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
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1, 3, 7 and 11-12 are rejected under 35 U.S.C. 103 as being unpatentable over Shi et al. (“Loss-Based Attention for Deep Multiple Instance Learning” – hereinafter Shi) and further in view of Hendrycks et al. (“A BASELINE FOR DETECTING MISCLASSIFIED AND OUT-OF-DISTRIBUTION EXAMPLES IN NEURAL NETWORKS” – hereinafter Hendrycks).
In regards to claim 1, Shi discloses a learning device comprising:
learn a discriminative model that discriminates a class to which second data belongs, the second data being data corresponding to an unknown object, (Shi page 5742 right column first paragraph cites “ The main goal of MIL is to learn a classification model with training bag labels in order to predict a test bag label.”, wherein test bag is unknown. Also see page 5744 “Cross-Entropy in Neural Networks” first paragraph and equation 1 teaches neural network prediction vector, wherein equation 1 converts each component into an probability that training image x belongs to class k. Shi page 5744 right column second paragraph teaches L-Layer neural network, which is the Discriminative model.) by using first training data that includes a group including a plurality of pieces of first data corresponding to a same object, and a first data label with respect to the group, wherein the learning includes: (Given a set of training images X = {X1, X2, ··· , Xn} representing n bags, each bag Xi = {xi,1, xi,2, ··· , xi,ni } consists of ni instances and yi ∈ {0, 1, ··· ,K − 1} is the corresponding bag label, where xi,t is the t-th instance in the i-th bag and yi,t ∈ {0, 1, ··· ,K − 1} is the corresponding instance label.)
computing a discrimination probability with respect to the first data by using the discriminative model; (Shi page 5744 “Loss-based Attention Mechanism” second paragraph – page 5745 paragraphs 1-2 to equation 7. Also see page 5744 equation 1 that teaches class probabilities using SoftMax.)
computing a weight that depends on a relative height of the discrimination score in the group; (Shi page 5744 equation 5 defines the weight of instance j in bag I, wherein the top of equation is the instance specific score and the denominator sums every instance in the bag, thus the weight depends on height. Also see equation 6 page 5744 text under equation 6-page 5745 left column to equation 7 teaches large weights correspond approximately to target-class exponential score normalized by the scores of all instances. )
computing a loss weighted by the computed weight; and (Shi equation 6 and the paragraph under it defines total lost and instance log, wherein instance loss is the last line. L1 is the bag loss, l2 is fused instance loss, L2 is a regularization term for predicting instance labels and λ balances bag and instance predictions. Equation 7 also retains the weighted instance loss term in line 3 of equation 7).
learning the discriminative model by using the computed loss. (Shi page 5744 right column second paragraph states “After obtaining zt, the loss function eq. (3) can be utilized to learn model parameters”; equation 7 teaches complete loss and page 5745 right column first paragraph teaches the attention mechanism updates both instance weights and model parameters through all instances; and page 5748 “Conclusion” section teaches “In this paper, we present a novel loss based attention mechanism to simultaneously learn instance weights and predictions, and bag predictions for deep multiple instance learning, by connecting the attention mechanism with the SoftMax and cross-entropy loss functions.”. Thus, is teaches using the computed loss to learn or teach the discriminative model.)
However, Shi does not explicitly disclose computing a discrimination score with respect to the first data by using the discriminative model.
Hendrycks disclose a memory containing program instructions and a processor coupled to the memory, wherein the processor is configured to execute the program instructions to: (Hendrycks does not explicitly disclose a memory and a processor executing program instructions, however it does so implicitly when it discloses assessing the performance of a neural network in several task in computer vision, natural language processing, and automatic speech recognition in the abstract and page 1 last paragraph. All of these task require a computer which includes memory and a process.) computing a discrimination score with respect to the first data by using the discriminative model. (Hendrycks discloses using the maximum class probability from the classifier’s SoftMax distribution as it per-example score in section 3 first paragraph wherein it cites “In what follows we retrieve the maximum/predicted class probability from a SoftMax distribution and thereby detect whether an example is erroneously classified or out-of-distribution. Specifically, we separate correctly and incorrectly classified test set examples and, for each example, compute the SoftMax probability of the predicted class, i.e., the maximum SoftMax probability.3 From these two groups we obtain the area under PR and ROC curves. These areas summarize the performance of a binary classifier discriminating with values/scores (in this case, maximum probabilities from the SoftMaxes) across different thresholds.”. This teaches using maximum probabilities as discrimination scores.)
It would have been obvious to one of ordinary skill in the art before the earliest effective filing date of the claimed invention to modify the teachings of the Shi with that of Hendrycks to allow for using the maximum probabilities from the SoftMax as discrimination score as both reference deal with neural network classification using SoftMax-based scores to evaluate the individual input examples. Do so creating a more accurate system by using Hendrkycks maximum SoftMax confidence as a discrimination score in Shi’s instance weight to assign greater weight to examples with higher confidence thereby reducing influence of samples with low confidence.
