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 .
Response to Amendment
This Office Action is in response to applicant’s communication filed 7 July 2026, in response to the Office Action mailed 6 April 2026. The applicant’s remarks and any amendments to the claims or specification have been considered, with the results that follow.
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.
Claim(s) 1-5 and 7-20 is/are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claim(s) recite(s) mathematical concepts. This judicial exception is not integrated into a practical application and does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception, as described below.
Step 1 for all claims:
Under the first part of the analysis, claims 13-18 recite a method and claims 1-5, 7-12, 19, and 20 recite a device. Accordingly, these claims fall within the four statutory categories of invention and the analysis proceeds to Step 2A, prongs 1 and 2, and Step 2B, as described below.
As per claim 1:
Under step 2A, prong 1, the claim recites an abstract idea including the following mathematical concept elements:
perform a 1x1 convolution on the input data to obtain a plurality of data groups – a 1x1 convolution is a mathematical calculation.
perform a group convolution on the plurality of data groups – the group convolution is a mathematical calculation.
perform a 1x1 convolution on the intermediate data – a 1x1 convolution is a mathematical calculation.
wherein the first number of convolutional layers equals a sum of the second number of convolutional layers and the third number of decomposed convolutional layers – this is a mathematical formula/calculation (sum).
If a claim, under the broadest reasonable interpretation covers a mathematical relationship between variables or numbers, a numerical formula or equation, or a mathematical calculation, it will be considered as falling within the “mathematical concepts” grouping of abstract ideas. See MPEP § 2106.04(a)(2).
Accordingly, at step 2A, prong one, the claim is directed to an abstract idea.
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
A device for implementing a tensor-train decomposition operation for a respective convolutional layer of a convolutional neural network (CNN) – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
the device being configured to – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
obtain the CNN comprising a first number of convolutional layers, wherein each convolutional layer is associated with a respective first ranking number – this is recited at a high level of generality and amounts to insignificant extra-solution activity as data gathering/storage that is limited to a particular type of data, generally linking the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(g) and (h), and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
and provide a decomposed CNN comprising a second number of convolutional layers and a third number of decomposed convolutional layers based on a training of the CNN – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
receive input data comprising a first number of channels at one of the third number of decomposed convolutional layers – this is recited at a high level of generality and amounts to insignificant extra-solution activity as data gathering/storage that is limited to a particular type of data, generally linking the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(g) and (h), and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
the plurality of data groups comprising a second number of channels – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
to obtain intermediate data comprising a third number of channels – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
to obtain output data comprising a fourth number of channels – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
and wherein each decomposed convolutional layer is associated with a respective second ranking number – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
A device for implementing a tensor-train decomposition operation for a respective convolutional layer of a convolutional neural network (CNN) – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
the device being configured to – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
obtain the CNN comprising a first number of convolutional layers, wherein each convolutional layer is associated with a respective first ranking number – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data). The courts have also found limitations directed to obtaining and storing information electronically, recited at a high level of generality, to be well-understood, routine, and conventional. See MPEP § 2106.05(d)(II) “receiving or transmitting data over a network,” "electronic record keeping,” and "storing and retrieving information in memory.”
and provide a decomposed CNN comprising a second number of convolutional layers and a third number of decomposed convolutional layers based on a training of the CNN – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
receive input data comprising a first number of channels at one of the third number of decomposed convolutional layers – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data). The courts have also found limitations directed to obtaining and storing information electronically, recited at a high level of generality, to be well-understood, routine, and conventional. See MPEP § 2106.05(d)(II) “receiving or transmitting data over a network,” "electronic record keeping,” and "storing and retrieving information in memory.”
