DETAILED ACTION
This action is in response to the amended filing on 08/11/2026. Claims 1-8 and 12-17 are pending for examination.
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 .
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1, 4-8, and 12-17 are rejected under 35 U.S.C. 103 as being unpatentable over Pan et al. (US 20210083994 A1, hereinafter Pan) in view of Uva (US 20250053273 A1, hereinafter Uva) in further view of KLAIMAN et al. (US 20210350176 A1, hereinafter Klaiman) in additional view of Eby et al. (US 20230386450 A1, hereinafter Eby)
Pan, Uva, and Klaiman were recited in the previous office action.
Regarding Claim 1, Pan teaches a method for automatically identifying outliers in a training dataset for a neural network NN (Paragraph [0165] The outlier ratio may be defined as the total number of outliers among the clusters, divided by the quantity of clusters (i.e., the count).), the method comprising:
generating, for each element of the training dataset, an embedding vector which is a numeric representation of the corresponding element; (Paragraph [0004] The training further includes generating training feature vectors from the training utterances. The training feature vectors include respective training feature vectors associated with each skill hot. The training further includes generating multiple set representations of the training feature vectors, where each set representation of the multiple set representations corresponds to a subset of the training feature vectors, and configuring the classifier model to compare input feature vectors to the multiple set representations. (Generation of a vector given a series of inputs))
storing, for each element of the training dataset, the embeddings vector with an association to the corresponding element of the training dataset; (Paragraph [0123] At block 505, the training system 350 accesses training utterances, also referred to as example utterances, for the various skill bots 116 associated with the master bot 114. For instance, the training utterances may be stored as skill bot data in the training data 354 accessible to the training system 350. Paragraph [0124] At block 510, the training system 350 generates training feature vectors from the training utterances accessed at block 505. As described above, the training utterances may include a subset associated with each skill bot 116 and, further, may include training utterances for each intent of each skill bot 116. For instance, the training system 350 may store an indication of the composite feature vectors in a storage device (e.g., on the digital assistant 106) accessible by the classifier model 324. The classifier model 324 may be configured to compare input utterances to such composite feature vectors, which act as set representations of the training feature vectors 620. (The training data (utterance) is stored, then associated with feature. The associated feature is then also stored))
marking the elements from the training set with embedding vectors having the dissimilarity score higher than or equal to a predetermined threshold value as outliers. (Paragraph [0218] In some embodiments, the input utterance 303 may be deemed unrelated to any of the available skill bots 116 if the input feature vector 1710 is far enough away (e.g., exceeding a predetermined distance deemed to be excessive) from every bot vector. However, in additional or alternative embodiments, an input utterance 303 is not deemed unrelated based on dissimilarity to bot vectors alone. For instance, if the input feature vector 1710 is dissimilar (i.e., not sufficiently similar) to all the bot vectors, this is not necessarily a dispositive indication that the input utterance 303 belongs in the none class 316. (Items deemed too dissimilar are marked as belonging to a "none class".))
removing the marked elements from the training dataset; (Paragraph [0047] If the input utterance is determined to be unrelated to any bot, then the input utterance may be deemed a “none class” and is not routed to any bot for handling. Instead, processing of the input utterance may end, or the master bot may prompt the user for clarification as to what the user intended. (The input utterance marked as "none" is not used (removed) to train))
identifying if the training dataset with removed marked elements is sufficient to train the neural network NN based on at least one sufficiency criterion; and (Paragraph [0159] In some embodiments, the training system 350 performs one, two, or more rounds of iteratively determining a set of clusters of the training feature vectors 620. Although this method 1400 utilizes two rounds of determining clusters, a single round or a greater number of rounds may be used in some embodiments. At block 1420, the training system 350 begins a first round of determining the clusters. The first round utilizes an iterative loop, in which the training system 350 generates clusters of the training feature vectors 620 and then determines the sufficiency of such clusters. As described below, when the clusters are deemed sufficient, the training system 350 may then terminate the iterations in the first round. (The system determines if the cluster set is sufficient to train the model))
Pan does not teach gaining access to the training set for the neural network NN comprising a plurality of elements;
In the same field of endeavor, Uva teaches: gaining access to the training set for the neural network NN comprising a plurality of elements; (Paragraph [0083] In some embodiments, training sets may include a variety of subject matters, such as, as nonlimiting examples, medical report documents, electronic health records, entity documents, business documents, inventory documentation, emails, user communications, advertising documents, newspaper articles, and the like. In some embodiments, training sets of an LLM may include information from one or more public or private databases. As a non-limiting example, training sets may include databases associated with an entity. In some embodiments, training sets may include portions of documents associated with audiovisual data correlated to examples of outputs. (If a training set includes information from a private database, then at some point access must be gained to use anything in there due to its private nature.)
