The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
DETAILED ACTION
Notice to Applicant
In response to the communication received on 03/02/2026, the following is a Final Office Action for Application No. 18185489.
Status of Claims
Claims 1-30 are pending.
Response to Amendments
Applicant’s amendments have been fully considered.
Response to Arguments
Applicant’s arguments with respect to the claims have been considered but are moot in light of the new grounds of rejection, as necessitated by amendment.
As per the 101 rejection, Applicant argues that the claims are in favor of eligibility per Prong One of Step 2A, however Examiner respectfully disagrees. Per Prong One of Step 2A, the identified recitation of an abstract idea falls within at least one of the Abstract Idea Groupings consisting of: Mathematical Concepts, Mental Processes, or Certain Methods of Organizing Human Activity. Particularly, the identified recitation falls within the Mental Processes including concepts performed in the human mind (including an observation, evaluation judgment, opinion) and/or Certain Methods of Organizing Human Activity including managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules of instructions). Since the recitation of the claims falls into at least one of the above Groupings, there is a basis for providing further analysis with regard to Prong Two of Step 2A to determine whether the recitation of an abstract idea is deduced to being directed to an abstract idea. Thus, the rejection is maintained.
Applicant argues that the claims are in favor of eligibility per Prong Two of Step 2A, however Examiner respectfully disagrees. Per Prong Two of Step 2A, this judicial exception is not integrated into a practical application because the claim as a whole does not integrate the identified abstract idea into a practical application. The processor and/or memory medium is recited at a high level of generality, i.e., as a generic processor performing a generic computer function of processing/transmitting data. This generic processor server limitation is no more than mere instructions to apply the exception using a generic computer component. Further, processor and/or memory medium to inter alia perform the function of taking one or more actions based on the generated output is mere instruction to apply an exception using a generic computer component which cannot integrate a judicial exception into a practical application. Accordingly, this/these additional element(s) does/do not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. In other words, the present claims use a generic processing device and memory medium to inter alia perform the function of taking one or more actions based on the generated output which is a concept that can be performed in the human mind. The processor is merely used to perform the function(s), and the processor does not integrate the abstract idea into a practical application since there are no meaningful limits on practicing the abstract idea. Thus, since the claims are directed to the determined judicial exception in view of the two prongs of Step 2A, the 2019 PEG flowchart is directed to Step 2B. Thus, the rejection is maintained.
Applicant argues that the claims are in favor of eligibility per Step 2B, however Examiner respectfully disagrees. Therein, the additional elements and combinations therewith are examined in the claims to determine whether the claims as a whole amounts to significantly more than the judicial exception. It is noted here that the additional elements are to be considered both individually and as an ordered combination. In this case, the claims each at most comprise additional elements of: processor and/or memory medium. Taken individually, the additional limitations each are generically recited and thus does not add significantly more to the respective limitations. Further, processor and/or memory medium to inter alia perform the function of taking one or more actions based on the generated output is mere instruction to apply an exception using a generic computer component which cannot provide an inventive concept in Step 2B (or, looking back to Step 2A, cannot integrate a judicial exception into a practical application). For further support, the Applicant’s specification supports the claims being directed to use of a generic computer/memory type structure. Taken as an ordered combination, the claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the limitations are directed to limitations referenced in Alice Corp. that are not enough to qualify as significantly more when recited in a claim with an abstract idea include the non-limiting or non-exclusive examples of MPEP § 2106.05. Thus, the rejection is maintained.
In an effort to further expedite prosecution, see: July 2024 Subject Matter Eligibility Examples, Example 47. Anomaly Detection. Per the analysis of claim 2 Example 47, the analysis refers to MPEP 2106.05(f) which provides the following considerations for determining whether a claim simply recites a judicial exception with the words “apply it” (or an equivalent), such as mere instructions to implement an abstract idea on a computer: (1) whether the claim recites only the idea of a solution or outcome i.e., the claim fails to recite details of how a solution to a problem is accomplished; (2) whether the claim invokes computers or other machinery merely as a tool to perform an existing process; and (3) the particularity or generality of the application of the judicial exception. Although the additional elements, e.g. (per Example 47) “using a trained ANN”, limits the identified judicial exceptions, e.g. (per Example 47) “detecting one or more anomalies in a data set using the trained ANN” and, e.g. (per Example 47) “analyzing the one or more detected anomalies using the trained ANN to generate anomaly data,” this type of limitation merely confines the use of the abstract idea to a particular technological environment, e.g. (per Example 47: neural networks) and thus fails to add an inventive concept to the claims. See MPEP 2106.05(h). As an exemplary direction for claim limitations to be eligible, see claims 1 and 3 of Example 47.