the plurality of data groups comprising a second number of channels – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
to obtain intermediate data comprising a third number of channels – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
to obtain output data comprising a fourth number of channels – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
and wherein each decomposed convolutional layer is associated with a respective second ranking number – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 2:
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
wherein: the group convolution is performed based on a kernel shared between the plurality of data groups – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
wherein: the group convolution is performed based on a kernel shared between the plurality of data groups – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 3:
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
wherein: the third number of channels is determined based on a number of data groups in the plurality of data groups – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
wherein: the third number of channels is determined based on a number of data groups in the plurality of data groups – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 4:
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
wherein: the third number of channels is further determined based on one or more hardware characteristics of the device – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
wherein: the third number of channels is further determined based on one or more hardware characteristics of the device – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 5:
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
wherein: each data group comprises a fifth number of channels, and wherein the second number of channels is determined based on the third number of channels and the fifth number of channels – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
wherein: each data group comprises a fifth number of channels, and wherein the second number of channels is determined based on the third number of channels and the fifth number of channels – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 7:
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
the device according to claim 6, further configured to – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
determine, for a respective convolutional layer of the CNN, a weighting pair based on: a weighted convolutional layer obtained by allocating a first weighting trainable parameter to the respective convolutional layer; and
a weighted decomposed convolutional layer obtained by allocating a second weighting trainable parameter to a decomposed convolutional layer determined for the respective convolutional layer – this is recited at a high level of generality and amounts to insignificant extra-solution activity as data gathering/storage that is limited to a particular type of data, generally linking the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(g) and (h), and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
the device according to claim 6, further configured to – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
determine, for a respective convolutional layer of the CNN, a weighting pair based on: a weighted convolutional layer obtained by allocating a first weighting trainable parameter to the respective convolutional layer; and
a weighted decomposed convolutional layer obtained by allocating a second weighting trainable parameter to a decomposed convolutional layer determined for the respective convolutional layer – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data). The courts have also found limitations directed to obtaining and storing information electronically, recited at a high level of generality, to be well-understood, routine, and conventional. See MPEP § 2106.05(d)(II) “receiving or transmitting data over a network,” "electronic record keeping,” and "storing and retrieving information in memory.”
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 8:
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
the device according to claim 7, further configured to – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
perform an initial training iteration of the CNN based on at least the weighting pair – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
the device according to claim 7, further configured to – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
perform an initial training iteration of the CNN based on at least the weighting pair – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 9:
The claim recites the following additional mathematical concept elements:
determine, after performing the initial training iteration, at least one convolutional layer having a minimal first weighting trainable parameter – determining the minimal parameter from a group of parameters is a mathematical calculation.
Accordingly, at step 2A, prong one, the claim is directed to an abstract idea.
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
the device according to claim 8, further configured to – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
the device according to claim 8, further configured to – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 10:
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
the device according to claim 9, further configured to – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
perform an additional training iteration of the CNN – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
based on substituting a weighting pair of the at least one convolutional layer having the minimal first weighting trainable parameter with a corresponding decomposed convolutional layer, and a remaining of the at least one weighting pair from a previous iteration – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
the device according to claim 9, further configured to – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
perform an additional training iteration of the CNN – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
based on substituting a weighting pair of the at least one convolutional layer having the minimal first weighting trainable parameter with a corresponding decomposed convolutional layer, and a remaining of the at least one weighting pair from a previous iteration – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 11:
The claim recites the following additional mathematical concept elements:
determining a respective convolutional layer having a minimal first weighting trainable parameter – determining the minimal parameter from a group of parameters is a mathematical calculation.
until a predetermined number of convolutional layers are substituted with corresponding decomposed convolutional layers – comparing the iteration/number of substituted layers value to a predetermined number is a mathematical calculation.
Accordingly, at step 2A, prong one, the claim is directed to an abstract idea.