It would have been obvious to a person having ordinary skill in the art before the effective filing date to have incorporated the concept of gaining access to a training data set for a neural network, as suggested by Uva into the system of Pan because both of these systems address the field of training data sets used by machine learning models. Doing so would improve Pan by allowing for a more efficient gathering or generating of training data used in machine learning models (Uva Paragraph [0004]).
The combination of Pan and Uva does not teach generating at least one additional element using a generative convolutional neural network trained on the remaining training dataset and adding the at least one additional element to the training dataset when the training dataset is determined to be insufficient to train the neural network NN after the elements with dissimilarity score higher than or equal to the predetermined threshold have been removed from the training dataset.
In the same field of endeavor, Klaiman teaches: generating at least one additional element using a generative convolutional neural network trained on the remaining training dataset and adding the at least one additional element to the training dataset when the training dataset is determined to be insufficient to train the neural network NN after the elements with dissimilarity score higher than or equal to the predetermined threshold have been removed from the training dataset. (Paragraph [0175] A “fully convolutional neural network” as used herein is a neural network composed of convolutional layers without any fully-connected layers or multilayer perceptrons (MLPs) usually found at the end of the network. Paragraph [0206] According to one embodiment, the received images as well as the generated tiles are multi-channel images. The number of tiles can be increased for enriching the training data set by creating modified copies of existing tiles having different sizes, magnification levels, and/or comprising some simulated artifacts and noise. In some cases, multiple bags can be created by sampling the instances in the bag repeatedly as described herein for embodiments of the invention and placing the selected instances in additional bags. This “sampling” may also have the positive effect of enriching the training data set. ())
It would have been obvious to a person having ordinary skill in the art before the effective filing date to have incorporated the concept of generating an additional element and adding it to the training dataset as suggested by Klaiman into the combination of Pan and Uva because both of these system address the field of training data sets used by machine learning models. Doing so would be beneficial and improve the combination of Pan and Uva, as easily viewing similar scores would allow professionals in a given field more efficient tools to practice their trade by specifying particular needs (Klaiman Paragraph [0005]).
The combination of Pan, Uva, and Klaiman does not teach: computing a centroid of all the embedding vectors of all the elements of the training dataset equal to an as the average of all the embedding vectors of all the elements of the training dataset; generating a dissimilarity score for each element of the training set by calculating a distance between the embedding vector corresponding to the element and the centroid;
In the same field of endeavor, Eby teaches computing a centroid of all the embedding vectors of all the elements of the training dataset equal to an as the average of all the embedding vectors of all the elements of the training dataset; (Paragraph [0047] As shown in FIG. 2, the server 106 performs a selection operation 210. During the selection operation 210, the server 106 selects an input utterance that could represent a sentence or utterance including one or more words (represented as tokens in the input utterance). In some embodiments, the input utterance is an utterance that is selected for use in Siamese network training. The server 106 also selects one or more target embedding vectors for each class among multiple classes of training utterances. From the target embedding vectors, the similarity to the input utterance will be calculated. In some embodiments, the training utterances are part of a training set of historical utterances labeled with one or more classes. Paragraph [0048] In still other embodiments, the server 106 selects an utterance as a representative target for each class, such as the utterance determined to be the example closest to either the mean or median embedding vector of that class. These selected class embedding vectors represent the “centroid” or “target” for each class and thus are referred to as the “class centroid” or “class target.” (In light of Paragraph [15] in the instant specification, a ‘training dataset’ is defined as a “training set corresponding to a single label for a neural network”. ‘Label’ and ‘Class’ are seen as synonymous for the purposes of examination.Eby teaches a training dataset for a Siamese network, which comprises one or more classes. The singular rather than plural embodiment will be used. (Paragraph [0047]) For this singular class which represents a training dataset for a model, a representative word (a centroid) is calculated from a class where each element is an embedding vector.))