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 therefor, subject to the conditions and requirements of this title.
Claims 1-30 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more. The claims fall within statutory class of process or machine; hence, the claims fall under statutory category of Step 1.
Step 2 is the two-part analysis from Alice Corp. (also called the Mayo test). The 2019 PEG makes two changes in Step 2A: It sets forth new procedure for Step 2A (called “revised Step 2A”) under which a claim is not “directed to” a judicial exception unless the claim satisfies a two-prong inquiry. The two-prong inquiry is as follows: Prong One: evaluate whether the claim recites a judicial exception (an abstract idea enumerated in the 2019 PEG, a law of nature, or a natural phenomenon). If claim recites an exception, then Prong Two: evaluate whether the claim recites additional elements that integrate the exception into a practical application of the exception. The claim(s) recite(s) the following abstract idea indicated by non-boldface font and additional limitations indicated by boldface font:
1. A processor-implemented method for machine learning, comprising: generating an intermediate output of a neural network for an input into the neural network, the neural network having a plurality of activation functions; selecting, based on one or more values associated with generating the intermediate output of the neural network, one or more activation functions to apply to the intermediate output; generating an output of the neural network based on the selected one or more activation functions and the intermediate output; and taking one or more actions based on the generated output.
[or]
13. A processor-implemented method for machine learning, comprising: training a neural network having a plurality of activation functions to apply to at least portions of an intermediate output generated by one or more layers of the neural network, the neural network including a selector configured to select at least one activation function of the plurality of activation functions to apply to the intermediate output; and deploying the trained neural network.
[or]
18. A processing system, comprising: a memory having executable instructions stored thereon; and a one or more processors configured to execute the executable instructions in order to cause the system to: generate an intermediate output of a neural network for an input into the neural network; select one or more activation functions to apply to the intermediate output; generate an output of the neural network based on the selected one or more activation functions and the intermediate output; and take one or more actions based on the generated output.
[or]
30. A processing system, comprising: a memory having executable instructions stored thereon; and one or more processors configured to execute the executable instructions in order to cause the system to: train a neural network having a plurality of activation functions to apply to at least portions of an intermediate output generated by one or more layers of the neural network, the neural network including a selector configured to select at least one activation function of the plurality of activation functions to apply to the intermediate output; and deploy the trained neural network.
The claim(s) recite(s) the following summarization of the abstract idea which includes activation functions used to generate outputs of a neural network executed by the additional element(s) of memory and/or processor. This falls into at least the Abstract Idea Grouping of Mental Processes since the information can be analyzed by an abstract evaluation judgment process. Thus, the identified recitation of an abstract idea falls within at least one of the Abstract Idea Groupings consisting of: Mathematical Concepts, Mental Processes, or Certain Methods of Organizing Human Activity since the identified recitation falls within the Mental Processes including concepts performed in the human mind (including an observation, evaluation judgment, opinion).
Per Prong One of Step 2A, the identified recitation of an abstract idea falls within at least one of the Abstract Idea Groupings consisting of: Mathematical Concepts, Mental Processes, or Certain Methods of Organizing Human Activity. Particularly, the identified recitation falls within the Mental Processes including concepts performed in the human mind (including an observation, evaluation judgment, opinion) and/or Certain Methods of Organizing Human Activity including managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules of instructions).
Per Prong Two of Step 2A, this judicial exception is not integrated into a practical application because the claim as a whole does not integrate the identified abstract idea into a practical application. The processor and/or memory is recited at a high level of generality, i.e., as a generic processor performing a generic computer function of processing/transmitting data. This generic processor and/or memory limitation is no more than mere instructions to apply the exception using a generic computer component. Further, taking one or more actions based on the generated output by a processor and/or memory is mere instruction to apply an exception using a generic computer component which cannot integrate a judicial exception into a practical application. Accordingly, this/these additional element(s) does/do not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. Thus, since the claims are directed to the determined judicial exception in view of the two prongs of Step 2A, the 2019 PEG flowchart is directed to Step 2B.
