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 Interpretation
Examiner notes that method claims 7-8 recite limitations with conditional language. As per MPEP 2111.04, “The broadest reasonable interpretation of a method (or process) claim having contingent limitations requires only those steps that must be performed and does not include steps that are not required to be performed because the condition(s) precedent are not met”. However, system claims 16-17 recite limitations with the same scope as claims 7-8. MPEP 2111.04 further states, “The broadest reasonable interpretation of a system (or apparatus or product) claim having structure that performs a function, which only needs to occur if a condition precedent is met, requires structure for performing the function should the condition occur. The system claim interpretation differs from a method claim interpretation because the claimed structure must be present in the system regardless of whether the condition is met and the function is actually performed”. Therefore, for compact prosecution purposes, claims 7-8 will be examined like claims 16-17 such that the conditions are occurring and therefore hold patentable weight.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefore, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101. Claims 1-9 are directed to a method, claims 10-18 are directed to a system, and claims 19-20 are directed to a non-transitory computer-readable medium; therefore, claims 1-20 fall within one of the four statutory categories (i.e., process, machine, manufacture, or composition of matter). However, claims 1-20 fall within the judicial exception of an abstract idea, specifically the abstract ideas of “Mental Processes” (including observation, evaluation, and opinion) and “Mathematical Concepts (including mathematical calculations and relationships)”.
Claim 1:
Claim 1 is directed to a method; therefore, the claim does fall within one of the four statutory categories (i.e., process, machine, manufacture, or composition of matter).
Claim 1 recites the following abstract ideas:
Step 2A Prong 1:
estimating, [by the at least one processor,] based on an output generated by the first gate, a first probability that the first proposed input is classifiable as being out-of-distribution (OOD) (mental step directed to observation, evaluation – a person could estimate a probability that an observed input is classifiable as being out-of-distribution in their mind based on an observed output from a first gate of a neural network. Wherein this limitation is executed by a processor is interpreted as merely implementing the abstract idea using a generic computer component (see MPEP 2106.05(f));
estimating, [by the at least one processor,] based on a respective output generated by each respective one of the at least second gate, a corresponding probability that the first proposed input is classifiable as being OOD (mental step directed to observation, evaluation – a person could estimate a corresponding probability that an observed input is classifiable as being out-of-distribution in their mind based on an observed output from at least a second gate of a neural network. Wherein this limitation is executed by a processor is interpreted as merely implementing the abstract idea using a generic computer component (see MPEP 2106.05(f)); and
determining, based on each of the first probability and each corresponding probability, whether the first proposed input is classifiable as being OOD (mental step directed to observation, evaluation – a person could determine whether an observed proposed input is classifiable as being out-of-distribution in their mind based on an observed or mentally determined first and corresponding probability).
Claim 1 recites the following additional elements:
receiving, by the at least one processor, a first proposed input to a first neural network at a first gate of the first neural network; and forwarding, by the at least one processor, the first proposed input to at least a second gate of the first neural network.
Step 2A Prong 2:
The processor is interpreted as a generic computer component merely used to implement the claimed abstract ideas. Receiving a proposed input to a first gate of a neural network and forwarding the proposed input to at least a second gate are each interpreted as steps directed to mere data gathering in the technological environment or field of use in which the claimed abstract ideas are performed. These additional elements, when considered as a whole with the aforementioned abstract ideas, do not integrate those abstract ideas into a practical application (see MPEP 2106.05(f), MPEP 2106.05(g), and MPEP 2106.05(h)).
Step 2B:
The processor is interpreted as a generic computer component. Receiving a proposed input to a first gate of a neural network and forwarding the proposed input to at least a second gate are each interpreted as steps directed to the well-understood, routine, conventional activity of transmitting and receiving data over a network. These additional elements, when considered as a whole with the aforementioned abstract ideas, do not amount to significantly more than those abstract ideas (see MPEP 2106.05(d)).
Claim 2 recites wherein the forwarding of the first proposed input to the at least second gate of the first neural network comprises skipping at least one layer of the first neural network (forwarding a proposed input to a second neural network gate by skipping a layer of the neural network is interpreted as mere data gathering and as the well-understood, routine, conventional activity of transmitting data over a network. This additional element, when considered as a whole with the aforementioned abstract ideas, does not integrate those abstract ideas into a practical application or amount to significantly more than the claimed abstract ideas (see MPEP 2106.05(d)(II) and MPEP 2106.05(g)).
