DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 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 1-7 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
In regards to claim 1,
Step 1: This part of the eligibility analysis evaluates whether the claim(s) falls within any statutory category. See MPEP 2106.03:
The claim directs to a statutory category – machine.
Step 2A – Prong 1: Judicial Exception Recited?
MPEP 2106.04(a)(2)(II) “The mathematical concepts grouping is defined as mathematical relationships, mathematical formulas or equations, and mathematical calculations.”
Further, the MPEP recites “It is important to note that a mathematical concept need not be expressed in mathematical symbols, because "[w]ords used in a claim operating on data to solve a problem can serve the same purpose as a formula." In re Grams, 888 F.2d 835, 837 and n.1, 12 USPQ2d 1824, 1826 and n.1 (Fed. Cir. 1989). See, e.g., SAP America, Inc. v. InvestPic, LLC, 898 F.3d 1161, 1163, 127 USPQ2d 1597, 1599 (Fed. Cir. 2018) (holding that claims to a ‘‘series of mathematical calculations based on selected information’’ are directed to abstract ideas); Digitech Image Techs., LLC v. Elecs. for Imaging, Inc., 758 F.3d 1344, 1350, 111 USPQ2d 1717, 1721 (Fed. Cir. 2014) (holding that claims to a ‘‘process of organizing information through mathematical correlations’’ are directed to an abstract idea); and Bancorp Servs., LLC v. Sun Life Assurance Co. of Can. (U.S.), 687 F.3d 1266, 1280, 103 USPQ2d 1425, 1434 (Fed. Cir. 2012) (identifying the concept of ‘‘managing a stable value protected life insurance policy by performing calculations and manipulating the results’’ as an abstract idea).”
Yes, the claim recites a mathematical concept, specifically:
calculate a recognition loss using the recognition result, a mixing matrix calculated based on the dataset for learning, and the weak label
This limitation encompasses a mathematical calculation in light of the specification (see para. [0067-0069])
wherein the weak label probability distribution is a probability distribution according to the weak label conditioned by a true correct answer class to which the recognition object data belongs and is reconfigurable
This limitation encompasses a mathematical calculation. See MPEP 2106.04(a)(2)(I)(C.)i. performing a resampled statistical analysis to generate a resampled distribution, SAP America, Inc. v. InvestPic, LLC, 898 F.3d 1161, 1163-65, 127 USPQ2d 1597, 1598-1600 (Fed. Cir. 2018), modifying SAP America, Inc. v. InvestPic, LLC, 890 F.3d 1016, 126 USPQ2d 1638 (Fed. Cir. 2018);
wherein calculating the recognition loss includes converting the recognition result into a conjugate vector, calculating a product of the conjugate vector and the mixing matrix, calculating a normalization term from the conjugate vector, and calculating a sum of the product and the normalization term and outputting the calculated sum as the recognition loss
These limitations encompass a series of mathematical calculations (conjugate, product, normalization (which is multiplying/dividing by some scalar), and sum).
Therefore, the claim recites a mathematical concept.
MPEP 2106.04(a)(2)(I) “Accordingly, the "mental processes" abstract idea grouping is defined as concepts performed in the human mind, and examples of mental processes include observations, evaluations, judgments, and opinions.”
Further, the MPEP recites “The courts do not distinguish between mental processes that are performed entirely in the human mind and mental processes that require a human to use a physical aid (e.g., pen and paper or a slide rule) to perform the claim limitation.”
Yes, the claim recites a mental process, specifically:
output a recognition result with respect to recognition object data in a dataset for learning that is a set of pairs of the recognition object data and a weak label attached to the recognition object data
This limitation encompasses providing an opinion (recognition result) based on an evaluation of a pair of recognition object data and a corresponding weak label. For example, provided an image of a car (the car is the object in the image) and a weak label of ‘bicycle’, one of ordinary art would be able to provide an opinion in the form of a score (numerical value).
wherein outputting the recognition result further includes performing learning using the recognition loss
This limitation encompasses an evaluation of the mathematical calculations (recognition loss) to provide an informed opinion (recognition result) wherein an observation and evaluation of the recognition loss would be learning.
Therefore, the claim recites a mental process.
Step 2A – Prong 2: Integrated into a Practical Solution?