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
the device according to claim 8, further configured to – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
iteratively perform, … substituting the weighting pair of the respective convolutional layer having the minimal first weighting trainable parameter with a corresponding decomposed convolutional layer, and performing a next training iteration – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
the device according to claim 8, further configured to – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
iteratively perform, … substituting the weighting pair of the respective convolutional layer having the minimal first weighting trainable parameter with a corresponding decomposed convolutional layer, and performing a next training iteration – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 12:
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
comprising an artificial intelligence accelerator adapted for tensor processing operation of the CNN – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
comprising an artificial intelligence accelerator adapted for tensor processing operation of the CNN – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f).
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 13, see the rejection of claim 1, above.
As per claim 14:
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
A tangible, non-transitory computer-readable medium having instructions thereon, which, upon being executed by a computer, cause the steps of the method of claim 13 to be performed – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f). [Examiner’s Note: see above for the method steps themselves.]
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
A tangible, non-transitory computer-readable medium having instructions thereon, which, upon being executed by a computer, cause the steps of the method of claim 13 to be performed – this amounts to mere instructions to apply the exception using a generic computer component, recited at a high level of generality. See MPEP § 2106.05(f). [Examiner’s Note: see above for the method steps themselves.]
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 15, see the rejection of claim 2, above.
As per claim 16, see the rejection of claim 3, above.
As per claim 17, see the rejection of claim 4, above.
As per claim 18, see the rejection of claim 5, above.
As per claim 19:
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
wherein the CNN processes image data, text data, or voice data – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
wherein the CNN processes image data, text data, or voice data – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
As per claim 20:
Under step 2A, prong two, the judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of:
wherein the input data is image data, text data, or voice data – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2A, prong two, these additional elements do not integrate the abstract idea into a practical application for the claim as a whole, because it does not impose any meaningful limits on practicing the abstract idea. See MPEP § 2106.04(d).
Under step 2B, the claims do not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the claim recites the additional elements of:
wherein the input data is image data, text data, or voice data – this amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Accordingly, at step 2B, these additional elements, both individually and in combination, do not amount to significantly more than the judicial exception. See MPEP § 2106.05.
Therefore, the claim is not eligible subject matter under 35 U.S.C. 101.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
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) 1-5 and 7-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Wang et al. (Wide Compression: Tensor Ring Nets, Feb 2018, pgs. 1-12) in view of Choudhury (US 2020/0410336 – cited in an IDS).
As per claim 1, Wang teaches a device for implementing a tensor-train decomposition operation for a convolutional layer of a convolutional neural network (CNN) [tensor ring networks (TR-Nets) compress the fully-connected and convolutional layers of deep neural networks (pg. 1, abstract, etc.) using tensor ring decomposition (pg. 2, section 3; etc.)], the device being configured to:
obtain the CNN comprising a first number of convolutional layers, wherein each convolutional layer is associated with a respective first ranking number [TRN compresses a CNN including fully connected and convolutional layers (pg. 1, abstract, etc.) where each is associated with a tensor ring rank R (pg. 2, section 3; etc.)];
receive input data comprising a first number of channels [each convolutional layer decomposes the number of channels entering the layer, which is 1 at the first input (pg. 5, fig. 5, etc.); where 1 is the first number of channels in this case];
perform a 1x1 convolution on the input data to obtain a plurality of data groups, the plurality of data groups comprising a second number of channels [in TRN, tensor decomposition is applied by combining kernel tensor factorization with convolution operations in three steps (11)-(13), where (11) is a convolutional layer from I feature maps to R2 feature maps with a 1x1 patch (pg. 5, section 3.2; etc.) and each convolutional layer decomposes the number of channels entering the layer (pg. 5, fig. 5, etc.); where the output channels of the 1x1 convolution are the second number of channels];
perform a group convolution on the plurality of data groups to obtain intermediate data comprising a third number of channels [in TRN, tensor decomposition is applied by combining kernel tensor factorization with convolution operations in three steps (11)-(13), where (12) contains R convolutional layers from R feature maps to R feature maps with a D x D patch (group convolution) (pg. 5, section 3.2; etc.) and each convolutional layer decomposes the number of channels entering the layer (pg. 5, fig. 5, etc.); where the output of the group convolution is the intermediate data with a third number of channels]; and
perform a 1x1 convolution on the intermediate data to obtain output data comprising a fourth number of channels [in TRN, tensor decomposition is applied by combining kernel tensor factorization with convolution operations in three steps (11)-(13), where (13) is a convolutional layer from R2 feature maps to O feature maps with a 1 x 1 patch. (pg. 5, section 3.2; etc.) and each convolutional layer decomposes the number of channels entering the layer (pg. 5, fig. 5, etc.); where the output of the third convolutional layer is the output data with a fourth number of channels].