Per Step 2B, the additional elements and combinations therewith are examined in the claims to determine whether the claims as a whole amounts to significantly more than the judicial exception. It is noted here that the additional elements are to be considered both individually and as an ordered combination. In this case, the claims each at most comprise additional elements of: processor and memory. Taken individually, the additional limitations each are generically recited and thus does not add significantly more to the respective limitations. Further, taking one or more actions based on the generated output by a processor and/or memory is mere instruction to apply an exception using a generic computer component which cannot provide an inventive concept in Step 2B (or, looking back to Step 2A, cannot integrate a judicial exception into a practical application). For further support, the Applicant’s specification supports the claims being directed to use of a generic computer/memory type structure at ¶0125 wherein “The means may include various hardware and/or software component(s) and/or module(s), including, but not limited to a circuit, an application specific integrated circuit (ASIC), or processor.” Taken as an ordered combination, the claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the limitations are directed to limitations referenced in Alice Corp. that are not enough to qualify as significantly more when recited in a claim with an abstract idea include, as a non-limiting or non-exclusive examples: i. Adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, e.g., a limitation indicating that a particular function such as creating and maintaining electronic records is performed by a computer, as discussed in Alice Corp., 134 S. Ct. at 2360, 110 USPQ2d at 1984 (see MPEP § 2106.05(f));
PNG
media_image1.png
18
19
media_image1.png
Greyscale
ii. Simply appending well-understood, routine, conventional activities previously known to the industry, specified at a high level of generality, to the judicial exception, e.g., a claim to an abstract idea requiring no more than a generic computer to perform generic computer functions that are well-understood, routine and conventional activities previously known to the industry, as discussed in Alice Corp., 134 S. Ct. at 2359-60, 110 USPQ2d at 1984 (see MPEP § 2106.05(d));
PNG
media_image1.png
18
19
media_image1.png
Greyscale
iii. Adding insignificant extra-solution activity to the judicial exception, e.g., mere data gathering in conjunction with a law of nature or abstract idea such as a step of obtaining information about credit card transactions so that the information can be analyzed by an abstract mental process, as discussed in CyberSource v. Retail Decisions, Inc., 654 F.3d 1366, 1375, 99 USPQ2d 1690, 1694 (Fed. Cir. 2011) (see MPEP § 2106.05(g)); or
PNG
media_image1.png
18
19
media_image1.png
Greyscale
v. Generally linking the use of the judicial exception to a particular technological environment or field of use, e.g., a claim describing how the abstract idea of hedging could be used in the commodities and energy markets, as discussed in Bilski v. Kappos, 561 U.S. 593, 595, 95 USPQ2d 1001, 1010 (2010) or a claim limiting the use of a mathematical formula to the petrochemical and oil-refining fields, as discussed in Parker v. Flook. The courts have recognized the following computer functions inter alia to be well-understood, routine, and conventional functions when they are claimed in a merely generic manner: performing repetitive calculations; receiving, processing, and storing data (e.g., the present claims); electronically scanning or extracting data; electronic recordkeeping; automating mental tasks (e.g., process/machine/manufacture for performing the present claims); and receiving or transmitting data (e.g., the present claims).
The dependent claims do not cure the above stated deficiencies, and in particular, the dependent claims further narrow the abstract idea without reciting additional elements that integrate the exception into a practical application of the exception or providing significantly more than the abstract idea. Since there are no elements or ordered combination of elements that amount to significantly more than the judicial exception, the claims are not eligible subject matter under 35 USC §101.
Thus, viewed as a whole, these additional claim element(s) do not provide meaningful limitation(s) to transform the abstract idea into a patent eligible application of the abstract idea such that the claim(s) amounts to significantly more than the abstract idea itself. Therefore, the claim(s) are rejected under 35 U.S.C. 101 as being directed to non-statutory subject matter.
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 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 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 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 1-30 are rejected under 35 U.S.C. 103 as being unpatentable over Trygg et al. (US 20200074269 A1) hereinafter referred to as Trygg in view of McDonald et al. (US 20230033694 A1) hereinafter referred to as McDonald.