MPEP 2106.05(f) Mere Instructions To Apply An Exception has found simply adding a general purpose computer or computer components after the fact to an abstract idea (e.g., a fundamental economic practice or mathematical equation) does not integrate a judicial exception into a practical application or provide significantly more. The following steps are mere instructions to apply:
a memory configured to store instructions
a processor configured to execute the instructions
MPEP 2106.05(g) Insignificant Extra-Solution Activity has found mere data gathering to be insignificant extra-solution activity. The following steps are insignificant extra-solution activities:
Mere data gathering:
wherein the dataset for learning has a weak label probability distribution,
(the dataset is provided for the intended use of learning wherein an attribute of the provided dataset is a weak label probability distribution)
The additional elements have been considered both individually and as an ordered combination in to determine whether they integrate the exception into a practical application. Therefore, no meaningful limits are imposed on practicing the abstract idea.
The claim is directed to the abstract idea.
Step 2B: Claim provides an Inventive Concept?
No, as discussed with respect to Step 2A, the additional limitation is mere data gathering (Insignificant Extra-Solution Activity) and a generic device do not impose any meaningful limits on practicing the abstract idea and therefore the claim does not provide an inventive concept in Step 2B.
The claim recites receiving data by generic device.
This has been determined to be insignificant extra-solution activity as found in MPEP § 2106.05(d)(II)(i): Receiving or transmitting data over a network, e.g., using the Internet to
gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary
computer to forward information); TLI Communications LLC v. AV Auto. LLC, 823 F.3d 607,
610, 118 USPQ2d 1744, 1745 (Fed. Cir. 2016) (using a telephone for image transmission); OIP
Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015)
(sending messages over a network); buy SAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112
USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network);
but see DDR Holdings, LLC v. Hotels.com, L.P., 773 F.3d 1245, 1258, 113 USPQ2d 1097, 1106
(Fed. Cir. 2014) ("Unlike the claims in Ultramercial, the claims at issue here specify how
interactions with the Internet are manipulated to yield a desired result‐‐a result that overrides
the routine and conventional sequence of events ordinarily triggered by the click of a hyperlink."
(emphasis added)).
The additional elements have been considered both individually and as an ordered
combination in the significantly more consideration.
The claim is ineligible.
In regards to claim 2,
Step 1: This part of the eligibility analysis evaluates whether the claim(s) falls within any statutory category. See MPEP 2106.03:
The claim directs to a statutory category – machine.
Step 2A Prong 1: The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
wherein the processor is configured to execute the instructions to calculate the mixing matrix based on the dataset for learning
This limitation directs to a mental process that can be performed in the human mind, by a human using pen and paper, or using a computer as a tool to perform the concept and encompasses providing an opinion (mixing matrix) based on an evaluation of the dataset. See MPEP 2106.04(a)(2)(III)
Step 2A Prong 2: The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
Step 2B: The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
In regards to claim 3,
Step 1: This part of the eligibility analysis evaluates whether the claim(s) falls within any statutory category. See MPEP 2106.03:
The claim directs to a statutory category – machine.
Step 2A Prong 1: The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
Step 2A Prong 2: The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
a learning model
This limitation directs to merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f)
update parameters based on the recognition loss, wherein outputting the recognition result further includes setting the parameters in a learning model
This limitation merely indicates a field of use or technological environment in which the judicial exception is performed. This type of limitation merely confines the use of the abstract idea (recognition loss) to a particular technological environment (learning model) and thus fails to add an inventive concept to the claims. See MPEP § 2106.05(h)
Step 2B: The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
In regards to claim 4,
Step 1: This part of the eligibility analysis evaluates whether the claim(s) falls within any statutory category. See MPEP 2106.03:
The claim directs to a statutory category – machine.
Step 2A Prong 1: The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
Step 2A Prong 2: The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
supply the dataset for learning
This limitation directs to mere data gathering of insignificant extra-solution activity. See MPEP § 2106.05(g)
Step 2B: The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
supply the dataset for learning
This limitation directs to mere data gathering of insignificant extra-solution activity. See MPEP § 2106.05(g)
This has been determined to be insignificant extra-solution activity as found in MPEP § 2106.05(d)(II)(i): Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); TLI Communications LLC v. AV Auto. LLC, 823 F.3d 607, 610, 118 USPQ2d 1744, 1745 (Fed. Cir. 2016) (using a telephone for image transmission); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network); buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network); but see DDR Holdings, LLC v. Hotels.com, L.P., 773 F.3d 1245, 1258, 113 USPQ2d 1097, 1106 (Fed. Cir. 2014) ("Unlike the claims in Ultramercial, the claims at issue here specify how interactions with the Internet are manipulated to yield a desired result‐‐a result that overrides the routine and conventional sequence of events ordinarily triggered by the click of a hyperlink." (emphasis added));
In regards to claim 5,
Step 1: This part of the eligibility analysis evaluates whether the claim(s) falls within any statutory category. See MPEP 2106.03:
The claim directs to a statutory category – machine.