While Wang also teaches decomposing convolutional layers of a CNN (see above), it has not been relied upon for teaching the device configured to: provide a decomposed CNN comprising a second number of convolutional layers and a third number of decomposed convolutional layers based on a training of the CNN; and receive input data at one of the third number of decomposed convolutional layers; wherein the first number of convolutional layers equals a sum of the second number of convolutional layers and the third number of decomposed convolutional layers, and wherein each decomposed convolutional layer is associated with a respective second ranking number.
Choudhury teaches a device [the invention may be implemented in a mobile of IoT device, etc. (paras. 0002, 0079, etc.)] configured to:
provide a decomposed CNN comprising a second number of convolutional layers and a third number of decomposed convolutional layers based on a training of the CNN [the system receives and trains a CNN model for a number of epochs, performs weighted decomposition for some of the convolutional layers based upon the determined rankings, and retrains a compressed/decomposed version of the model, (paras. 0030-31; figs. 3-4; etc.); where the low rank factorization produces a (third) number of decomposed convolutional layers, and the remaining convolutional layers are the second number of convolutional layers];
receive input data at one of the third number of decomposed convolutional layers [the system receives and trains a CNN model for a number of epochs, performs weighted decomposition for some of the convolutional layers based upon the determined rankings, and retrains a compressed/decomposed version of the model, (paras. 0030-31; figs. 3-4; etc.); where the low rank factorization produces a (third) number of decomposed convolutional layers receiving inputs];
wherein the first number of convolutional layers equals a sum of the second number of convolutional layers and the third number of decomposed convolutional layers, and wherein each decomposed convolutional layer is associated with a respective second ranking number [the system receives and trains a CNN model for a number of epochs, performs weighted decomposition for some of the convolutional layers based upon the determined rankings, and retrains a compressed/decomposed version of the model, (paras. 0030-31; figs. 3-4; etc.); where the low rank factorization produces a (third) number of decomposed convolutional layers, and the remaining convolutional layers are the second number of convolutional layers, which sums to the first number of layers].
Wang and Choudhury are analogous art, as they are within the same field of endeavor, namely decomposition of neural networks.
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to include layer-wise ranking and decomposition, as taught by Choudhury, in the convolutional layer decomposition in the system/method taught by Wang.
Choudhury provides motivation as [One way to reduce the amount of required computing resources is to compress the model using one or more compression mechanisms, which include, for example, pruning, weight sharing, encoding, and performing low rank decomposition of connection weights (para. 0019, etc.) and at least one embodiment of the present invention may provide a beneficial effect such as, for example, reducing size and/or flops of neural networks compared to existing compression techniques. One or more embodiments of the present invention may also provide a beneficial effect such as, for example, allowing deep learning models with a greater number of layers and parameter to be used on device with limited resources (such as mobile phones and other edge devices in the context of IOT, for example) (para. 0079), which is also a desired outcome contemplated by Wang (see, e.g., Wang: pg. 1, section 1; etc.)].
As per claim 2, Wang/Choudhury teaches wherein: the group convolution is performed based on a kernel shared between the plurality of data groups [in TRN, tensor decomposition is applied by combining kernel tensor factorization with convolution operations in three steps (11)-(13), where (12) contains R convolutional layers from R feature maps to R feature maps with a D x D patch (group convolution) (Wang: pg. 5, section 3.2; etc.); which is a group convolution based on a shared kernel].