Claim 3 recites wherein the forwarding of the first proposed input to the at least second gate of the first neural network further comprises forwarding the first proposed input to a final gate of the first neural network (forwarding a proposed input to a final neural network gate is interpreted as mere data gathering and as the well-understood, routine, conventional activity of transmitting data over a network. This additional element, when considered as a whole with the aforementioned abstract ideas, does not integrate those abstract ideas into a practical application or amount to significantly more than the claimed abstract ideas (see MPEP 2106.05(d)(II) and MPEP 2106.05(g)).
Claim 4 recites wherein the estimating of the first probability comprises calculating a first deep deterministic uncertainty (DDU) value with respect to the first gate, and the estimating of each corresponding probability comprises calculating a respective DDU value with respect to each respective one of the at least second gate (mental step directed to observation, evaluation – a person could calculate a respective deep deterministic uncertainty value in their mind to estimate a corresponding probability for each output from a first and at least second gate. Examiner notes that the broadest reasonable interpretation of calculating a deep deterministic uncertainty value also includes the mathematical calculations as described in paragraphs [0089]-[0090] of Applicant’s specification).
Claim 5 recites wherein the estimating of the first probability comprises calculating a first energy score with respect to the first gate, and the estimating of each corresponding probability comprises calculating a respective energy score with respect to each respective one of the at least second gate (mental step directed to observation, evaluation – a person could calculate a respective energy score in their mind to estimate a corresponding probability for each output from a first and at least second gate. Examiner notes that the broadest reasonable interpretation of calculating an energy score also includes the mathematical calculations as described in paragraph [0092] of Applicant’s specification).
Claim 6 recites wherein the determining comprises determining whether at least a first predetermined number of estimated probabilities exceed a first predetermined threshold value (mental step directed to observation, evaluation – a person could determine whether a predetermined number of observed or estimated probabilities exceed an observed or mentally predetermined threshold value in their mind).
Claim 7 recites when the first predetermined number of estimated probabilities exceeds the first predetermined threshold value, determining that the first proposed input is OOD and discarding the first proposed input (mental step directed to observation, evaluation – a person could determine that an observed proposed input is out-of-distribution and discard, or decide not to utilize, that input in their mind having observed or mentally determined that a predetermined number of observed or estimated probabilities exceeds an observed or mentally predetermined threshold value).
Claim 8 recites when the first proposed input has been forwarded to a final gate of the first neural network and an estimation of a respective probability that the first proposed input is OOD has been performed with respect to the final gate and the first predetermined number of estimated probabilities has not exceed the first predetermined threshold value, determining that the first proposed input is not OOD and retaining the first proposed input (mental step directed to observation, evaluation – a person could determine that an observed proposed input is not out-of-distribution in their mind having observed or mentally determined that that the proposed input has been forwarded to a final neural network gate and a predetermined number of observed or estimated probabilities does not exceed an observed or mentally predetermined threshold value).
Claim 9 recites wherein the first neural network is usable for performing a classification task that relates to at least one from among high frequency trading, a deep learning model that is installed in a drone, a deep learning model that is installed in a self-driving automobile, and financial fraud detection (Examiner notes that this limitation does not actively perform a classification task and is interpreted as the intended use of the neural network, which does not provide additional patentable weight (see MPEP 2103). However, if this limitation actively claimed a classification task related to at least one from among high frequency trading, a deep learning model that is installed in a drone, a deep learning model that is installed in a self-driving automobile, and financial fraud detection, it would be interpreted as merely describing the field of use or technological environment in which the claimed abstract ideas are performed (see MPEP 2106.05(h)).
Claim 10 is a system claim and its limitation is included in claim 1. The only difference is that claim 10 requires a system comprising a processor, memory, and communication interface, which are interpreted as generic computer components used to merely implement the abstract ideas as identified in the analysis of claim 1 (see MPEP 2106.05(f)). Therefore, claim 10 is rejected for the same reasons as claim 1.
Claim 11 is a system claim and its limitation is included in claim 2. Claim 11 is rejected for the same reasons as claim 2.
Claim 12 is a system claim and its limitation is included in claim 3. Claim 12 is rejected for the same reasons as claim 3.
Claim 13 is a system claim and its limitation is included in claim 4. Claim 13 is rejected for the same reasons as claim 4.