Trygg teaches:
Claim 1. A processor-implemented method for machine learning, comprising:
generating an intermediate output of a neural network for an input into the neural network, the neural network having a plurality of activation functions (Fig. 3 and ¶0038 Further, in the method according to the above-stated aspect, the steps of obtaining the first sets of intermediate output values and constructing the latent variable model may be performed for two or more of the plurality of hidden layers wherein the steps of obtaining the second set of intermediate output values and mapping the second set of intermediate output values to the second set of projected values may be performed concerning said two or more of the plurality of hidden layers.. ¶0147 After linear projection, an activation function f.sub.1 is applied to z to form activation a.sub.1. The activation function f.sub.1 may be a non-linear function. Common choices of activation functions include the rectified linear function f(x)=max(0,x), the sigmoid function f(x)=(1+e.sup.−x).sup.−1, the softmax function f(x)=e.sup.x.sup.j/Σ.sub.m=1.sup.ne.sup.x.sup.m, among others);
selecting, based on one or more values associated with generating the intermediate output of the neural network, one or more activation functions of the plurality of activation functions to apply to the intermediate output (Fig. 3 and ¶0147 After linear projection, an activation function f.sub.1 is applied to z to form activation a.sub.1. The activation function f.sub.1 may be a non-linear function. Common choices of activation functions include the rectified linear function f(x)=max(0,x), the sigmoid function f(x)=(1+e.sup.−x).sup.−1, the softmax function f(x)=e.sup.x.sup.j/Σ.sub.m=1.sup.ne.sup.x.sup.m, among others.);
generating an output of the neural network based on the selected one or more activation functions and the intermediate output (Fig. 3 and ¶0114 The intensity value of each pixel of the input image may be considered as an input value to an input node of an input layer of the exemplary CNN. The exemplary CNN shown in FIG. 2 comprises four convolutional layers C1, C2, C3, C4, two max pool layers MP1, MP2 and an output layer with a softmax function as the activation function of nodes included in the output layer. ¶0169 At step S32, the application 10 may transform the new observation. For example, the application 10 may obtain the activation vector from at least one layer of the deep neural network 100 according to equation (6) stated above, of the new observation. At step S34, the application 10 may project activation using the latent variable model constructed at step S18 as stated above. For example, the application 10 may obtain, from the activation vector obtained at step S32, a corresponding set of projected values (e.g., T.sub.A,new) in the sub-space found by constructing the latent variable model (see equation (9))); and
taking one or more actions based on the generated output (Fig. 3 and ¶0120 Finally, the output of the max pool layer MP2 may be provided to the output layer with a softmax function. The output layer may include one or more output nodes corresponding to one or more groups (or categories) into which the input image may be classified. While this example refers to specific parameters (e.g., a number of filters, a dropout percentage, number of convolutional layers, stride, etc.), the methods and systems are not limited to these embodiments, as a range of values for each parameter is contemplated herein. In some examples, concerning the exemplary CNN shown in FIG. 2, the outputs from the max pool layers MP1 and MP2 may be provided to the outlier detection module 104 (FIG. 1) for detecting whether or not the input image is an outlier with respect to the training dataset used for training the exemplary CNN shown in FIG. 2. For instance, the outlier detection module 104 may construct, for each of the max pool layers MP1 and MP2, a latent variable model using the outputs from the respective max pool layers MP1 and MP2 for possible input images in the training dataset).
Although not explicitly taught by Trygg, McDonald teaches in the analogous art of efficient binary representations from neural networks:
selecting, based on one or more values associated with generating the intermediate output of the neural network, one or more activation functions of the plurality of activation functions to apply to the intermediate output (¶0006 Accordingly, a first example embodiment may involve persistent storage containing a representation of a neural network including an input layer, and output layer, and a hidden layer, wherein nodes of the hidden layer incorporate serialized activation functions, wherein the serialized activation functions for each of the nodes include a sigmoid function and a Beta function, wherein the sigmoid function is applied to weighted outputs from nodes of a previous layer of the neural network, wherein the Beta function is applied to a conductance hyper-parameter and respective outputs of the sigmoid function, and wherein outputs of the Beta function are provided to a subsequent layer of the neural network. One or more processors may be configured to train the neural network until the outputs of the sigmoid function for the nodes of the hidden layer are substantially binary. ¶0106 The solid arrows between pairs of nodes represent connections through which intermediate values flow, and are each associated with a respective weight (e.g., any real number) that is applied to the respective intermediate value. Each node performs an operation on its input values and their associated weights to produce an output value. In some cases this operation may involve a dot-product sum of the products of each input value and associated weight. An activation function may be applied to the result of the dot-product sum to produce the output value. Other operations are possible. ¶0149 In order to show this, a probabilistic model for the stochastic activation function is developed. The probability density functions of the random variables S and Y are obtained, corresponding to the output of the intermediate sigmoid and the final output of stochastic activation, respectively. These functions will have parameters that allow controlling how close the variables are to binary. Using the derived probability density functions, an expression of D.sub.KL(S∥Y) is obtained that proves D.sub.KL(S∥Y) reaches a minimum when the values of s.sub.i are binary. An empirical analysis illustrates how this activation function behaves in practice, in terms of binarization properties relative to hyper-parameters (namely κ) as well as the accuracy versus representational fidelity trade-off.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the efficient binary representations from neural networks of McDonald with the system for data analysis of Trygg for the following reasons:
(1) a finding that there was some teaching, suggestion, or motivation, either in the references themselves or in the knowledge generally available to one of ordinary skill in the art, to modify the reference or to combine reference teachings, e.g. Trygg ¶0005 teaches that it is desirable to obtain general applicability for real-time systems in cases where making multiple inferences is impractical and/or for existing systems that do not fit into the constraints;
(2) a finding that there was reasonable expectation of success since the only difference between the claimed invention and the prior art being the lack of actual combination of the elements in a single prior art reference, e.g. Trygg teaches a deep neural network is provided for processing images and at least a part of a training dataset used for training the deep neural network, and McDonald teaches persistent storage that contains a representation of a neural network including an input layer, and output layer, and a hidden layer, wherein nodes of the hidden layer incorporate serialized activation functions; and
(3) whatever additional findings based on the Graham factual inquiries may be necessary, in view of the facts of the case under consideration, to explain a conclusion of obviousness, e.g. Trygg at least the above cited paragraphs, and McDonald at least the inclusively cited paragraphs.
