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 .
Status of Claims
Claims 1-16 are pending and examined herein.
Claims 3-10 and 13-15 are objected to.
Claims 1-16 are rejected under 35 U.S.C. 112(b).
Claims 12-14 are rejected under 35 U.S.C. 101 for being directed to non-statutory subject matter.
Claims 1-16 are rejected under 35 U.S.C. 101 for being directed to an abstract idea.
Claims 1, 3, and 12-16 are rejected under 35 U.S.C. 102.
Claim Objections
Claim 3-10 and 13-15 are objected to because of the following informalities:
Claims 3-10 and 13-15 appear to be missing a comma after the preamble of each claim.
Claim 5 appears to be missing a dash in “no longer-desired data”.
Claim 5 should likely have paragraph breaks after the semicolons.
Claim 6 has an inappropriately capitalized N in “aN”.
Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-16 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 1 recites the limitations
"the training data" in the second paragraph
“the influence of which” in the second paragraph
“the no-longer-desired data” in the third paragraph
“the difference between …” in the fourth paragraph
“the coefficients of the intermediate trained model”
“the coefficients of the original trained model”
There is insufficient antecedent basis for these limitations in the claim. Each limitation will be interpreted as if they started with “a” instead of “the”, except “the no-longer-desired data” which will be interpreted to refer to “the samples the influence of which on the modified model is no longer desired”.
Claim 1 recites the limitation “a modified trained model” in the preamble of the claim and the last paragraph of the claim. It is uncertain as to if the “a modified trained model” in the last paragraph of the claim refers to the “a modified trained model” in the preamble or if it is another modified trained model.
Dependent claims 2-11 fail to resolve the issues and are rejected with the same rationales.
Claims 6 and 7 both recite the limitation “the coefficients of the model” twice. This is indefinite as it is uncertain as to which model “the model” refers to. For purposes of examination, the limitation will be interpreted as “the coefficients of the intermediate model and the coefficients of the original model”.
Dependent claims 7-9 fail to resolve the issue and are rejected with the same rationale.
Claim 2 recites the limitation “a modified trained model” in the last paragraph of the claim. This is indefinite as it is uncertain as to if this “a modified trained model” is the same as “a modified trained model” as in claim 1. For purposes of examination, this will be interpreted as “a second modified trained model”.
Dependent claims 4-11 fail to resolve the issue and are rejected with the same rationale.
Claim 3 recites the limitation “the differences”. It is unclear if this is intended to refer to the single difference introduced in claim 1 and if so, what the other difference is, as the plural form of the words implies another difference. Therefore, the claim is rendered indefinite. For purposes of examination, this limitation will be interpreted as “the difference”.
Claim 11 recites the limitations “multiple intermediate models” and “a single intermediate model”. This is indefinite as it is unclear how these intermediate models relate to the intermediate trained model in claim 1. For purposes of examination, “a single intermediate model” will be interpreted to be “the intermediate trained model” in claim 1.
Claim 12 recites the limitation “a data storage device” in both the second and last paragraphs of the claims. It is unclear whether the second “a data storage device” refers to the first “a data storage device” or if it is another data storage device. For purposes of examination, the second “a data storage device” will be interpreted as the same “a data storage device” as in the second paragraph of the claim.
Claims 12 and 16 recite the limitation “some training data samples” in the fourth paragraph of the claim. This is indefinite as it is unclear how “some training data samples” relates to the “training data samples” introduced in the second paragraph of the claim. For purposes of examination, this limitation will be interpreted as “some of the training data samples”.
Claims 12 and 16 recite the limitation “a trained machine learning model” in the fifth paragraph of each claim. This is indefinite as it is unclear whether “a trained machine learning model” refers to the “a machine learning model” that is trained in the third paragraph of the claim or if it is another model. For purposes of examination, this limitation will be interpreted as “a second trained machine learning model”.
Dependent claims 13-15 fail to resolve the issues and are rejected with the same rationales.
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 12-14 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter.
The claim(s) does/do not fall within at least one of the four categories of patent eligible subject matter because they recite a system without reciting any hardware, and are therefore directed to software per se. A “computing device” and a “data storage device”, in their broadest reasonable interpretation, includes software. A “cloud-based computing device” also includes software. It is recommended that “computing device” be amended to “processor” and “data storage device” to “memory”. Claim 15 recites a “GPU” and is therefore directed to a statutory category.
Claims 1-16 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
MPEP § 2109(III) sets out steps for evaluating whether a claim is drawn to patent-eligible subject
matter. The analysis of claims 1-16, in accordance with these steps, follows.