As per claim 3, Wang/Choudhury teaches wherein: the third number of channels is determined based on a number of data groups in the plurality of data groups [each convolutional layer decomposes the number of channels entering the layer (Wang: pg. 5, fig. 5, etc.); where the third number of channels is based upon the channels entering the third convolutional layer, which is based on the second, which is based on the first].
As per claim 4, Wang teaches the device according to claim 3, as described above.
While Wang teaches that the decomposition is based on resource constraints (see, e.g., Wang: pg. 1, section 1), it has not been relied upon for teaching wherein: the third number of channels is further determined based on one or more hardware characteristics of the device.
Choudhury teaches wherein: the third number of channels is further determined based on one or more hardware characteristics of the device [a reconstructed tensor may be generated based on the core and factor matrices that essentially approximate the weight tensor of, e.g., a given layer of a neural network such that the reconstructed tensor has a smaller rank than the weight tensor, thereby reducing the required size and/or flops (para. 0021, etc.) and the determination of ranking for the layers and factorization is based upon the resource constraints of the system (paras. 0030-31, etc.); so, the number of channels output by the layer is based upon the resource constraints of the system, which is a hardware characteristic of the device].
Wang and Choudhury are analogous art, as they are within the same field of endeavor, namely decomposition of neural networks.
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to utilize the system’s resource constraints as part of the factorization/decomposition of the layer, which determines the number of channels, as taught by Choudhury, in the factorization/decomposition of the layers in the system/method taught by Wang.
Choudhury provides motivation as [One way to reduce the amount of required computing resources is to compress the model using one or more compression mechanisms, which include, for example, pruning, weight sharing, encoding, and performing low rank decomposition of connection weights (para. 0019, etc.) and at least one embodiment of the present invention may provide a beneficial effect such as, for example, reducing size and/or flops of neural networks compared to existing compression techniques. One or more embodiments of the present invention may also provide a beneficial effect such as, for example, allowing deep learning models with a greater number of layers and parameter to be used on device with limited resources (such as mobile phones and other edge devices in the context of IOT, for example) (para. 0079), which is also a desired outcome contemplated by Wang (see, e.g., Wang: pg. 1, section 1; etc.)].
As per claim 5, Wang/Choudhury teaches wherein: each data group comprises a fifth number of channels, and wherein the second number of channels is determined based on the third number of channels and the fifth number of channels [each convolutional layer decomposes the number of channels entering the layer (Wang: pg. 5, fig. 5, etc.); where the number of channels of the grouped convolution (second number of channels, above) is based upon the number of channels in each member of the group (the fifth number of channels) and the desired output (third number of channels)].
As per claim 7, Wang/Choudhury teaches the device according to claim 6, further configured to determine, for a respective convolutional layer of the CNN, a weighting pair based on:
a weighted convolutional layer obtained by allocating a first weighting trainable parameter to the respective convolutional layer [the filter significance determination module performs function based on the set of statistics to determine filter significance (W) for each layer of the trained model, the significance values are then provided as input to the weighted decomposition module, which multiplies the filter values with the filter significance. At 336, the module 308 determines a rank based at least in part on one or more resource constraints 314, and then performs a low rank factorization process as represented by 334. A compressed version 312 of the trained model 302 is output by the weighted decomposition module 308 (Choudhury: paras. 0030-31; figs. 3-4; etc.); where the determined rank weighting for the convolutional layer before decomposition is the first weighting trainable parameter]; and
a weighted decomposed convolutional layer obtained by allocating a second weighting trainable parameter to a decomposed convolutional layer determined for the respective convolutional layer [the filter significance determination module performs function based on the set of statistics to determine filter significance (W) for each layer of the trained model, the significance values are then provided as input to the weighted decomposition module, which multiplies the filter values with the filter significance. At 336, the module 308 determines a rank based at least in part on one or more resource constraints 314, and then performs a low rank factorization process as represented by 334. A compressed version 312 of the trained model 302 is output by the weighted decomposition module 308 (Choudhury: paras. 0030-31; figs. 3-4; etc.); where the determined weight for the convolutional layer after decomposition is the second weighting trainable parameter].