Therefore, it would be obvious to one skilled in the art at the time of the invention to combine the efficient binary representations from neural networks of McDonald with the system for data analysis of Trygg. The rationale to support a conclusion that the claim would have been obvious is that "a person of ordinary skill in the art would have been motivated to combine the prior art to achieve the claimed invention and whether there would have been a reasonable expectation of success in doing so." DyStar Textilfarben GmbH & Co. Deutschland KG v. C.H. Patrick Co., 464 F.3d 1356, 1360, 80 USPQ2d 1641, 1645 (Fed. Cir. 2006). See MPEP 2143(G).
Trygg teaches:
Claim 2. The method of Claim 1, wherein: each activation function of the one or more activation functions is associated with a range of values for the intermediate output; and selecting the one or more activation functions to apply to the intermediate output comprises selecting an activation function associated with the range of values in which the value of the intermediate output lies (¶0146 In order to provide a possibility to detect outliers during prediction in a similar way as PLS, a feed-forward neural network may be considered as a series of non-linear transformations. In other words, an activation vector a.sub.i of an observation x (input data) from layer i (=1, 2, 3, 4, . . . ) of the given network may be given by the nested series of transformations as follows: a.sub.if.sub.i(W.sub.if.sub.i−1(W.sub.i−1f.sub.i−2( . . . f.sub.1(W.sub.1x)))) where f.sub.k (k=1, 2, . . . , i) may be activation functions and W.sub.k (k=1, 2, . . . , i) may be weight matrices. The activation vector a.sub.i of the observation x may be considered as an intermediate output from the layer i of the given network and may include element values corresponding to outputs from respective nodes of the layer i when the observation x is input to the given network).
Trygg teaches:
Claim 3. The method of Claim 1, wherein selecting the one or more activation functions to apply to the intermediate output comprises: segmenting the intermediate output into a plurality of segments; and for each respective segment of the plurality of segments, selecting a respective activation function to apply to the respective segment (¶0207 The methods and systems as described herein may be applied to phase microscopy image analysis. The task of identifying nuclei from phase contrast images is challenging and typically relies on fluorescent markers binding to nuclei to provide an extra image channel highlighting locations of nuclei. By applying deep CNN:s, a model may be generated and trained that segments background from cells as well as identifies nuclei of cells from phase contrast images without using fluorescence labels. If the CNN is trained on phase contrast images from multiple cell types, the CNN may also identify new previously unseen cell types).
Trygg teaches:
Claim 4. The method of Claim 3, wherein generating the output of the neural network comprises: generating segment-specific activation outputs based on a value of each respective segment of the plurality of segments and the selected respective activation function for the respective segment; and concatenating the segment-specific activation outputs into a combined output (¶0196 To provide a latent variable model for detecting these outliers, features (e.g., intermediate outputs) were extracted from the LSTM-layers L1, L2, L3 of the predictive maintenance model (see FIG. 12). For the first two LSTM-layers L1, L2, the average across time for each LSTM-node was used to summarize each sliding window. For the last LSTM-layer L3, simply the last cycle in the time sequence was used. The output from the three LSTM-layers L1, L2, L3 were then concatenated into a single matrix).
Trygg teaches:
Claim 5. The method of Claim 3, wherein segmenting the intermediate output into the plurality of segments comprises segmenting the intermediate output into a high segment and a low segment, the high segment corresponding to bits of the intermediate output having a position above a split point and the low segment corresponding to bits of the intermediate output having a position below the split point (¶0165 At step S24, the application 10 may determine a threshold value for the distance. Additionally or alternatively, a threshold value for the squared approximation residual may be determined. The threshold value(s) may later be used for determining whether a new observation (e.g., input image) is an outlier with respect to the training dataset. For obtaining the threshold value(s), the distances and/or the squared approximation residuals calculated at step S22 may be used ¶0172 At step S38, the application 10 may determine whether or not the distance calculated at step S36 is larger than the threshold determined at step S24. If yes at step S38, the process may proceed to step S40 and the application 10 may determine that the new observation is an outlier. In this case, the system may report the model prediction as an unreliable prediction, since the new observation is determined to be an outlier. The process may end after step S40. If no at step S38, the process may proceed to step S42 and the application 10 may determine that the prediction made by the deep neural network 100 for the new observation can be trusted. In this case, the system may report the model prediction as a reliable prediction, since the new observation is determined to be a non-outlier. The process may end after step S42).