Step 2A Prong 1: The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
Step 2A Prong 2: The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
wherein the dataset for learning is an expert dataset or a PU dataset
This limitation merely indicates a field of use or technological environment in which the judicial exception is performed. This type of limitation merely confines the use of the abstract idea to a particular technological environment (expert or PU) and thus fails to add an inventive concept to the claims. See MPEP § 2106.05(h)
Step 2B: The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
wherein the dataset for learning is an expert dataset or a PU dataset
This limitation merely indicates a field of use or technological environment in which the judicial exception is performed. This type of limitation merely confines the use of the abstract idea to a particular technological environment (expert or PU) and thus fails to add an inventive concept to the claims. See MPEP § 2106.05(h)
Claim 6 (method) and Claim 7 (manufacture) are rejected on the same grounds under 35 U.S.C. 101 as claim 1 as they are substantially similar, respectively, Mutatis mutandis.
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.
Claim(s) 1-4 and 6-7 are rejected under 35 U.S.C. 103 as being unpatentable over Patrini, Giorgio, et al. "Loss factorization, weakly supervised learning and label noise robustness." arXiv preprint arXiv:1602.02450 (2016). (“Patrini-2016”) in view of Patrini, Giorgio, et al. "Making Deep Neural Networks Robust to Label Noise: a Loss Correction Approach." arXiv e-prints (2017): arXiv-1609. (“Patrini-2017”) in further view of Sun, Shiliang, and John Shawe-Taylor. "Sparse semi-supervised learning using conjugate functions." Journal of Machine Learning Research 11 (2010): 2423-2455. (“Sun”)
In regards to claim 1,
Patrini-2016 teaches A learning device comprising: a memory configured to store instructions; and a processor configured to execute the instructions to: output a recognition result with respect to recognition object data in a dataset for learning that is a set of pairs of the recognition object data and a weak label attached to the recognition object data;
(Patrini-2016, Section 2.1, “We denote vectors in bold as x and 1{p} the indicator of p being TRUE. We define [m] .= {1, . . . , m} and [x]+ .= max(0, x). In binary classification, a learning sample S = {(xi , yi), i ∈ [m]} is a sequence of (observation, label) pairs [a set of pairs of the recognition object data and a weak label attached to the recognition object data], the examples, drawn from an unknown distribution D over X×Y, with X ⊆ R d and Y = {−1, 1}. Expectation (or average) over (x, y) ∼ D (S) is denoted as ED (ES). Given a hypothesis (or model) h ∈ H, h : X → R, a loss is a function l : Y×R → R. A loss gives a penalty l(y, h(x)) when predicting the value h(x) [output a recognition result ie predicted value] and the observed label is y.”)
Patrini-2016 teaches and calculate a recognition loss using the recognition result, a mixing matrix calculated based on the dataset for learning, and the weak label,
(Patrini-2016, Section 2.1, “Given a hypothesis (or model) h ∈ H, h : X → R, a loss is a function l : Y×R → R. A loss gives a penalty l(y, h(x)) when predicting the value h(x) [calculate a recognition loss using the recognition result] and the observed label is y [the weak label]. We consider margin losses, i.e. l(y, h(x)) = l(yh(x))”)
Patrini-2016 teaches wherein the dataset for learning has a weak label probability distribution, wherein the weak label probability distribution is a probability distribution according to the weak label conditioned by a true correct answer class to which the recognition object data belongs and is reconfigurable,
(Patrini-2016, Section 2.1, “Finally, we discuss the meaning of WSL — and in particular of weakly supervised binary classification. The difference with the above is at training time: we learn on a sample ˜S drawn from a noisy distribution D˜ that may flip, aggregate or suppress labels [a weak label probability distribution, wherein the weak label probability distribution is a probability distribution according to the weak label conditioned by a true correct answer class ie the labels are altered which alters their probability distribution from the ‘true correct answer’ to which the recognition object data belongs and is reconfigurable wherein reconfigurable is the iterative sampling of the noisy distribution], while observations are the same. Still, the purpose of learning is unchanged: to minimize the true risk.”)