As per claim 8, Wang/Choudhury teaches the device according to claim 7, further configured to:
perform an initial training iteration of the CNN based on at least the weighting pair [The filter significance determination module 306 performs a few epochs of training (forward and backward pass) as shown at 322, and then computes a set of statistics at 324. The trained model 302 is retrained at 332 using the dataset 304 and the resulting tensor. A compressed version 312 of the trained model 302 is then output by the weighted decomposition module 308 (Choudhury: paras. 0030-31; figs. 3-4; etc.); where the retraining iterations are based upon the ranking and decomposed weightings].
As per claim 9, Wang/Choudhury teaches the device according to claim 8, further configured to:
determine, after performing the initial training iteration, at least one convolutional layer having a minimal first weighting trainable parameter [At 326, the filter significance determination module 306 performs one or more of a step, sigmoid, and logistic function based on the set of statistics to determine filter significance (W) for each layer of the trained model. The filter significance values are then provided as input to the weighted decomposition module 308. The weighted decomposition module 308 multiplies the filter values with the filter significance as shown at 338 to obtain an input tensor for a given layer of the neural network. At 336, the module 308 determines a rank based at least in part on one or more resource constraints 314, and then performs a low rank factorization process as represented by 334 (Choudhury: paras. 0030-31; figs. 3-4; etc.); where the low rank factorization is the determination of the minimal first weighting trainable parameter].
As per claim 10, Wang/Choudhury teaches the device according to claim 9, further configured to:
perform an additional training iteration of the CNN, based on substituting a weighting pair of the at least one convolutional layer having the minimal first weighting trainable parameter with a corresponding decomposed convolutional layer, and a remaining of the at least one weighting pair from a previous iteration [The trained model 302 is retrained at 332 using the dataset 304 and the resulting tensor. A compressed version 312 of the trained model 302 is then output by the weighted decomposition module 308 (Choudhury: paras. 0030-31; figs. 3-4; etc.); which includes the decomposed convolutional layer and its associated minimal first weighting and ranking (see above)].
As per claim 11, Wang/Choudhury teaches the device according to claim 8, further configured to:
iteratively perform, determining a respective convolutional layer having a minimal first weighting trainable parameter, substituting the weighting pair of the respective convolutional layer having the minimal first weighting trainable parameter with a corresponding decomposed convolutional layer, and performing a next training iteration, until a predetermined number of convolutional layers are substituted with corresponding decomposed convolutional layers [At 326, the filter significance determination module 306 performs one or more of a step, sigmoid, and logistic function based on the set of statistics to determine filter significance (W) for each layer of the trained model. The filter significance values are then provided as input to the weighted decomposition module 308. The weighted decomposition module 308 multiplies the filter values with the filter significance as shown at 338 to obtain an input tensor for a given layer of the neural network. At 336, the module 308 determines a rank based at least in part on one or more resource constraints 314, and then performs a low rank factorization process as represented by 334. The low factorization process generates a resulting tensor having a rank equal to the determined rank, wherein the resulting tensor approximates the input tensor. The trained model 302 is retrained at 332 using the dataset 304 and the resulting tensor. A compressed version 312 of the trained model 302 is then output by the weighted decomposition module 308 (Choudhury: paras. 0030-31; figs. 3-4; etc.); which ranking, low rank factorization/decomposition, and retraining is iteratively performing the steps].