Trygg teaches:
Claim 6. The method of Claim 1, wherein the one or more activation functions to apply to the intermediate output are selected based on one or more statistical measurements associated with the intermediate output (¶0165 At step S24, the application 10 may determine a threshold value for the distance. Additionally or alternatively, a threshold value for the squared approximation residual may be determined. The threshold value(s) may later be used for determining whether a new observation (e.g., input image) is an outlier with respect to the training dataset. For obtaining the threshold value(s), the distances and/or the squared approximation residuals calculated at step S22 may be used. ¶0201 In some other examples, activations from two or more hidden layers may be used for performing the outlier detection. For instance, in order to avoid the difficulty in selecting which layer to use for outlier detection, the measures from all (hidden) layers may be combined using Gaussian Kernel Density estimation, …Kernel density estimations can be calculated separately for training set Mahalanobis-distances and residual sum of squares, but combining all layers. The probabilities of each image can be approximated under the resulting kernel density functions using Monte Carlo integration).
Trygg teaches:
Claim 7. The method of Claim 6, wherein the one or more statistical measurements comprise a mean and standard deviation relative to a distribution of possible values for the intermediate output (¶0165 At step S24, the application 10 may determine a threshold value for the distance. Additionally or alternatively, a threshold value for the squared approximation residual may be determined. The threshold value(s) may later be used for determining whether a new observation (e.g., input image) is an outlier with respect to the training dataset. For obtaining the threshold value(s), the distances and/or the squared approximation residuals calculated at step S22 may be used.).
Trygg teaches:
Claim 8. The method of Claim 7, wherein selecting the one or more activation functions comprises: selecting a first activation function when the value of the intermediate output is within a threshold number of standard deviations from the mean, and selecting a second activation function when the value is more than the threshold number of standard deviations from the mean (¶0165 At step S24, the application 10 may determine a threshold value for the distance. Additionally or alternatively, a threshold value for the squared approximation residual may be determined. The threshold value(s) may later be used for determining whether a new observation (e.g., input image) is an outlier with respect to the training dataset. For obtaining the threshold value(s), the distances and/or the squared approximation residuals calculated at step S22 may be used.).
Trygg teaches:
Claim 9. The method of Claim 6, wherein the one or more statistical measurements are related to a probability density function describing probabilities associated with different values for the intermediate output (¶0201 In some other examples, activations from two or more hidden layers may be used for performing the outlier detection. For instance, in order to avoid the difficulty in selecting which layer to use for outlier detection, the measures from all (hidden) layers may be combined using Gaussian Kernel Density estimation, …Kernel density estimations can be calculated separately for training set Mahalanobis-distances and residual sum of squares, but combining all layers. The probabilities of each image can be approximated under the resulting kernel density functions using Monte Carlo integration).
Trygg teaches:
Claim 10. The method of Claim 1, wherein selecting the one or more activation functions to apply to the intermediate output is based, at least in part, on the input into the neural network (¶0145 The method performed by the outlier detection module 104 to detect prediction-time outliers in a deep learning system may be based on the fact that a neural network may function by transforming input data. When input data is fed through a deep neural network, multiple intermediate representations of the data may exist, where the intermediate representations may be used for prediction (e.g., of a group into which the input data is classified in case the deep neural network is configured to solve a classification problem).).
Trygg teaches:
Claim 11. The method of Claim 1, wherein selecting the one or more activation functions to apply to the intermediate output is based, at least in part, on weights in the neural network (¶0146 The activation vector a.sub.i of the observation x may be considered as an intermediate output from the layer i of the given network and may include element values corresponding to outputs from respective nodes of the layer i when the observation x is input to the given network. Each of these activations a.sub.k may provide a feature representation of the input data. Although the weight matrices may be commonly obtained by supervised training using back-propagation, the activations may simply provide transformed, or pre-processed, representations of the input data.).
Trygg teaches:
Claim 12. The method of Claim 1, wherein selecting the one or more activation functions to apply to the intermediate output comprises searching for a post-activation value of the intermediate output in a lookup table based on a value of the intermediate output (¶0146 The activation vector a.sub.i of the observation x may be considered as an intermediate output from the layer i of the given network and may include element values corresponding to outputs from respective nodes of the layer i when the observation x is input to the given network. Each of these activations a.sub.k may provide a feature representation of the input data. Although the weight matrices may be commonly obtained by supervised training using back-propagation, the activations may simply provide transformed, or pre-processed, representations of the input data.).