Claim 14 is a system claim and its limitation is included in claim 5. Claim 14 is rejected for the same reasons as claim 5.
Claim 15 is a system claim and its limitation is included in claim 6. Claim 15 is rejected for the same reasons as claim 6.
Claim 16 is a system claim and its limitation is included in claim 7. Claim 16 is rejected for the same reasons as claim 7.
Claim 17 is a system claim and its limitation is included in claim 8. Claim 17 is rejected for the same reasons as claim 8.
Claim 18 is a system claim and its limitation is included in claim 9. Claim 18 is rejected for the same reasons as claim 9.
Claim 19 is a system claim and its limitation is included in claim 1. The only difference is that claim 19 requires a non-transitory computer-readable medium, which is interpreted as a generic computer component used to merely implement the abstract ideas as identified in the analysis of claim 1 (see MPEP 2106.05(f)). Therefore, claim 19 is rejected for the same reasons as claim 1.
Claim 20 is a non-transitory computer-readable medium claim and its limitation is included in claim 2. Claim 20 is rejected for the same reasons as claim 2.
Viewed as a whole, these additional claim elements do not provide meaningful limitations to transform the abstract idea into a patent eligible application of the abstract idea such that the claims amount to significantly more than the abstract idea itself. Therefore, the claims are rejected under 35 U.S.C. 101.
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 1-3, 5-12, and 14-20 are rejected under 35 U.S.C. 103 as being unpatentable over Wang et al (“SkipNet: Learning Dynamic Routing in Convolutional Networks”, herein Wang) in view of Ni et al (US 20230076575 A1, herein Ni).
Regarding claim 1, Wang teaches a method being implemented by at least one processor, the method comprising: receiving, by the at least one processor, a first proposed input to a first neural network at a first gate of the first neural network; (section 3 para. 1 recites “SkipNets are convolutional networks in which individual layers are selectively included or excluded for a given input. The per-input selection of layers is accomplished using small gating networks that are interposed between layers. The gating networks map the output of the previous layer or group of layers to a binary decision to execute or bypass the subsequent layer or group of layers as illustrated in Fig. 2” (i.e., fig. 2 shows where input data x is received by a first layer of a neural network, such as a convolutional layer from fig. 3));
estimating, by the at least one processor based on an output generated by the first gate, a first probability that the first proposed input is classifiable [as being out-of-distribution (OOD)] (section 1 para. 5 recites “we treat the probabilistic gate outputs as an initial skipping policy and use REINFORCE to refine the policy without relaxation”. Section 3.1 para. 1 recite “we evaluate two feed-forward convolutional gate designs (Fig. 2a). The FFGate-I (Fig. 11a) design is composed of two 3x3 convolutional layers with stride of 1 and 2 respectively followed by a global average pooling layer and a fully connected layer to output a single dimension vector”. Section 4 para. 1 recites “We evaluate a range of SkipNet architectures and our proposed training procedure on four image classification benchmarks” (i.e., at least a first gated layer, such as a convolutional layer from fig. 2-3, can output a probabilistic output usable for a classification task));
forwarding a first proposed input to at least a second gate of the first neural network (section I para. 2 recites “we introduce SkipNets (see Fig. 1) which are modified residual networks with gating units that dynamically select which layers of a convolutional neural network should be skipped during inference” (i.e., forwarding an input to at least a second gate of a neural network));
estimating, by the at least one processor based on a respective output generated by each respective one of the at least second gate, a corresponding probability that the first proposed input is classifiable [as being OOD] (section 1 para. 5 recites “we treat the probabilistic gate outputs as an initial skipping policy and use REINFORCE to refine the policy without relaxation”. Section 3 para. 1 recites “The gating networks map the output of the previous layer or group of layers to a binary decision to execute or bypass the subsequent layer or group of layers as illustrated in Fig. 2” (i.e., at least a second gated layer, such as a convolutional layer from fig. 2-3, can output a probabilistic output usable for a classification task));
and determining, based on each of the first probability and each corresponding probability, whether the first proposed input is classifiable [as being OOD] (section 1 para. 5 recites “we treat the probabilistic gate outputs as an initial skipping policy and use REINFORCE to refine the policy without relaxation”. Section 3.1 para. 1 recite “we evaluate two feed-forward convolutional gate designs (Fig. 2a). The FFGate-I (Fig. 11a) design is composed of two 3x3 convolutional layers with stride of 1 and 2 respectively followed by a global average pooling layer and a fully connected layer to output a single dimension vector”. Section 4 para. 1 recites “We evaluate a range of SkipNet architectures and our proposed training procedure on four image classification benchmarks” (i.e., the gates, such as the convolutional layers in fig. 2-3, output probabilistic outputs usable for a classification task)).