Step 1 Analysis:
Step 1 is to determine whether the claim is directed to a statutory category (process, machine,
manufacture, or composition of matter. Claims 1-11 are directed to a process, claims 12-15, if amended, would be directed to a machine, and claim 16 is directed to an article of manufacture. Though claims 12-15 are not directed to a statutory category, for purposes of examination, they will be treated as if they were directed to a statutory category in this section. Claims 1-11 and 16 are directed to statutory categories, and claims 12-15 would be directed to statutory categories if amended, and the analysis proceeds.
Step 2A Prong One, Step 2A Prong Two, and Step 2B Analysis:
Step 2A Prong One asks if the claim recites a judicial exception (abstract idea, law of nature, or natural phenomenon). If the claim recites a judicial exception, analysis proceeds to Step 2A Prong Two, which asks if the claim recites additional elements that integrate the abstract idea into a practical application. If the claim does not integrate the judicial exception, analysis proceeds to Step 2B, which asks if the claim amounts to significantly more than the judicial exception. If the claim does not amount to significantly more than the judicial exception, the claim is not eligible subject matter under 35 U.S.C. 101.
None of the claims represent an improvement to technology.
Regarding claim 1, the following are abstract ideas:
identifying a subset of the training data consisting of samples the influence of which on the modified model is no longer desired; and (One could practically perform identification of unwanted samples in the human mind. This is the mental process of evaluation.)
constructing a representation of the difference between the coefficients of the intermediate trained model and the coefficients of the original trained model; and (Constructing a representation of the difference between coefficients can be practically performed in the human mind. This is a mental process. Additionally, the difference between coefficients is a mathematical formula, which is a mathematical concept.)
calculating a modified trained model by modifying the original trained model by applying to the coefficients of the original trained model a transformation function that depends on the difference between the coefficients of the intermediate trained model and the coefficients of the original trained model. (This is a mathematical calculation, which is a mathematical concept.)
The following claim elements are additional elements which, taken alone or in combination with the other additional elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
A method of modifying an original trained machine learning model into a modified trained model, the method comprising: (This describes generic machine learning; this limitation amounts to mere instructions to apply an exception.)
incrementally further training, using samples of the no-longer-desired data, the original trained model into an intermediate trained model; and (This describes generic machine learning; this limitation amounts to mere instructions to apply an exception.)
Regarding claim 2, the rejection of claim 1 is incorporated herein. The following are abstract ideas:
calculating the difference between the coefficients of the intermediate trained model and the original trained model after each batch is applied individually; and (This is a mathematical calculation, which is a mathematical concept.)
computing the average over all batches of the difference per coefficient; and (This is a mathematical calculation, which is a mathematical concept.)
calculating a modified trained model by modifying the original trained model by applying to the coefficients of the original trained model a transformation function that depends on such averaged difference. (This is a mathematical calculation, which is a mathematical concept.)
The following claim elements are additional elements which, taken alone or in combination with the other additional elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
applying batched subsets of the no-longer-desired data; and (This recites the generic machine learning concept of batch processing; this amounts to mere instructions to apply an exception.)
Regarding claim 3, the rejection of claim 1 is incorporated herein. The following is an abstract idea:
wherein the transformation function is an element-wise subtraction of the differences from the coefficients of the original trained model. (This is a mathematical calculation, which is a mathematical concept.)
Claim 3 does not recite any additional elements.
Regarding claim 4, the rejection of claim 2 is incorporated herein. The following is an abstract idea:
wherein the transformation function is an element-wise subtraction of the averaged differences from the coefficients of the original trained model. (This is a mathematical calculation, which is a mathematical concept.)
Claim 4 does not recite any additional elements.
Regarding claim 5, the rejection of claim 2 is incorporated herein. The following is an abstract idea:
such test model is computed by modifying the original trained model by applying to the coefficients of the original trained model a transformation function that depends on the difference between the coefficients of the intermediate trained model and the original trained model; (Applying a transformation function is a mathematical calculation, which is a mathematical concept.)
The following claim elements are additional elements which, taken alone or in combination with the other additional elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
wherein a test model is created after some portion of the no longer-desired data has been used for the further training of the original trained model; (Creating a model is a generic machine learning process; this amounts to mere instructions to apply an exception.)
the accuracy of the test model is evaluated against one or more chosen sets of data samples; the further training of the original trained model is halted when such computed accuracy reaches a desired value. (Evaluating a model and training until specific accuracy are generic machine learning processes; this amounts to mere instructions to apply an exception.)
Regarding claim 6, the rejection of claim 2 is incorporated herein. The following is an abstract idea:
wherein the transformation function is expressed as aN approximation of a polynomial series expansion in the coefficients of the model and the expansion series term coefficients are calculated analytically by taking the partial derivatives of the transformation function with respect to the coefficients of the model. (The transformation function describes a mathematical equation/formula and it is used for a mathematical calculation, both of which are mathematical concepts.)
