DETAILED ACTION
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 § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor, or for pre-AIA the applicant regards as the invention.
Claims 1, 11 and 20 recites the limitation "the positive sample in the feature space" in line 7, 11 and 9, respectively. There is insufficient antecedent basis for this limitation in the claims.
Claims 1, 11 and 20 recites the limitation "the positive sample" in line 11, 16 and 13, respectively. There is insufficient antecedent basis for this limitation in the claims. It is not clear whether the recited “positive sample” has for antecedent basis “a positive sample from the training dataset” or “a positive sample in the feature space”.
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 of this title, 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.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied 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.
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.
Claims 1, 2, 11, 12 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over David Hoffmann et al. [US 20230025169 A1] in view of Bogdan Georgescu et al. [US 20160174902 A1].
Regarding claim 1, David teaches:
1. A method for training a neural network based model (i.e. A method for training an encoder- Abstract…a method is provided for training an encoder that maps data samples x of measurement data onto representations z that can be evaluated by machine- ¶0007), comprising:
receiving, via a communication interface, a training dataset comprising a plurality of training samples (i.e. a set of training samples is provided- Abstract…In step 110, a set X of training samples x is provided; here, in the context of a specified application a relation is defined concerning the degree to which two samples x.sub.1 and x.sub.2 are similar to one another- ¶0056);
encoding (i.e. A further important use case of the encoder ƒθ (x) is the classification of data samples, such as images- ¶0037), by the neural network based model, at least a subset of the plurality of training samples into representations in a feature (i.e. representations z- ¶0057, fig. 1A) space(i.e. a function ƒθ (x), parameterized with trainable parameters θ, is provided that maps samples x onto representations z- ¶0057);
determining, for a given query, a positive sample from the training dataset (i.e. for this query sample q a set P, ordered in a ranked order, of positive samples p from the set X that are similar to the query sample q- ¶0061) based on a relationship between the given query and the positive sample in the feature space (i.e. In step 140, a similarity measure h(x1, x2) is provided that assigns to samples x1 and x2 a similarity of the depictions gλ (ƒθ (x1)) and gλ (ƒθ (x2)) in the working space- ¶0059);
selecting, for the given query, one or more negative samples from the training dataset (i.e. a set N of negative samples n from the set X that are no longer similar to the query sample q- ¶0061);
computing a loss (i.e. a cost function L) based on the positive sample and the one or more negative samples (i.e. the positive samples pi ∈ Pi of the respective rank level i are evaluated as positive samples, the positive samples pi ∈ Pi for rank levels j<i remain not taken into account, and [0068] the positive samples pi ∈ Pi for rank levels j. i are evaluated as negative samples, can be selected as contribution Li to cost function L- ¶0065-0068); and
training the neural network based model based on the loss (i.e. Curve a was obtained after ƒθ (x) was trained with the method described above, the cost function L having been assembled from contributions Li,out- ¶0099-0102).
However, David does not teach explicitly:
neural network; that are within a reconfigurable distance to the positive sample in the feature space.
In the same field of endeavor, Bogdan teaches:
neural network (i.e. FIG. 2 illustrates an exemplary auto-encoder neural network- ¶0008… According to an advantageous implementation, complex image patterns can be encoded in hierarchical features by learning one or more hidden layers by stacking deep neural network architectures, as described above- ¶0048); that are within a reconfigurable distance to the positive sample in the feature space (i.e. For example, in the case in which the first marginal search space is the position of the target anatomical objects, image patches centered at the ground truth center position of the target anatomical object in each annotated training image are selected as positive training samples and one or more random image patches located more than a predetermined distance (e.g., 5 mm) from the ground truth center position of the target anatomical object are randomly selected from each annotated training image as the negative training samples- ¶0041).
It would have been obvious to one with ordinary skill in the art before the effective filing date of the claimed invention, to modify the teachings of David with the teachings of Bogdan to improve the speed of applying the marginal space deep neural network architecture over the input hypotheses (Bogdan- ¶0086).
Regarding claim 2, David and Bogdan teach all the limitations of claim 1.
However, David does not teach explicitly:
wherein the determining, selecting, computing and training are iteratively performed across a plurality of iterations.