generating a dissimilarity score for each element of the training set by calculating a distance between the embedding vector corresponding to the element and the centroid; and (Paragraph [0047] Also, in some embodiments, the server 106 passes each training utterance in the training set through a pre-trained language model (such as a large language model 220 described in greater detail below), groups the training utterances into clusters by class, and calculates either the mean or median embedding vector of each class. Paragraph [0051] Based on the pooling operation 230, the utterance embedding vector has been determined for the given input utterance and for all class targets. The server 106 executes a distance calculation layer 240, which determines how similar the input utterance is to each class of training utterances. If the input utterance is not similar to any class of training utterances, the input utterance can be considered to be “unhandled.” (A pre-trained model classifies elements in the training dataset as similar and places them in a class, meaning that training elements in a class are marked as similar. Furthermore, when an input utterance which serves as training data (Paragraph [0047]) is attempted to be added to the class, a distance similarity score is calculated. If deemed to be too dissimilar to the rest of the training items in the training dataset, the training utterance is marked as 'unhandled')).
It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to have incorporated the concepts of computing a centroid for a training dataset and then generating a dissimilarity score for the elements in the training dataset as taught by Eby into the combination of Pan, Uva, and Klaiman as all of these references are in the same field of training machine learning models, and doing so would prove to be advantageous by way of allowing for the increased capacity of neural network to accomplish similarity-based machine learning tasks in far more efficient ways (Eby Paragraph [0003]).
Regarding Claim 4, the combination of Pan, Uva, Klaiman, and Eby teaches all of the limitations of Claim 1, including: wherein the gaining access to the training dataset for the neural network NN comprises gaining access to one or more user devices, cluster/cloud storages, or databases which respectively store the training dataset. (Uva Paragraph [0083] In some embodiments, training sets may include a variety of subject matters, such as, as nonlimiting examples, medical report documents, electronic health records, entity documents, business documents, inventory documentation, emails, user communications, advertising documents, newspaper articles, and the like. In some embodiments, training sets of an LLM may include information from one or more public or private databases. As a non-limiting example, training sets may include databases associated with an entity. In some embodiments, training sets may include portions of documents associated with audiovisual data correlated to examples of outputs. (If a training set includes information from a private database, then at some point access must be gained to use anything in there due to its private nature.))
Regarding Claim 5, the combination of Pan, Uva, Klaiman, and Eby teaches all of the limitations of Claim 1, including: wherein storing the embedding vector further comprises storing metadata as the association indicating a relationship of each of the elements of the training dataset to the embedding vectors. (Pan paragraph [0004] In some embodiments, a system described herein includes a training system and a master bot. The training system is configured to train a classifier model. Training the classifier model includes accessing training utterances associated with skill bots, where the training utterances comprising respective training utterances associated with each skill bot of the skill bots. Each skill bot is configured to provide a dialog with a user. The training further includes generating training feature vectors from the training utterances. The training feature vectors include respective training feature vectors associated with each skill hot. Paragraph [0070] In certain embodiments, the master bot 114 is configured to be aware of the available skill bots 116. For instance, the master bot 114 may have access to metadata that identifies the various available skill bots 116, and for each skill bot 116, the capabilities of the skill bot 116 including the tasks that can be performed by the skill bot 116. (As each skill bot is defined by metadata and each skill bot is defined by its relationship to a vector, the metadata therefore indicates a relationship with the training vectors of the skill bots)).