Trygg teaches:
Claim 13. A processor-implemented method for machine learning, comprising:
training a neural network having a plurality of activation functions to apply to at least portions of an intermediate output generated by one or more layers of the neural network, the neural network including a selector configured to select, based on one or more values associated with generating the intermediate output of the neural network, at least one activation function of the plurality of activation functions to apply to the intermediate output (¶0008 a computer-implemented method for data analysis is provided. The method includes: obtaining a deep neural network for processing data and at least a part of a training dataset used for training the deep neural network, the deep neural network comprising a plurality of hidden layers, the training dataset including possible observations that can be input to the deep neural network.. Fig. 3 and ¶0038 Further, in the method according to the above-stated aspect, the steps of obtaining the first sets of intermediate output values and constructing the latent variable model may be performed for two or more of the plurality of hidden layers wherein the steps of obtaining the second set of intermediate output values and mapping the second set of intermediate output values to the second set of projected values may be performed concerning said two or more of the plurality of hidden layers ¶0147 After linear projection, an activation function f.sub.1 is applied to z to form activation a.sub.1. The activation function f.sub.1 may be a non-linear function. Common choices of activation functions include the rectified linear function f(x)=max(0,x), the sigmoid function f(x)=(1+e.sup.−x).sup.−1, the softmax function f(x)=e.sup.x.sup.j/Σ.sub.m=1.sup.ne.sup.x.sup.m, among others); and
deploying the trained neural network (Fig. 3 and ¶0167 The right hand side of FIG. 3 shows an exemplary process performed by the computing system 1 for detecting outliers. At step S30, the application 10 may receive a new observation. For example, in case the deep neural network 100 is a CNN as shown in FIG. 2, an image to be input to the CNN may be received as the new observation).
Although not explicitly taught by Trygg, McDonald teaches in the analogous art of efficient binary representations from neural networks:
a selector configured to select, based on one or more values associated with generating the intermediate output of the neural network, at least one activation function of the plurality of activation functions to apply to the intermediate output (¶0006 Accordingly, a first example embodiment may involve persistent storage containing a representation of a neural network including an input layer, and output layer, and a hidden layer, wherein nodes of the hidden layer incorporate serialized activation functions, wherein the serialized activation functions for each of the nodes include a sigmoid function and a Beta function, wherein the sigmoid function is applied to weighted outputs from nodes of a previous layer of the neural network, wherein the Beta function is applied to a conductance hyper-parameter and respective outputs of the sigmoid function, and wherein outputs of the Beta function are provided to a subsequent layer of the neural network. One or more processors may be configured to train the neural network until the outputs of the sigmoid function for the nodes of the hidden layer are substantially binary. ¶0106 The solid arrows between pairs of nodes represent connections through which intermediate values flow, and are each associated with a respective weight (e.g., any real number) that is applied to the respective intermediate value. Each node performs an operation on its input values and their associated weights to produce an output value. In some cases this operation may involve a dot-product sum of the products of each input value and associated weight. An activation function may be applied to the result of the dot-product sum to produce the output value. Other operations are possible. ¶0149 In order to show this, a probabilistic model for the stochastic activation function is developed. The probability density functions of the random variables S and Y are obtained, corresponding to the output of the intermediate sigmoid and the final output of stochastic activation, respectively. These functions will have parameters that allow controlling how close the variables are to binary. Using the derived probability density functions, an expression of D.sub.KL(S∥Y) is obtained that proves D.sub.KL(S∥Y) reaches a minimum when the values of s.sub.i are binary. An empirical analysis illustrates how this activation function behaves in practice, in terms of binarization properties relative to hyper-parameters (namely κ) as well as the accuracy versus representational fidelity trade-off.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the efficient binary representations from neural networks of McDonald with the system for data analysis of Trygg for the following reasons:
(1) a finding that there was some teaching, suggestion, or motivation, either in the references themselves or in the knowledge generally available to one of ordinary skill in the art, to modify the reference or to combine reference teachings, e.g. Trygg ¶0005 teaches that it is desirable to obtain general applicability for real-time systems in cases where making multiple inferences is impractical and/or for existing systems that do not fit into the constraints;
(2) a finding that there was reasonable expectation of success since the only difference between the claimed invention and the prior art being the lack of actual combination of the elements in a single prior art reference, e.g. Trygg teaches a deep neural network is provided for processing images and at least a part of a training dataset used for training the deep neural network, and McDonald teaches persistent storage that contains a representation of a neural network including an input layer, and output layer, and a hidden layer, wherein nodes of the hidden layer incorporate serialized activation functions; and
(3) whatever additional findings based on the Graham factual inquiries may be necessary, in view of the facts of the case under consideration, to explain a conclusion of obviousness, e.g. Trygg at least the above cited paragraphs, and McDonald at least the inclusively cited paragraphs.