Patrini-2016 teaches calculating a normalization term from the conjugate vector,
(Patrini-2016, “The Factorization Theorem 3 generalizes Patrini et al. (2014, Lemma 1) that works for symmetric proper losses (SPLs), e.g. logistic, square and Matsushita losses. Given a permissible generator l (Kearns & Mansour, 1996; Nock & Nielsen, 2009), i.e.
PNG
media_image1.png
19
128
media_image1.png
Greyscale
is strongly convex, differentiable and symmetric with respect to 1/2, SPLs are defined as
PNG
media_image2.png
21
165
media_image2.png
Greyscale
, where
PNG
media_image3.png
17
17
media_image3.png
Greyscale
is the convex conjugate of
PNG
media_image4.png
18
11
media_image4.png
Greyscale
. Then, since
PNG
media_image5.png
21
145
media_image5.png
Greyscale
PNG
media_image6.png
47
364
media_image6.png
Greyscale
[ a normalization term ie loss_o from the conjugate vector; wherein the conjugate vector is provided from Sun]”)
Patrini-2016 teaches and calculating a sum of the product and the normalization term and outputting the calculated sum as the recognition loss, and wherein outputting the recognition result further includes performing learning using the recognition loss.
(Patrini-2016, Section 1, “We introduce linear-odd losses (LOLs), a definition not demanding smoothness or convexity and embracing many losses of practical interest, e.g. logistic and square. They decompose into an even and an odd function (Figure 1) as
PNG
media_image7.png
30
168
media_image7.png
Greyscale
[product ie loss_e (wherein loss_e is provided by Patrini-2017 and Sun) and the normalization term ie loss_o (wherein loss_o is obtained from in part of the loss function provided by Patrini-2017 and Sun)]
Where
PNG
media_image8.png
21
182
media_image8.png
Greyscale
is label-independent.”; wherein the loss function is used for learning)
However, Patrini-2016 does not explicitly teach recognition object data calculate a recognition loss using the recognition result, a mixing matrix calculated based on the dataset for learning…
calculating a product of the conjugate vector and the mixing matrix,
wherein calculating the recognition loss includes converting the recognition result into a conjugate vector
Patrini-2017 teaches recognition object data
(Patrini-2017, Section 1, “We apply our loss corrections to image recognition on MNIST [recognition object data; wherein MNIST is recognizing digits in an image], CIFAR-10, CIFAR-100 and sentiment analysis on IMDB; we simulate corruption by artificially injecting noise on the training labels.”)
Patrini-2017 teaches calculate a recognition loss using the recognition result, a mixing matrix calculated based on the dataset for learning…
calculating a product of the conjugate vector and the mixing matrix,
(Patrini-2017, Section 4.2, “Alternatively, we can correct the model predictions. Following [39], we start by observing that a neural network learned with no loss correction would result in a predictor for noisy labels pˆ(y˜|x). We can make explicit the dependency on T [mixing matrix]. For instance, with cross-entropy we have: [a recognition loss]
PNG
media_image9.png
124
455
media_image9.png
Greyscale
[ a product of the conjugate vector ie the model prediction as a conjugate and the mixing matrix ie matrix T; wherein the product of the prediction (as the conjugate provided by Sun) and the matrix is a loss (in this case, cross-entropy loss)]
or in matrix form
PNG
media_image10.png
23
251
media_image10.png
Greyscale
This loss compares the noisy label y˜ to averaged noisy prediction corrupted by T. We call this procedure “forward” correction.”)
However, Patrini-2017 does not explicitly teach wherein calculating the recognition loss includes converting the recognition result into a conjugate vector
Sun teaches wherein calculating the recognition loss includes converting the recognition result into a conjugate vector,
(Sun, Section 2.2, “Define conjugate vector z = [z1,...,zℓ+u] ⊤ with entries being conjugate variables. The Fenchel-Legendre conjugate (which is also often called convex conjugate or conjugate function) f ∗ ε (z [recognition result]) is
PNG
media_image11.png
48
492
media_image11.png
Greyscale
[into a conjugate vector]
The domain of the conjugate function consists of z ∈ R ℓ+u for which the supremum is finite (i.e., bounded above) (Boyd and Vandenberghe, 2004).”)