However, while at least section 3.1 of Wang teaches estimating outputs from gated convolutional layers, Wang does not explicitly teach a method for automatically detecting out-of-distribution inputs to neural networks, and estimating that a proposed input is classifiable as out-of-distribution (OOD).
Ni teaches a method for automatically detecting out-of-distribution inputs to neural networks (para. [0004] recites “The method includes learning a meta-training model that simultaneously classifies dialysis in-distribution events and detects out-of-distribution (OOD) events”),
and estimating that a proposed input is classifiable as out-of-distribution (OOD) (para. [0075]-[0076] recite “The meta-trained ODMP component 130 uses the prototype network 150, the dictionary 145, and the attention component 146 to estimate the mean and variance (148) of the new support set. With the estimated mean and variance (148), the ODMP component 180 performs OOD detection by computing the energy score E(x) and uses a pre-defined threshold to determine OOD samples” (i.e., estimating whether an input is out-of-distribution based at least in part on the corresponding outputs, such as the probabilistic outputs from the convolutional layers in Wang)).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine these teachings by combining the layer-skipping classification model from Wang with the classification model from Ni to classify out-of-distribution data. Wang teaches in at least section 3.1 that the convolutional layers can be connected to an LSTM gate, such as the LSTM gates from at least paragraph [0050] of Ni, to compute a final decision, or classification. Wang also teaches in at least section 1 paragraph 2 that “Not only can skipping policies significantly reduce the average cost of model inference, they also provide insight into the diminishing return and role of individual layers”. Accordingly, one of ordinary skill in the art would be motivated to connect the probabilistic convolutional layers from Wang to the LSTM layers from Ni and potentially skip layers in this combined model using the method from Wang to provide additional insight and reduce the cost of model inference when determining out-of-distribution data.
Regarding claim 2, the combination of Wang and Ni teaches the method of claim 1 as mentioned above, wherein the forwarding of the first proposed input to the at least second gate of the first neural network comprises skipping at least one layer of the first neural network (Wang section I para. 2 recites “we introduce SkipNets (see Fig. 1) which are modified residual networks with gating units that dynamically select which layers of a convolutional neural network should be skipped during inference. Fig. 3c and section 3.1 para. 2 recite “we introduce a recurrent gate (RNNGate) design which enables parameter sharing and allows gates to re-use computation across stages. We adopt a single layer Long Short-Term Memory (LSTM) with hidden unit size of 10. At each gate, we project the LSTM output to a one-dimensional vector to compute the final gate decision” (i.e., forwarding an input to at least a second LSTM gate of a neural network. Examiner notes that at least Fig. 1 shows examples where one or more layers can be skipped)).
Regarding claim 3, the combination of Wang and Ni teaches the method of claim 2 as mentioned above, wherein the forwarding of the first proposed input to the at least second gate of the first neural network further comprises forwarding the first proposed input to a final gate of the first neural network (Wang section I para. 2 recites “we introduce SkipNets (see Fig. 1) which are modified residual networks with gating units that dynamically select which layers of a convolutional neural network should be skipped during inference. Fig. 3c and section 3.1 para. 2 recite “we introduce a recurrent gate (RNNGate) design which enables parameter sharing and allows gates to re-use computation across stages. We adopt a single layer Long Short-Term Memory (LSTM) with hidden unit size of 10. At each gate, we project the LSTM output to a one-dimensional vector to compute the final gate decision” (i.e., forwarding an input to at least a second LSTM gate of a neural network. Examiner notes that at least Fig. 1 shows examples where one or more layers can be skipped to forward data to a final layer)).