Regarding Claim 6, the combination of Pan, Uva, Klaiman, and Eby teaches all of the limitations of Claim 1, including: displaying a list of the dissimilarity scores with visual representations of the elements of the training dataset. (Klaiman Paragraph [0051] Computing, for each of the images whose tiles were used for computing a respective similarity score, a respective similarity heat map as a function of the similarity scores, the pixel color and/or pixel intensities of the similarity heat map being indicative of the similarity of the tiles in the said image to the selected tile; and Paragraph [0052] displaying the similarity heat map. Paragraph [0261] To each of the tile pairs in the first set, e.g. pair 812, the label “similar” is assigned. To each of the tile pairs in the second set, e.g. pair 814, the label “dissimilar” is assigned. (Displaying a list of similar and dissimilar items))
Regarding Claim 7, the combination of Pan, Uva, Klaiman, and Eby teaches all of the limitations of Claim 1, including: wherein the predetermined threshold value is configurable by using a graphical user interface. (Klaiman Paragraph [0034] According to embodiments, the GUI enables the user to select whether the relevance heat map is computed based on the numerical values of the MIL or based on the weights of the attention-MLL or based on the combined score. This may allow a user to identify if the output of the MIL and of the attention MLL in respect to the predictive power of a tile is significantly different. Paragraph [0097] According to other embodiments, the tile pairs are generated by selecting at least some or all of the tiles of each received image as a starting tile; for each starting tile, selecting all or a predefined number of “nearby tiles”, wherein a “nearby tile” is a tile within a first circle centered around the starting tile, whereby the radius of this circle is identical to a first spatial proximity threshold; for each starting tile, selecting all or a predefined number of “distant tiles”, wherein a “distant tile” is a tile outside of a second circle centered around the starting tile, whereby the radius of the said circle is identical to a second spatial proximity threshold; the selection of the predefined number can be performed by randomly choosing this number of tiles within the respective image area. The first and second proximity threshold may be identical, but preferably, the second proximity threshold is larger than the first proximity threshold. For example, the first proximity threshold can be 1 mm and the second proximity threshold can be 10 mm. Then, a first set of tile pairs is selected, whereby each tile pair comprises the start tile and a nearby tile located within the first circle. Each tile pair in the first set is assigned the label “similar” tissue patterns. In addition, a second set of tile pairs is selected, whereby each pair in the said set comprises the start tile and one of the “distant tiles”. Each tile pair in the second set is assigned the label “dissimilar” tissue patterns. For example, this embodiment may be used for creating “binary” labels “similar” or “dissimilar”. (The radius of the distance between tiles can be configured through usage of a GUI. These threshold values for dissimilarity are adjustable through usage of the GUI))
Regarding Claim 8, the combination of Pan, Uva, Klaiman, and Eby teaches all of the limitations of Claim 7, including: refreshing a list of the elements and the corresponding dissimilarity scores displayed in response to the a user changing, on the graphical user interface, the predetermined threshold value (Klaiman Paragraph [0046] The feature that the tiles in the report gallery are selectable and a selection triggers the performing of a similarity search for identifying and displaying other tiles having a similar feature vector/tissue pattern as the user-selected tile may enable a user to freely select any image tile in the report tile gallery he or she is interested in... The display and the GUI can be refreshed automatically after the similarity search has completed. Paragraph [0047] According to some embodiments, the computation and display of the similarity search gallery comprises the computation and display of a similarity heat map. The heat map encodes similar tiles and respective feature vectors in colors and/or in pixel intensities. Image regions and tiles having similar feature vectors are represented in the heat map with similar colors and/or high or low pixel intensities. Hence, a user can quickly get an overview of the distribution of particular tissue pattern signatures in a whole slide image. The heat map can easily be refreshed simply by selecting a different tile, because the selection automatically induces a re-computation of the feature vector similarities based on the feature vector of the newly selected tile. (As established, the search for distant or near tiles requires selecting a proximity threshold. Therefore, as selecting a new tile automatically refreshes the computed similarity, the limitation is met.))
Regarding Claims 12-17, they are system claims that correspond to the method claims 1 and 4-8 above. Therefore, they are rejected for the same reasons as method claims 1 and 4-8 above.