As per claim 12, Wang/Choudhury teaches comprising an artificial intelligence accelerator adapted for tensor processing operations of the CNN [the processor(s) used to implement the invention can include special purpose processors/hardware for executing the operations (Choudhury: paras. 0049-51), which is an AI accelerator adapted for tensor processing operations of the CNN, as it is hardware specifically designed for the CNN tensor operations described by the embodiments of Choudhury (see above)].
As per claim 13, see the rejection of claim 1, above.
As per claim 14, Wang/Choudhury teaches a tangible, non-transitory computer-readable medium having instructions thereon, which, upon being executed by a computer, cause the steps of the method of claim 13 to be performed [the system may be implemented in resource-constrained devices, such as a mobile, wearable, or IOT device including application, storage, and memory constraints (Wang: pg. 1, abstract and section 1; etc.); which includes a computer executing instructions from memory].
As per claim 15, see the rejection of claim 2, above.
As per claim 16, see the rejection of claim 3, above.
As per claim 17, see the rejection of claim 4, above.
As per claim 18, see the rejection of claim 5, above.
As per claim 19, Wang/Choudhury teaches wherein the CNN processes image data, text data, or voice data [the CNN can be used for image classification (Wang: abstract, etc.)].
As per claim 20, Wang/Choudhury teaches wherein the input data is image data, text data, or voice data [the CNN can be used for classifying images (Wang: abstract, etc.)].
Response to Arguments
The objection to claim 12 has been withdrawn due to the amendments filed.
Applicant's arguments filed 7 July 2026 have been fully considered but they are not persuasive.
Applicant argues that the claims are not drawn to mathematical concepts, methods of organizing human activity or a mental process.
However, the mathematical concepts of the claimed invention are identified in the rejections above.
Applicant also argues that the claims in question are mere based on, or involve, a mathematical concept and relate to training a machine learning model to facilitate a hardware-friendly tensor-train decomposition and do not recite any specific mathematical relationships, formulas or calculations.
However, the mathematical relationships, formulas, and/or calculations recited by the claims are identified above. For example, “perform a 1x1 convolution on the input data to obtain a plurality of data groups” is a specific mathematical calculation recited in claim 1. Additionally, utilizing the claimed mathematical calculations for implementing a tensor-train decomposition operation for a respective convolutional layer of a convolutional neural network amounts to generally linking the use of the judicial exception to a particular technological environment or field of use by limiting it to a particular data source or type. See MPEP § 2106.05(h) and Electric Power, 830 F.3d at 1354, 119 USPQ2d at 1742 (limiting application of abstract idea to power grid data).
Applicant further argues that the claims “are directed to methods of training a machine learning model to facilitate a hardware-friendly tensor-train decomposition of one or more convolutional layers of a convolutional neural network,” which provides a practical application via improvements in machine learning. Applicant argues that this solves a problem of “neural networks being computationally intensive to train and having varying degrees of accuracies” in a computationally efficient way by performing training a machine learning model with reduced computational complexity of convolutional neural networks by this facilitation.
However, applicant has – at most – described an improvement to the calculations being performed for training. As described above, this is a mathematical concept. Therefore, (assuming that the invention provides these improvements) this amounts to an improvement to an abstract idea rather than to a computer or technology. See MPEP 2106.05(a). It appears that any benefits to the computer itself are based solely on the use of an improvement to the abstract idea(s), using generic computer components to apply the abstract idea(s). Additionally, the Federal Circuit has also indicated that mere automation of manual processes or increasing the speed of a process, where these purported improvements come solely from the capabilities of a general-purpose computer are not sufficient to show an improvement in computer functionality. FairWarning IP, LLC v. latric Sys., 839 F.3d 1089, 1095, 120 USPQ2d 1293, 1296 (Fed. Cir. 2016); Credit Acceptance Corp. v. Westlake Services, 859 F.3d 1044, 1055, 123 USPQ2d 1100, 1108-09 (Fed. Cir. 2017). The Federal Circuit has also indicated that a claim must include more than conventional implementation on generic components or machinery to qualify as an improvement to an existing technology. Affinity Labs of Tex. v. DirecTV, LLC, 838 F.3d 1253, 1264-65, 120 USPQ2d 1201, 1208-09 (Fed. Cir. 2016); TLI Communications LLC v. AVAuto, LLC, 823 F.3d 607, 612-613, 118 USPQ2d 1744, 1747-48 (Fed. Cir. 2016). Claims must also include more than just instructions to perform the method on a generic component or machinery to qualify as an improvement to an existing technology (MPEP § 2106.05(a)). Finally, to find a valid improvement to a computer or technology the specification must disclose the improvement and the claim must include the necessary components to realize the improvement. MPEP 2106.05(d)(1).