Regarding claim 5, the combination of Wang and Ni teaches the method of claim 1 as mentioned above, wherein the estimating of the first probability comprises calculating a first energy score with respect to the first gate, and the estimating of each corresponding probability comprises calculating a respective energy score with respect to each respective one of the at least second gate (Ni para. [0075]-[0076] recite “The meta-trained ODMP component 130 uses the prototype network 150, the dictionary 145, and the attention component 146 to estimate the mean and variance (148) of the new support set. With the estimated mean and variance (148), the ODMP component 180 performs OOD detection by computing the energy score E(x) and uses a pre-defined threshold to determine OOD samples” (i.e., calculating an energy score based at least in part on the corresponding outputs, such as the probabilistic outputs from the convolutional layers in Wang)).
Regarding claim 6, the combination of Wang and Ni teaches the method of claim 1 as mentioned above, wherein the determining comprises determining whether at least a first predetermined number of estimated probabilities exceed a first predetermined threshold value (Ni para. [0076] recites “With the estimated mean and variance (148), the ODMP component 180 performs OOD detection by computing the energy score E(x) and uses a pre-defined threshold to determine OOD samples as: F(x) = { 1, E(x) > t; 0, E(x) ≤ t.” (i.e., determining whether the estimate from the model, such as the probabilistic outputs from the convolutional layers in Wang, exceeds a predetermined threshold value)).
Regarding claim 7, the combination of Wang and Ni teaches the method of claim 6 as mentioned above, further comprising: when the first predetermined number of estimated probabilities exceeds the first predetermined threshold value, determining that the first proposed input is OOD and discarding the first proposed input (Ni para. [0076]-[0077] recite “With the estimated mean and variance (148), the ODMP component 180 performs OOD detection by computing the energy score E(x) and uses a pre-defined threshold to determine OOD samples as: F(x) = { 1, E(x) > t; 0, E(x) ≤ t. Then for those regarded as in-distribution samples, the ODMP system 100 computes the classification probability as the predictive score of events” (i.e., determining that an input is out-of-distribution based on a threshold comparison to the output of the model, which could include the probabilistic outputs from the convolutional layers in Wang. Examiner notes that the classification step only utilizes inputs regarded as in-distribution, which falls under the broadest reasonable interpretation of discarding out-of-distribution inputs)).
Regarding claim 8, the combination of Wang and Ni teaches the method of claim 6 as mentioned above, further comprising: when the first proposed input has been forwarded to a final gate of the first neural network and an estimation of a respective probability that the first proposed input is OOD has been performed with respect to the final gate and the first predetermined number of estimated probabilities has not exceed the first predetermined threshold value, determining that the first proposed input is not OOD and retaining the first proposed input (Ni para. [0075]-[0077] recite “The meta-trained ODMP component 130 uses the prototype network 150, the dictionary 145, and the attention component 146 to estimate the mean and variance (148) of the new support set. With the estimated mean and variance (148), the ODMP component 180 performs OOD detection by computing the energy score E(x) and uses a pre-defined threshold to determine OOD samples as: F(x) = { 1, E(x) > t; 0, E(x) ≤ t. Then for those regarded as in-distribution samples, the ODMP system 100 computes the classification probability as the predictive score of events” (i.e., determining that an input is not out-of-distribution based on a threshold comparison to an output of the model, which could include the probabilistic outputs from the convolutional layers in Wang. Examiner notes that the classification step only utilizes, or retains, inputs that have been determined to be in-distribution, or not out-of-distribution)).
Regarding claim 9, the combination of Wang and Ni teaches the method of claim 1 as mentioned above, wherein the first neural network is usable for performing a classification task that relates to at least one from among high frequency trading, a deep learning model that is installed in a drone, a deep learning model that is installed in a self-driving automobile, and financial fraud detection (Examiner notes that this limitation does not actively perform a classification task and is interpreted as the intended use of the neural network, which does not provide additional patentable weight (see MPEP 2103). However, Examiner notes that Ni teaches a neural network deep learning model usable for a classification task in at least paragraphs [0032] and [0077]).
Claim 10 is a system claim and its limitation is included in claim 1. The only difference is that claim 10 requires a system (at least para. [0006] of Ni teaches a non-transitory computer-readable medium). Therefore, claim 10 is rejected for the same reasons as claim 1.
Claim 11 is a system claim and its limitation is included in claim 2. Claim 11 is rejected for the same reasons as claim 2.
Claim 12 is a system claim and its limitation is included in claim 3. Claim 12 is rejected for the same reasons as claim 3.
Claim 14 is a system claim and its limitation is included in claim 5. Claim 14 is rejected for the same reasons as claim 5.