Patrini-2016 and Patrini-2017 are both considered to be analogous to the claimed invention because they are in the same field of improving loss functions in supervised learning. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Patrini-2016 to incorporate the teachings of Patrini-2017 in order to a forward loss correction procedure along with noise estimation to build an end-to-end framework that demonstrates noise robustness. (Patrini-2017, Abstract, “We present a theoretically grounded approach to train deep neural networks, including recurrent networks, subject to class-dependent label noise. We propose two procedures for loss correction that are agnostic to both application domain and network architecture. They simply amount to at most a matrix inversion and multiplication, provided that we know the probability of each class being corrupted into another. We further show how one can estimate these probabilities, adapting a recent technique for noise estimation to the multi-class setting, and thus providing an end-to-end framework. Extensive experiments on MNIST, IMDB, CIFAR10, CIFAR-100 and a large scale dataset of clothing images employing a diversity of architectures — stacking dense, convolutional, pooling, dropout, batch normalization, word embedding, LSTM and residual layers — demonstrate the noise robustness of our proposals. Incidentally, we also prove that, when ReLU is the only non-linearity, the loss curvature is immune to class-dependent label noise.”)
Sun is considered to be analogous to the claimed invention because they are in the same field of supervised learning using conjugate functions. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Patrini-2016 and Patrini-2017 to incorporate the teachings of Sun in order to provide a conjugate function to the proper loss as to bound the loss from above (Sun, Section 2.2, “The domain of the conjugate function consists of z ∈ R ℓ+u for which the supremum is finite (i.e., bounded above)”)
In regards to claim 2,
Patrini-2016 and Patrini-2017 and Sun teaches The learning device according to claim 1,
Patrini-2017 teaches wherein the processor is configured to execute the instructions to calculate the mixing matrix based on the dataset for learning.
(Patrini-2017, Section 1, “We introduce two alternative procedures for loss correction, provided that we know a stochastic matrix T summarizing the probability of one class being flipped into another under noise [calculate the mixing matrix based on the dataset for learning]. The first procedure, a multiclass extension of [28, 30] applied to neural networks, is called “backward” as it multiplies the loss by T −1 . The second, inspired by [39], is named “forward” as it multiplies the network predictions by T.”)
(Patrini-2017, Section 4.3, “In particular, let X0 be any set of features vectors. This can be the training set itself, but not necessarily: we do not require this sample to have any label at all and therefore any unlabeled sample from the same distributions can be used as well. We can approximate T with two steps:
PNG
media_image12.png
71
381
media_image12.png
Greyscale
”)
Patrini-2017 further teaches an example of matrix T used for MNIST:
(Patrini-2017, Section 5.1, “We artificially corrupt labels by a parametric matrix T. The rationale is to mimic some of the structure of real mistakes for similar classes, e.g. CAT → DOG. Transitions are parameterized by N ∈ [0, 1] such that ground truth and wrong class have probability respectively of 1 − N, N. An example of T used for MNIST with N = 0.7 is on the left:
PNG
media_image13.png
220
782
media_image13.png
Greyscale
”)
In regards to claim 3,
Patrini-2016 and Patrini-2017 and Sun teaches The learning device according to claim 2,
Patrini-2017 teaches wherein the processor is configured to execute the instructions to update parameters based on the recognition loss, wherein outputting the recognition result further includes setting the parameters in a learning model.
(Patrini-2017, Section 3, “An n-layer neural network2 [learning model] comprises a transformation h: X → Rc, where h = (h(n) ◦ h(n−1) ◦ · · · ◦ h(1)) is the composition of a number of intermediate transformations —the layers — defined by:
PNG
media_image14.png
55
305
media_image14.png
Greyscale
where W(i) ∈ R d (i)×d (i−1) and b (i) ∈ R d (i) are parameters to be estimated3 [update parameters], and σ is any activation function that acts coordinate-wise, such as the ReLU σ(x)i = max(0, xi). Observe that the final layer applies a linear projection, unlike all preceding layers. To simplify notation, we write:
PNG
media_image15.png
31
210
media_image15.png
Greyscale
with the base case x (0) .= x, so that e.g. x (1) is exactly the representation in the first layer. The coordinates of h(x) represent the relative weights that the model assigns to each class i = 1, . . . , c to be predicted. The predicted label is thus given by arg maxi∈[c] hi(x). In the training phase, the output of the final layer is contrasted with the true label y via two steps. First, h(·) passes through the softmax function e hi(x)/ Pc k=1 e hk(x) . The softmax output can be interpreted as a vector approximating the class-conditional probabilities p(y|x); we denote it by pˆ(y|x) ∈ ∆c−1 . Next, we measure the discrepancy between label y = e i and network output by a loss function ` : Y × ∆c−1 → R, for example by means of cross-entropy: [based on the recognition loss wherein this cross-entropy loss can be replaced by the recognition loss given by Patrini-2016 in view of Patrini-2017 and Sun, wherein outputting the recognition result further includes setting the parameters in a learning model]
PNG
media_image16.png
47
403
media_image16.png
Greyscale
”)
In regards to claim 4,
Patrini-2016 and Patrini-2017 and Sun teaches The learning device according to claim 1,
Patrini-2017 teaches wherein the processor is configured to execute the instructions to supply the dataset for learning.
