the one or more processors based at least in part on the first accuracy loss value and the second accuracy loss value, one or more parameters of the steganography encoder (As noted above, the MSE is used to modify (i.e., train) the weights of the neural network (i.e., interpretive model).) based at least in part on the first accuracy loss value and the second accuracy loss value . 07-21-aia AIA Claim (s) 27 is/are rejected under 35 U.S.C. 103 as being unpatentable over US 20210150769, hereinafter referred to as Babaheidarian, in view of EP 3534283, hereinafter referred to as Elkind et al . Regarding claim 27 (currently amended) , Babaheidarian discloses the method of claim 24, but not wherein: the first data is a first text sequence, the first interpreted data is a second text sequence, and the identified difference comprises one or more words or characters that differ between the first text sequence and the second text sequence; and encoding the second data into the synthetically generated media file using the steganography encoder to generate the encoded media file comprises includes at least one of : encoding the one or more words or characters into the synthetically generated encoded media file; encoding a tokenized version of the one or more words or characters into the synthetically generated encoded media file; or encoding a vector embedding based on the one or more words into the synthetically generated encoded media file. Elkind et al. is cited to disclose wherein the first data is a first text sequence (“ In some examples, the disclosed systems can analyze textual strings, symbols, and natural language expressions ,” Elkind et al., para [0037].), the first interpreted data is a second text sequence (Elkind et al., para [0035]. The “predicted output” is first interpreted data.), and the identified difference comprises one or more words or characters that differ between the first text sequence and the second text sequence (“ The output of the network system is compared with the known input, and the difference between the predicted output and the known input are used to modify the weight parameters of the network system so that the neural network system more accurately classifies its input data ,” Elkind et al., para [0035].) ; and encoding the second data into the synthetically generated media file using the steganography encoder to generate the encoded media file comprises includes at least one of : encoding the one or more words or characters into the synthetically generated encoded media file (Elkind et al., para [0122].) ; encoding a tokenized version of the one or more words or characters into the synthetically generated encoded media file (Elkind et al., para [0122].) ; or encoding a vector embedding based on the one or more words into the synthetically generated encoded media file (Elkind et al., para [0122].) . Elkind et al. benefits Babaheidarian by tokenizing the input text into smaller, discrete units that can then be mapped to unique numerical identifiers or embeddings, thereby allowing the model to process and understand the textual input. Therefore, it would be obvious to combine the teachings of Babaheidarian with those of Elkind et al. as an important step of performing text encoding in Babaheidarian. Claims 28-30 cancelled. Conclusion 07-40 AIA THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a). 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 extension fee 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 date of this final action. Any inquiry concerning this communication or earlier communications from the examiner should be directed to ANNE L THOMAS-HOMESCU whose telephone number is (571)272-0899. The examiner can normally be reached on Mon-Fri 8-6. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Bhavesh Mehta can be reached on 5712727453. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000. /ANNE L THOMAS-HOMESCU/Primary Examiner, Art Unit 2656 Application/Control Number: 18/697,860 Page 2 Art Unit: 2656 Application/Control Number: 18/697,860 Page 3 Art Unit: 2656 Application/Control Number: 18/697,860 Page 4 Art Unit: 2656 Application/Control Number: 18/697,860 Page 5 Art Unit: 2656 Application/Control Number: 18/697,860 Page 6 Art Unit: 2656 Application/Control Number: 18/697,860 Page 7 Art Unit: 2656 Application/Control Number: 18/697,860 Page 8 Art Unit: 2656 Application/Control Number: 18/697,860 Page 9 Art Unit: 2656 Application/Control Number: 18/697,860 Page 10 Art Unit: 2656 Application/Control Number: 18/697,860 Page 11 Art Unit: 2656 Application/Control Number: 18/697,860 Page 12 Art Unit: 2656 Application/Control Number: 18/697,860 Page 15 Art Unit: 2656 Application/Control Number: 18/697,860 Page 16 Art Unit: 2656 Application/Control Number: 18/697,860 Page 17 Art Unit: 2656 Application/Control Number: 18/697,860 Page 18 Art Unit: 2656 Application/Control Number: 18/697,860 Page 19 Art Unit: 2656 Application/Control Number: 18/697,860 Page 20 Art Unit: 2656
Prosecution Insights