In the same field of endeavor, Bogdan teaches:
wherein the determining, selecting, computing and training (i.e. FIG. 16 illustrates an algorithm for injecting sparsity into a deep neural network using iterative re-weighted L1 norm regularization according to an embodiment of the present invention. As illustrated in FIG. 16, at 1602, the iteration count l is set to zero, the re-weighting coefficients are initialized to one for all weights of all of the filters. At 1604, the weights w(l) of are estimated using a gradient backpropagation step based on the adapted cost function expressed in Equation (8)) are iteratively performed across a plurality of iterations (i.e. The steps of refining the model using the trained boundary detector and projecting the refined model to the learned shape space can be iterated until convergence or for a predetermined number of iterations- ¶0066… A challenge arising with the use of deep neural networks as the discriminating engine in each stage of the marginal space pipeline is the high class imbalance. This imbalance can reach ratios of 1:1000 positive to negative samples, impacting both the training efficiency and stochastic sampling of the gradient during learning, resulting in a bias of the classifier towards the overrepresented negative class. A deep neural network architecture cannot be trained with an SGD approach on such an unbalanced set, and simply re-weighting the penalties for the network cost function further worsens the vanishing gradient effect. According to an advantageous embodiment of the present invention, a negative filtering cascade of shallow (i.e., one hidden layer) neural networks to filter negative responses. FIG. 19 illustrates a negative filtering cascade used to balance a training set according to an embodiment of the present invention. As shown in FIG. 19, in each stage (1902, 1904, and 1906) of the cascade 1900, a shallow, sparse neural network is trained and its decision boundary is adaptively tuned to eliminate as many true negative hypotheses from the training set as possible. The remaining hypotheses, classified as positives, are propagated to the next stage of the cascade where the same filtering procedure is repeated unless a balanced sample set is achieved. In order to train a network within the cascade, we iterate at epoch level over the complete positive training set, while at each batch level, we randomly sample the negative training space to obtain a balanced training batch- ¶0103).
It would have been obvious to one with ordinary skill in the art before the effective filing date of the claimed invention, to modify the teachings of David with the teachings of Bogdan to improve the speed of applying the marginal space deep neural network architecture over the input hypotheses (Bogdan- ¶0086).
Regarding claim 11, apparatus claim 11 is drawn to the apparatus using/performing the same method as claimed in claim 1. Therefore, apparatus claim 11 corresponds to method claim 1, and is rejected for the same reasons of obviousness as used above.
Regarding claim 12, apparatus claim 12 is drawn to the apparatus using/performing the same method as claimed in claim 2. Therefore, apparatus claim 12 corresponds to method claim 2, and is rejected for the same reasons of obviousness as used above.
Regarding claim 20, computer-readable medium storing instructions claim 20 corresponds to the same method as claimed in claim 1, and therefore is also rejected for the same reasons of obviousness as listed above.
Claims 3, 7, 13 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over David Hoffmann et al. [US 20230025169 A1] in view of Bogdan Georgescu et al. [US 20160174902 A1] and further in view of Xiao Qin et al. [US 20220245460 A1].
Regarding claim 3, David and Bogdan teach all the limitations of claim 2.
However, David and Bogdan do not teach explicitly:
decreasing the reconfigurable distance across a first portion of the plurality of iterations; and increasing the reconfigurable distance across a second portion of the plurality of iterations.
In the same field of endeavor, Xiao teaches:
decreasing the reconfigurable distance across a first portion of the plurality of iterations; and increasing the reconfigurable distance across a second portion of the plurality of iterations (i.e. The margin μ is also introduce as a hyperparameter to further control the hardness—the higher the μ the easier the case- ¶0065… With reference to the pseudocode, the training process starts by first initializing the GNN model. The process involves multiple training iterations (epoch). In line 3, the code generates the representation of each node. In line 4, in early on (epoch<threshold), the training involves random negative sampling which aims to let the model quickly learn the easy cases. In line 6, the ASA negative sampler will be triggered aiming to feed more challenging samples for the GNN to learn, The ASA sampler uses the model G (trained from the last iteration), node embedding E, margin (c) and a decay option. If the decay option is enabled, the margin will control the hardness and make the samples goes from easy to hard as the training progress. In line 8, given the positive sample and negative samples, measure how much mistake GNN makes on these samples. In line 9, the code computes and decides how the GNN should be updated (backprop). And, in line 10 the model is updated. Lines 8-10 are a stomatitis gradient decent process- ¶0071).