Claim 6 does not recite any additional elements.
Regarding claim 7, the rejection of claim 6 is incorporated herein. The following is an abstract idea:
wherein the transformation function is expressed as an approximation of a polynomial series expansion in the coefficients of the model and the expansion series term coefficients are calculated by numerically approximating the partial derivatives of the transformation function with respect to the coefficients of the model. (The transformation function describes a mathematical equation/formula and it is used for a mathematical calculation, both of which are mathematical concepts.)
Claim 7 does not recite any additional elements.
Regarding claim 8, the rejection of claim 6 is incorporated herein. The following is an abstract idea:
in which the polynomial expansion is a Taylor series expansion. (This describes a mathematical equation, which is a mathematical concept.)
Claim 8 does not recite any additional elements.
Regarding claim 9, the rejection of claim 7 is incorporated herein. The following is an abstract idea:
in which the polynomial expansion is a Taylor series expansion. (This describes a mathematical equation, which is a mathematical concept.)
Claim 9 does not recite any additional elements.
Regarding claim 10, the rejection of claim 2 is incorporated herein. The following claim elements are additional elements which, taken alone or in combination with the other additional elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
wherein the intermediate model is created in stages, by training on batches of data and calculating approximate models after each training batch. (This describes the generic machine learning process of batch processing. This is mere instructions to apply an exception.)
Regarding claim 11, the rejection of claim 2 is incorporated herein. The following claim elements are additional elements which, taken alone or in combination with the other additional elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
wherein multiple intermediate models are created simultaneously by training on the same batch of data, and such multiple models are combined to yield a single intermediate model. (This describes the generic machine learning process of ensemble learning. This is mere instructions to apply an exception.)
Regarding claim 12, the following are abstract ideas:
the computing device further programmed to implement a machine learning algorithm that will train a machine learning model by sequentially operating on such training data samples and adjusting the elements of the machine learning model in response to those training data samples; and (This limitation recites an algorithm, and sequential operations, which are mathematical calculations, which are mathematical concepts.)
the computing device further programmed to calculate a representation of a difference between the trained machine learning model and the further-trained machine learning model; and (This recites mathematical calculations, which are mathematical concepts.)
the computing device further programmed to apply a transformation function to the trained machine learning model, such transformation function being at least in part based on such calculated representation of a difference between the trained machine learning model and the further-trained machine learning model; and (This recites mathematical calculations, which are mathematical concepts.)
The following claim elements are additional elements which, taken alone or in combination with the other additional elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
A machine learning system comprising: (This limitation recites generic machine learning; this is mere instructions to apply an exception.)
a computing device programmed to receive from a data storage device a sequence of training data samples; and (Receiving data is a known process in computing. This amounts to mere instructions to apply an exception.)
the computing device further programmed to allow designation of some training data samples as training data samples to be no longer desired; and (This is the insignificant extra-solution activity of selecting a particular type of data to be manipulated. See MPEP § 2106.05(g), ‘Selecting a particular data source or type of data to be manipulated’, ex. i – iv.)
the computing device further programmed to allow additional training of a trained machine learning model, such training using some or all of the data samples designated as no longer desired; and (This recites generic machine learning training; this amounts to mere instructions to apply an exception.)
a data storage device containing the training data samples. (This recites a generic computer component and a generic computer process. This amounts to mere instructions to apply an exception.)
Regarding claim 13, the rejection of claim 12 is incorporated herein. The following claim elements are additional elements which, taken alone or in combination with the other additional elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
wherein the computing device is a general-purpose computing device. (This recites a generic computer component. This amounts to mere instructions to apply an exception.)
Regarding claim 14, the rejection of claim 12 is incorporated herein. The following claim elements are additional elements which, taken alone or in combination with the other additional elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
wherein at least a portion of the computing device is a cloud-based computing device. (This recites a generic computer component. This amounts to mere instructions to apply an exception.)
Regarding claim 15, the rejection of claim 12 is incorporated herein. The following claim elements are additional elements which, taken alone or in combination with the other additional elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
wherein at least a portion of the computing device is a GPU. (This recites a generic computer component. This amounts to mere instructions to apply an exception.)
Regarding claim 16, the following claim elements are additional elements which, taken alone or in combination with the other additional elements, do not integrate the judicial exception into a practical application nor amount to significantly more than the judicial exception:
A non-transitory computer readable medium encoded with computer executable instructions comprising instructions for (This recites generic computer components and processes. This amounts to mere instructions to apply an exception.)
The remainder of claim 16 recites substantially similar subject matter to claim 12 and is rejected with the same rationale, mutatis mutandis.
Claim Rejections - 35 USC § 102
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 the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claim(s) 1, 3, and 12-16 is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Graves (“Amnesiac Machine Learning”, 2020).