Claim 15 is a system claim and its limitation is included in claim 6. Claim 15 is rejected for the same reasons as claim 6.
Claim 16 is a system claim and its limitation is included in claim 7. Claim 16 is rejected for the same reasons as claim 7.
Claim 17 is a system claim and its limitation is included in claim 8. Claim 17 is rejected for the same reasons as claim 8.
Claim 18 is a system claim and its limitation is included in claim 9. Claim 18 is rejected for the same reasons as claim 9.
Claim 19 is a system claim and its limitation is included in claim 1. The only difference is that claim 19 requires a non-transitory computer-readable medium (at least para. [0005] of Ni teaches a non-transitory computer-readable medium). Therefore, claim 19 is rejected for the same reasons as claim 1.
Claim 20 is a non-transitory computer-readable medium claim and its limitation is included in claim 2. Claim 20 is rejected for the same reasons as claim 2.
Claims 4 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Wang et al (“SkipNet: Learning Dynamic Routing in Convolutional Networks”, herein Wang) in view of Ni et al (US 20230076576 A1, herein Ni), in further view of Mukhoti et al (“Deep Deterministic Uncertainty: A New Simple Baseline”, herein Mukhoti).
Regarding claim 4, the combination of Wang and Ni teaches the method of claim 1 as mentioned above, estimating of the first probability [comprises calculating a first deep deterministic uncertainty (DDU) value] with respect to the first gate, and estimating each corresponding probability [comprises calculating a respective DDU value] with respect to each respective one of the at least second gate (Ni fig. 4 and para. [0050] recite “The temporal channel includes several Long Short-Term Memory (LSTM) layers for processing the temporal features” (i.e., the model has multiple, or at least a first and second, LSTM layers, which one of ordinary skill in the art would recognize are gated layers, to process the input data). Ni para. [0075]-[0076] recite “The meta-trained ODMP component 130 uses the prototype network 150, the dictionary 145, and the attention component 146 to estimate the mean and variance (148) of the new support set. With the estimated mean and variance (148), the ODMP component 180 performs OOD detection by computing the energy score E(x) and uses a pre-defined threshold to determine OOD samples” (i.e., determining whether an input is out-of-distribution based at least in part on the corresponding outputs, such as the probabilistic outputs from the convolutional layers in Wang)).
However, the combination of Wang and Ni does not explicitly teach calculating a deep deterministic uncertainty value.
Mukhoti teaches calculating a deep deterministic uncertainty value (section I para. 4 recites “the combination of using GDA (i.e., Gaussian Discriminant Analysis) for epistemic uncertainty and the softmax predictive distribution for aleatoric uncertainty after training with feature-space regularisation, e.g. residual connections with spectral normalisation, provides a simple baseline which we call Deep Deterministic Uncertainty (DDU)”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine these teachings by adapting the model from Wang (as modified by Ni) to utilize the uncertainty calculation from Mukhoti. Ni teaches a softmax calculation from at least paragraph [0052] and Mukhoti is specifically directed to modifying a softmax prediction with a GDA step to calculate a deep deterministic uncertainty. As both Ni and Mukhoti are directed to detecting out-of-distribution model inputs, one of ordinary skill in the art would recognize that the known softmax computation from Ni could be modified with the known uncertainty calculation method from Mukhoti to “to obtain good epistemic and aleatoric uncertainty estimates as an alternative to deep ensembles without requiring the complexities or computational cost of the current state-of-the-art”, as Mukhoti teaches in section 6.
Claim 13 is a system claim and its limitation is included in claim 4. Claim 13 is rejected for the same reasons as claim 4.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
US 20220253695 A1 (Mozer et al) teaches a parallel cascaded neural network with skip connections to determine out-of-distribution data in fields such as autonomous vehicles or medical data processing.
“Energy-based Out-of-distribution Detection” (Liu et al) teaches a unified framework for calculating an energy score for out-of-distribution detection during classifier training.
“Dynamic Neural Networks: A Survey” (Han et al) teaches an overview of dynamic neural networks and associated early exit and layer skipping methods.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LEAH M FEITL whose telephone number is (571) 272-8350. The examiner can normally be reached on M-F 0900-1700 EST.
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, Viker Lamardo can be reached on (571) 270-5871. 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.
/L.M.F./ Examiner, Art Unit 2147
/MARC S SOMERS/Primary Examiner, Art Unit 2159
Prosecution Insights