(Patrini, Section 1, “We apply our loss corrections to image recognition on MNIST…”)
Claims 6 and 7 are rejected on the same grounds under 35 U.S.C. 103 as claim 1
Claim(s) 5 is rejected under 35 U.S.C. 103 as being unpatentable over Patrini-2016 in view of Patrini-2017 and Sun in further view of du Plessis, M C., Niu, G., and Sugiyama, M. Convex formulation for learning from positive and unlabeled data. In 32 th ICML, pp. 1386–1394, 2015. (“Plessis”)
In regards to claim 5,
Patrini-2016 and Patrini-2017 and Sun teaches The learning device according to claim 1,
Plessis teaches wherein the dataset for learning is an expert dataset or a PU dataset.
(Plessis, Section 1, “Let us consider the problem of learning a classifier only from positive and unlabeled data. This problem, which refer to as PU classification, arises in various practical situations under different guises. For example:
The goal of identification is to find samples in an unlabeled dataset that are similar to the samples provided by a user. Such a situation occurs, e.g., in automatic face tagging: a user provides a set of images of himself, and the task is to automatically tag photos in the user’s photo album.”)
Plessis is considered to be analogous to the claimed invention because they are in the same field of learning from positive and unlabeled data. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Patrini-2016 in view of Patrini-2017 and Sun to incorporate the teachings of Plessis in order to provide a positive and unlabeled dataset as the convex loss function provided by Patrini-2016 in view of Patrini-2017 and Sun would solve the bias problem and derive a tractable loss. (Plessis, Abstract, “We discuss binary classification from only positive and unlabeled data (PU classification), which is conceivable in various real-world machine learning problems. Since unlabeled data consists of both positive and negative data, simply separating positive and unlabeled data yields a biased solution. Recently, it was shown that the bias can be canceled by using a particular non-convex loss such as the ramp loss. However, classifier training with a non-convex loss is not straightforward in practice. In this paper, we discuss a convex formulation for PU classification that can still cancel the bias. The key idea is to use different loss functions for positive and unlabeled samples. However, in this setup, the hinge loss is not permissible. As an alternative, we propose the double hinge loss. Theoretically, we prove that the estimators converge to the optimal solutions at the optimal parametric rate. Experimentally, we demonstrate that PU classification with the double hinge loss performs as accurate as the non-convex method, with a much lower computational cost.”) Also, see Patrini-2016 for a direct reference to Plessis: (Patrini-2016, “The linear-odd losses of du Plessis et al. (2015) This work shows that a linear-odd condition on a convex l allows one to derive a tractable, i.e. still convex, loss for learning with positive and unlabelled data. The approach is similar to ours as it isolates a label-free term in the loss, with the goal of leveraging on the unlabelled examples too.”)
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
NPL: Kanamori, Takafumi, Akiko Takeda, and Taiji Suzuki. "Conjugate relation between loss functions and uncertainty sets in classification problems." The Journal of Machine Learning Research 14.1 (2013): 1461-1504.
US Pub No. US20080247598A1: Movellan et al. teachers Weak hypothesis generation apparatus and method, learning apparatus and method, detection apparatus and method, facial expression learning apparatus and method, facial expression recognition apparatus and method, and robot apparatus
NPL: Zhang, Yivan, Nontawat Charoenphakdee, and Masashi Sugiyama. "Learning from indirect observations." arXiv preprint arXiv:1910.04394 (2019).
NPL: Kong, Yuqing, and Grant Schoenebeck. "Water from two rocks: Maximizing the mutual information." Proceedings of the 2018 ACM Conference on Economics and Computation. 2018.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JASMINE THAI whose telephone number is (703)756-5904. The examiner can normally be reached M-F 8-4.
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, Michael Huntley can be reached at (303) 297-4307. 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.
/J.T.T./Examiner, Art Unit 2129
/MICHAEL J HUNTLEY/Supervisory Patent Examiner, Art Unit 2129