Therefore, it would be obvious to one skilled in the art at the time of the invention to combine the efficient binary representations from neural networks of McDonald with the system for data analysis of Trygg. The rationale to support a conclusion that the claim would have been obvious is that "a person of ordinary skill in the art would have been motivated to combine the prior art to achieve the claimed invention and whether there would have been a reasonable expectation of success in doing so." DyStar Textilfarben GmbH & Co. Deutschland KG v. C.H. Patrick Co., 464 F.3d 1356, 1360, 80 USPQ2d 1641, 1645 (Fed. Cir. 2006). See MPEP 2143(G).
Trygg teaches:
Claim 14. The method of Claim 13, wherein: each activation function of the plurality of activation functions is associated with a range of values for the intermediate output; and the selector is configured to select an activation function associated with a range of values in which the value of the intermediate output lies from the plurality of activation functions (¶0146 In order to provide a possibility to detect outliers during prediction in a similar way as PLS, a feed-forward neural network may be considered as a series of non-linear transformations. In other words, an activation vector a.sub.i of an observation x (input data) from layer i (=1, 2, 3, 4, . . . ) of the given network may be given by the nested series of transformations as follows: a.sub.if.sub.i(W.sub.if.sub.i−1(W.sub.i−1f.sub.i−2( . . . f.sub.1(W.sub.1x)))) where f.sub.k (k=1, 2, . . . , i) may be activation functions and W.sub.k (k=1, 2, . . . , i) may be weight matrices. The activation vector a.sub.i of the observation x may be considered as an intermediate output from the layer i of the given network and may include element values corresponding to outputs from respective nodes of the layer i when the observation x is input to the given network.).
Trygg teaches:
Claim 15. The method of Claim 13, wherein the selector is configured to select the at least one activation function to apply to the intermediate output based on a partitioning of the intermediate output into a plurality of segments, and wherein each segment of the plurality of segments is processed using a different activation function from the plurality of activation functions (¶0207 The methods and systems as described herein may be applied to phase microscopy image analysis. The task of identifying nuclei from phase contrast images is challenging and typically relies on fluorescent markers binding to nuclei to provide an extra image channel highlighting locations of nuclei. By applying deep CNN:s, a model may be generated and trained that segments background from cells as well as identifies nuclei of cells from phase contrast images without using fluorescence labels. If the CNN is trained on phase contrast images from multiple cell types, the CNN may also identify new previously unseen cell types.).
Trygg teaches:
Claim 16. The method of Claim 13, wherein the selector is configured to select the at least one activation function to apply to the intermediate output based on one or more statistical measurements associated with the intermediate output (¶0022 In the present disclosure, the term “latent variable model” may be a statistical model that relates a set of observable variables to a set of latent variables. In various embodiments and examples described herein, a (first or second) “set of intermediate output values” may be considered as the set of observable variables for the latent variable model. Further, in various embodiments and examples described herein, a “set of projected values” may be considered as the set of latent variables for the latent variable model).
Trygg teaches:
Claim 17. The method of Claim 13, wherein training the neural network comprises: generating one or more statistical measurements associated with the intermediate output; selecting the at least one activation function based on the generated one or more statistical measurements; and backpropagating a result of applying the selected at least one activation function to the intermediate output through at least a portion of the neural network (¶0146 The activation vector a.sub.i of the observation x may be considered as an intermediate output from the layer i of the given network and may include element values corresponding to outputs from respective nodes of the layer i when the observation x is input to the given network. Each of these activations a.sub.k may provide a feature representation of the input data. Although the weight matrices may be commonly obtained by supervised training using back-propagation, the activations may simply provide transformed, or pre-processed, representations of the input data.).
As per claims 18-29 and 30, the system and system tracks the method and method of claims 1-12 and 13, respectively, resulting in substantially similar limitations. The same cited prior art and rationale of claims 1-12 and 13 are applied to claims 18-29 and 30, respectively. Trygg discloses that the embodiment may be found as a system (Figs. 1-3).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KURTIS GILLS whose telephone number is (571)270-3315. The examiner can normally be reached on M-F 8-5 PM.
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, Jerry O’Connor can be reached on 5712726787. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/KURTIS GILLS/Primary Examiner, Art Unit 3624
Prosecution Insights