Claim 3, The learning device according to claim 1 or 2, wherein the processor is further configured to execute the instructions to the learning means is configured to compute a value obtained by normalizing a strictly monotone increasing function f(s) of the discrimination score with respect to the first data by a total value in the group, as a weight of the first data. (Hendrycks discloses using the maximum class probability from the classifier’s SoftMax distribution as it per-example score in section 3 first paragraph wherein it cites “…These areas summarize the performance of a binary classifier discriminating with values/scores (in this case, maximum probabilities from the SoftMaxes) across different thresholds.”. This teaches using maximum probabilities as discrimination scores.” Shi page 5744 equation 5 teaches an increasing transformation through exponentials, wherein the exponentials are increase, which the examiner maps to strictly monotone increasing functions of the discrimination score. Also Shi equation 5 divides each instance transformed value by the sum of all the instances, which is normalizing, and αi,j in equations 5 and 6 is the weight of the first data.)
In regards to claim 7, Shi in view of Hendrycks disclose the learning device according to claim 1, wherein when N represents a number of discrimination classes of the discriminative model, the discrimination score is computed by using a maximum value of a SoftMax output of an N component of the discriminative model. (Hendrycks discloses using the maximum class probability from the classifier’s SoftMax distribution as it per-example score in section 3 first paragraph wherein it cites “…These areas summarize the performance of a binary classifier discriminating with values/scores (in this case, maximum probabilities from the SoftMaxes) across different thresholds.”. This teaches using maximum probabilities as discrimination scores.” Also Shi page 5744 “Cross-Entropy in Neural Networks” teaches K representing a number of discrimination classes in the model.)
In regards to claim 11, it is the method embodiment of claim 1 with similar limitations and thus rejected using the same reasoning found claim 1.
In regards to claim 12, it is the non-transitory computer-readable medium of claim 1 with similar limitations and thus rejected using the same reason as that found in claim 1.
Claims 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over Shi et al. (“Loss-Based Attention for Deep Multiple Instance Learning” – hereinafter Shi) in view of Hendrycks et al. (“A BASELINE FOR DETECTING MISCLASSIFIED AND OUT-OF-DISTRIBUTION EXAMPLES IN NEURAL NETWORKS” – hereinafter Hendrycks) and further in view of De Freitas et al. (“Using Deep Learning for Trajectory Classification” – hereinafter Nicksson).
In regards to claim 9, Shi in view of Hendrycks discloses the learning device according to claim 1 but does not explicitly disclose wherein the first data is time-series data.
Nicksson discloses wherein the first data is time-series data. (Nicksson abstract teaches time-series data wherein it cites “Trajectory classification is an efficient way to analyze trajectory, consisting of building a prediction model to classify a new trajectory (or sub-trajectory) in a single-class or multi-class.”, this is time-series data as trajectory is an order sequence of positions indexed over time.)
It would have been obvious for one of ordinary skill in the art before the earliest effective filing date of the claimed invention to modify the teachings of the Shi in view of Hendrycks with the teachings of Nicksson in order to allow the first data to be time-series data as both references deal using deep learning models for classification. The benefit of doing so it allows the system to be multi-modal and handles various types of data.
In regards to claim 10, Shi in view of Hendrycks in view of Nicksson disclose the learning device according to claim 1, wherein the first data is time-series data representing a moving locus of an object obtained by observation.(Nicksson Page 1 introduction first paragraph cites, “We strongly believe that these data provide a unique opportunity for understanding the patterns and behaviors of several moving objects, such as people, animals, transportation modes, hurricanes, among others.”; Nicksson Page 1 introduction section paragraph 3 cites “The trajectory classification problem consists of building a prediction model to classify a new trajectory in a single-class or multi-class. The model is trained and learns the patterns (or classes) from a historical labeled trajectory (or sub-trajectory) data. Ba sic examples of trajectory classification are: (i) deter mining the transportation mode of the moving object like car, bus, bike, taxi, airplane, and train; (ii) determine what a user’s next stopping point like home, school, cafe, office, and restaurant; (iii) identifying who is the user of a trajectory (TUL problem).”. These together disclose a time-order sequence of spatial coordinates of a moving object generated by GPS tracking device, and thus a sequence that represents the moving locus or path of the object.)
Allowable Subject Matter
Claims 2 and 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 PAULINHO E SMITH whose telephone number is (571)270-1358. The examiner can normally be reached Mon-Fri. 10AM-6PM CST.
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, Abdullah Kawsar can be reached at 571-270-3169. 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.
/PAULINHO E SMITH/Primary Examiner, Art Unit 2127
Prosecution Insights