Claims 2-3 are rejected under 35 U.S.C. 103 as being unpatentable over Pan in view of Uva in further view of Klaiman and in additional view of Eby, as applied in Claim 1 above, and in further view of BYDLON (EP 4154810 A1, hereinafter Bydlon).
Regarding Claim 2, the combination of Pan, Uva, Klaiman, and Eby teaches the invention as claimed in Claim 1 above. The combination of Pan, Uva, Klaiman, and Eby fails to disclose the dissimilarity score for each element is calculated using a neural network trained using a metric learning method.
Bydlon was referenced in the prior office action.
In the same field of endeavor, Bydlon teaches: the dissimilarity score for each element is calculated using the neural network trained using a distance metric learning method. (Paragraph [0176] In embodiments, the auto-labeler ALL is operable to interface with the transformer stage TS' to obtain the latent representation of some or each frame or interval of frames (windows) s of the labelled data of the prototype set. For example, a trained encoder g.sub.e may be used, and the auto-labeler ALL computes a distance measure between some or all frames/windows within a known label, and some or all encoder generated representations of the unlabeled training data. One suitable distance measure may be the cosine similarity, or squared Euclidean distance (L2 norm), or other. (Network g calculates distance, which is synonymous with similarity. Network g is metric learning.))
It would have been obvious to a person having ordinary skill in the art before the effective filing date to have incorporated the concept of calculating a dissimilarity score using a neural network trained in a metric learning method as suggested by Bydlon into the combination of Pan, Uva, Klaiman, and Eby because all of these systems address the field of training data sets used by machine learning models. Doing so would improve the combination of Pan and Uva by including the advantages of efficient machine learning model training and applying it to a non-medical field (Bydlon Paragraph [0008]).
Regarding Claim 3, the combination of Pan, Uva, Klaiman, Eby, and Bydlon teaches the invention as claimed in Claim 2 including: wherein the distance metric learning method implements at least one of a Center Loss, Triplet Loss, Contrastive Loss, Softmax Loss, A-Softmax Loss, Large Margin Cosine Loss (LMCL), or Arcface Loss methods. (Bydlon Paragraph [0163] Network g may be trained in an iterative process, in which predicted (training output) labels ĉ.sub.t are compared to the ground-truth labels c.sub.i using a cost function F, such as cross-entropy loss, log loss, etc. Training continues in multiple iterations until convergence is achieved, i.e., predictions match the ground-truth (within a pre-define margin), where c.sub.i, ĉ.sub.t ∈ {1.. K}, and K denotes the number of classes. (Cross entropy loss falls under the same umbrella as softmax loss. For softmax loss you need a softmax activation layer and a cross entropy loss function both in operation. Therefore, it is part of the same structure))
Response to Arguments
Applicant’s arguments filed on 08/11/2026 on the remark (page 9) with respect to objected claims have been fully considered and are considered persuasive due to the amended filing.
Applicant’s arguments filed 08/11/2026, pages 9-10, with respect to claims 12 and 19-20 under 35 U.S.C 112(f) rejection have been fully considered and are persuasive due to amendments filed or cancellation of claims. Therefore, the interpretation of claims 12 and 19-20 under 35 U.S.C 112(f) is withdrawn.
Applicant’s arguments filed 08/11/2026, pages 10-13, with respect to claims 1-8, and 12-17 under 35 U.S.C 101 rejection have been fully considered and are persuasive. Therefore, the rejections of claims 1-8, and 12-17 under 35 U.S.C 101 is withdrawn.
Applicant’s arguments filed on 08/11/2026 on the remark (pages 14-16) with respect to claim 1-8 and 12-17 under 35 U.S.C 103 have been considered but are moot because the new ground of rejection (see rejection above).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Wu et al. (US 20220026519 A1) methods and systems for movement tracking, including vectors.
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 JUSTIN A CARDOSO whose telephone number is (571)272-8512. The examiner can normally be reached M-F 7:30 - 5:00, alternate Friday's off.
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, Jennifer Welch can be reached at (571) 272-7212. 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.
/JUSTIN CARDOSO/
Patent Examiner, Art Unit 2143
/JENNIFER N WELCH/Supervisory Patent Examiner, Art Unit 2143