Regarding claim 1, Graves teaches
A method of modifying an original trained machine learning model into a modified trained model, the method comprising: (Page 1 states "In this paper, we address the question of efficiently removing learned data from a trained neural network without harming the performance of the network." Removing data from a trained neural network is modifying an original trained machine learning model into a modified trained model.)
identifying a subset of the training data consisting of samples the influence of which on the modified model is no longer desired; and (Page 3 states "Amnesiac unlearning involves selectively undoing the learning steps that involved the sensitive data. During training, the model owner keeps a list of which examples appeared in which batches as well as the parameter updates from each batch. When a data removal request comes, the model owner undoes the parameter updates from only the batches containing sensitive data." The batches containing sensitive data is interpreted as the subset of training data consisting of samples the influence of which on the modified model is no longer desired.)
incrementally further training, using samples of the no-longer-desired data, the original trained model into an intermediate trained model; and (Page 3 states "We can view model training as a series of parameter updates to the initial model parameters (that, in the case of neural networks, are randomly initialized). We begin from initial model parameters
θ
i
n
i
t
i
a
l
. Model
M
is then trained for
E
epochs each consisting of
B
batches, and the parameters are updated after each batch by an amount
∆
θ
e
,
b
." The initial model is interpreted as the original trained model. Model
M
is interpreted as the intermediate trained model. Page 3 further states "During training, we keep a list
S
B
of which batches contained the sensitive data. This can be in the form of an index of batches for each example in the training data, an index of batches for each class, or any other form desired." As the batches that contained the sensitive data is used for training, the samples of the no-longer-desired data are used for incremental training.)
constructing a representation of the difference between the coefficients of the intermediate trained model and the coefficients of the original trained model; and (Page 3 states "Once training is completed, we can produce a protected model
M
'
using amnesiac unlearning as follows: we simply remove the parameter updates from each batch
s
b
∈
S
B
(the list of sensitive data batches) from the learned parameters
θ
M
.
θ
M
'
=
θ
i
n
i
t
i
a
l
+
∑
e
=
1
E
∑
b
=
1
B
∆
θ
e
,
b
-
∑
s
b
=
1
S
B
∆
θ
s
b
=
θ
M
-
∑
s
b
=
1
S
B
∆
θ
s
v
" As
∑
s
b
=
1
S
B
∆
θ
s
b
are the parameter (coefficient) updates from the no-longer-desired data that the intermediate trained model was trained on,
∑
s
b
=
1
S
B
∆
θ
s
b
represents the coefficients of the intermediate model, which is subtracted from the original trained model
θ
i
n
i
t
i
a
l
.)
calculating a modified trained model by modifying the original trained model by applying to the coefficients of the original trained model a transformation function that depends on the difference between the coefficients of the intermediate trained model and the coefficients of the original trained model. (Page 3 states “
θ
M
'
=
θ
i
n
i
t
i
a
l
+
∑
e
=
1
E
∑
b
=
1
B
∆
θ
e
,
b
-
∑
s
b
=
1
S
B
∆
θ
s
b
=
θ
M
-
∑
s
b
=
1
S
B
∆
θ
s
v
”.
θ
M
'
is interpreted as the modified trained model.
θ
i
n
i
t
i
a
l
is the original trained model, and the transformation function is
θ
i
n
i
t
i
a
l
+
∑
e
=
1
E
∑
b
=
1
B
∆
θ
e
,
b
-
∑
s
b
=
1
S
B
∆
θ
s
b
. As explained above, this representation depends on the difference between the coefficients of the models.)
Regarding claim 3, the rejection of claim 1 is incorporated herein. Grave teaches
wherein the transformation function is an element-wise subtraction of the differences from the coefficients of the original trained model. (As stated above,
θ
i
n
i
t
i
a
l
+
∑
e
=
1
E
∑
b
=
1
B
∆
θ
e
,
b
-
∑
s
b
=
1
S
B
∆
θ
s
b
is interpreted as the transformation function, which includes the subtraction of the difference between the coefficients of the models. One of ordinary skill in the art would realize that subtracting parameters from other parameters is an element-wise subtraction.)
Regarding claim 12, Graves teaches
A machine learning system comprising: (Page 1 states "In this paper, we address the question of efficiently removing learned data from a trained neural network without harming the performance of the network." Removing data from a trained neural network is modifying an original trained machine learning model into a modified trained model. Page 4 states "All experiments were conducted on the Amazon Sagemaker platform using an ml.g4dn.xlarge instance with 4 vCPUs, 1 GPU, and 16GB of memory." The system used is interpreted as the machine learning system.)
a computing device programmed to receive from a data storage device a sequence of training data samples; and (Page 3 states “Model
M
is then trained for
E
epochs each consisting of