It would have been obvious to one with ordinary skill in the art before the effective filing date of the claimed invention, to modify the teachings of David and Bogdan with the teachings of Xiao to make use of the positive relationship during the evaluation to control a hardness accordingly (Xiao- ¶0047).
Regarding claim 7, David, Bogdan and Xiao teach all the limitations of claim 3.
However, David and Bogdan do not teach explicitly:
wherein the increasing is at a constant rate across the first portion of the plurality of iterations.
In the same field of endeavor, Xiao teaches:
wherein the increasing is at a constant rate across the first portion of the plurality of iterations (i.e. Different decay functions, (e.g., exponential and linear decay function for μ) are used to increase the hardness as the training progresses.). It would have been obvious to one with ordinary skill in the art before the effective filing date of the claimed invention, to modify the teachings of David and Bogdan with the teachings of Xio to make use of the positive relationship during the evaluation to control a hardness accordingly (Xiao- ¶0047).
Regarding claim 13, apparatus claim 13 is drawn to the apparatus using/performing the same method as claimed in claim 3. Therefore, apparatus claim 13 corresponds to method claim 3, and is rejected for the same reasons of obviousness as used above.
Regarding claim 17, apparatus claim 17 is drawn to the apparatus using/performing the same method as claimed in claim 7. Therefore, apparatus claim 17 corresponds to method claim 7, and is rejected for the same reasons of obviousness as used above.
Claims 8 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over David Hoffmann et al. [US 20230025169 A1] in view of Bogdan Georgescu et al. [US 20160174902 A1] and further in view of Vidya Rajagopal et al. [US 20210064826 A1].
Regarding claim 8, David and Bogdan teach all the limitations of claim 1.
However, David and Bogdan do not teach explicitly:
wherein the neural network based model is a recommender model that is trained to generate a recommended item for an intelligent agent who is conducting a multi-turn conversation with a user.
In the same field of endeavor, Vidya teaches:
wherein the neural network based model is a recommender model that is trained to generate a recommended item for an intelligent agent who is conducting a multi-turn conversation with a user (i.e. The present disclosure describes a system for an AI based virtual agent trainer, which may include a portion or all components as shown in FIG. 3. The system 300 may be an AI based virtual agent trainer. The system for the AI based virtual agent trainer 300 may include an utterance generator 310, a conversation builder 320, a conversation simulator 330, a conversation analyzer 340, a maturity scorer and recommender 350… The AI based virtual agent trainer 300 may include one or more AI machine learning networks, including but not limited to, K-means, term frequency-inverse document frequency (TF-IDF), random forest, deep neural network (DNN) classifier, sequence to sequence (Seq2Seq) model, recurrent neural network (RNN), and linear regression. The AI based virtual agent trainer 300 may include one or more voice processing network, for example but not limited to, a text-to-speech (TTS) generator/simulator and a speech-to-text (STT) generator/simulator- ¶0054-0056).
It would have been obvious to one with ordinary skill in the art before the effective filing date of the claimed invention, to modify the teachings of David and Bogdan with the teachings of Vidya to improve accuracy and authenticity of the virtual agents (Vidya- ¶0005).
Regarding claim 18, apparatus claim 18 is drawn to the apparatus using/performing the same method as claimed in claim 8. Therefore, apparatus claim 18 corresponds to method claim 8, and is rejected for the same reasons of obviousness as used above.
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over David Hoffmann et al. [US 20230025169 A1] in view of Bogdan Georgescu et al. [US 20160174902 A1] and further in view of Prashant Iyengar et al. [US 20240177524 A1].
Regarding claim 10, David and Bogdan teach all the limitations of claim 1.
However, David and Bogdan do not teach explicitly:
wherein the given query and the positive sample belong to a pair of a user utterance and a corresponding agent response in a prior conversation.
In the same field of endeavor, Prashant teaches:
wherein the given query and the positive sample belong to a pair of a user utterance and a corresponding agent response in a prior conversation (i.e. the processor is configured to detect pairs of dialogue adjacency between the user and the agent to identify the exact meaning of words in the conversation- ¶0007… the processor is configured to combine at least one answer by the agent from the created record of the set of questions and responses based on the pairs of dialogue adjacency when at least one new question is asked by the user- ¶0008).
It would have been obvious to one with ordinary skill in the art before the effective filing date of the claimed invention, to modify the teachings of David and Bogdan with the teachings of Prashant to develop an AI that enables the fastest user interaction which in turn improves users conversational experience (Prashant- ¶0003).
Claims 10 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over David Hoffmann et al. [US 20230025169 A1] in view of Bogdan Georgescu et al. [US 20160174902 A1] further in view of Xiao Qin et al. [US 20220245460 A1] and even further in view of Prashant Iyengar et al. [US 20240177524 A1].
Regarding claim 10, David, Bogdan and Xiao teach all the limitations of claim 3.
However, David, Bogdan and Xiao do not teach explicitly:
further comprising: receiving, from a user interface, a user utterance; and encoding, by the trained neural network based model, the user utterance into an utterance representation; and generating, by a decoder head, a recommended response based on the utterance representation.
In the same field of endeavor, Prashant teaches:
further comprising: receiving, from a user interface (i.e. Input/output (I/O) devices (including but not limited to keyboards, displays, pointing devices, etc.) can be coupled to the system either directly or through intervening I/O controllers. Network adapters may also be coupled to the system to enable the data processing system to become coupled to other data processing systems or remote printers or storage devices through intervening private or public networks. Modems, cable modem and Ethernet cards are just a few of the currently available types of network adapters- ¶), a user utterance (i.e. the processor is configured to combine at least one answer by the agent from the created record of the set of questions and responses based on the pairs of dialogue adjacency when at least one new question is asked by the user- ¶0008); and encoding, by the trained neural network based model, the user utterance into an utterance representation; and generating, by a decoder head, a recommended response based on the utterance representation (i.e. In one aspect, a processor-implemented method for automatically classifying an activity of a user during a proposal by an agent to the user based on micro-expression and emotion of the user using an artificial intelligence model is provided. The method includes capturing a face of the agent and a face of the user appropriately and recording a voice of the agent and a voice of the user clearly by a facial micro-expression unit that is mounted at a position close to the user, the facial micro-expression unit includes at least one camera or a microphone to capture an interactive sequence of audio-visual information during the proposal by the agent. The method includes acquiring, by an expression analyzer, the interactive sequence of audio-visual information continuously in real-time from the facial micro-expression unit and processing the sequence of the audio-visual information using the artificial intelligence model. The method includes training the artificial intelligence model using historical interactive sequences of audio-visual information between the user and the agent, historical activities of the user, historical proposals, historical users, historical agents, historical succeeding responses by, (a) segmenting, using semantic analysis technique, a conversation of the historical interactive sequence of audio-visual information during a historical proposal by analyzing a context of the conversation to determine an exact meaning of words in the conversation based on the context, (b) determining, using a micro-expression analysis technique and a gait analysis technique, at least one historical emotion of the user and historical intensity of the at least one emotion, (c) correlating the historical audio-visual session with the at least one emotion of the user, and the intensity of the at least one emotion to classify an activity into at least one of positive events or negative events, the positive events refer to a successful proposal and the negative events refer to an unsuccessful proposal, and (d) obtaining a trained artificial intelligence model by associating a historical succeeding response of the user during the historical proposal based on the classification of the activity; (ii) determine, using the trained artificial intelligence model, the at least one emotion, and the intensity of the at least one emotion of the user during the proposal in real-time. The method includes providing a successful response to the agent by creating, a record of a set of questions and responses from at least one of (i) the determined at least one emotion, and the determined intensity of the at least one emotion of the user during the proposal in real-time, or (ii) the response is detected if a question is from the standard training record- ¶0014).
It would have been obvious to one with ordinary skill in the art before the effective filing date of the claimed invention, to modify the teachings of David, Bogdan and Xiao with the teachings of Prashant to develop an AI that enables the fastest user interaction which in turn improves users' conversational experience (Prashant- ¶0003).
Regarding claim 19, apparatus claim 19 is drawn to the apparatus using/performing the same method as claimed in claim 10. Therefore, apparatus claim 19 corresponds to method claim 10, and is rejected for the same reasons of obviousness as used above.
Allowable Subject Matter
Claims 4-6 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to CLIFFORD HILAIRE whose telephone number is (571)272-8397. The examiner can normally be reached 5:30-1400.
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, SATH V PERUNGAVOOR can be reached at (571)272-7455. 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.
CLIFFORD HILAIRE
Primary Examiner
Art Unit 2488
/CLIFFORD HILAIRE/Primary Examiner, Art Unit 2488