Applicant also argues that the cited art does not teach wherein the first number of convolutional layers equals a sum of the second number of convolutional layers and the third number of decomposed convolutional layers, and wherein each decomposed layer is associated with a respective ranking number.
However, Wang teaches TRN compresses a CNN including fully connected and convolutional layers (pg. 1, abstract, etc.) where each is associated with a tensor ring rank R (pg. 2, section 3; etc.), while Choudhury teaches the system receives and trains a CNN model for a number of epochs, performs weighted decomposition for some of the convolutional layers based upon the determined rankings, and retrains a compressed/decomposed version of the model, (paras. 0030-31; figs. 3-4; etc.). Here the low rank factorization produces a (third) number of decomposed convolutional layers, and the remaining convolutional layers are the second number of convolutional layers, which sums to the first number of layers.
Applicant further argues that Choudhury makes no mention of the number of layers of the compressed version of the trained model being related to the number of layers of the trained neural network.
However, as described above, Choudhury teaches producing a compressed version of the layers of the trained model. Choudhury also teaches the process 500 may include using the determined relevancies of the one or more of the filters and the channels to adjust at least one weight tensor of the trained model, wherein the at least one weight tensor comprises weights for the one or more of the filters and the channels of the at least one layer. The compressing may include: applying a weighted low rank factorization process to the adjusted at least one tensor to generate at least one resulting tensor that approximates the at least one weight tensor of the trained model, wherein a target rank of the at least one resulting tensor is smaller than a rank of the at least one weight tensor of the trained model (Choudhury: para. 0033, etc.) which describes that the compressed version of the trained model is based upon the layers of the model.
Conclusion
The following is a summary of the treatment and status of all claims in the application as recommended by M.P.E.P. 707.07(i): claim 6 is cancelled; claims 1-5 and 7-20 are rejected.
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Cheng (US 2021/0241094) – discloses using reinforcement learning for rank selection in tensor decomposition of neural network layers.
Sather (US 12,061,988) – discloses layer-wise decomposition of weight tensors, including decomposing a layer into multiple layers.
Kim et al. (Compression of Deep Convolutional Neural Networks for Fast and Low Power Mobile Applications, Feb 2016, pgs. 1-16) – discloses decomposition/compression of CNNs.
Lebedev et al. (Speeding-up Convolutional Neural Networks Using Fine-tuned CP-Decomposition, April 2015, pgs. 1-11) – discloses low-rank CP-decomposition of convolutional layers.
The examiner requests, in response to this Office action, that support be shown for language added to any original claims on amendment and any new claims. That is, indicate support for newly added claim language by specifically pointing to page(s) and line number(s) in the specification and/or drawing figure(s). This will assist the examiner in prosecuting the application.
When responding to this office action, Applicant is advised to clearly point out the patentable novelty which he or she thinks the claims present, in view of the state of the art disclosed by the references cited or the objections made. He or she must also show how the amendments avoid such references or objections. See 37 CFR 1.111(c).
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 nonprovisional extension fee (37 CFR 1.17(a)) 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 mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to GEORGE GIROUX whose telephone number is (571)272-9769. The examiner can normally be reached M-F 10am-6pm.
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, Omar Fernandez Rivas can be reached at 571-272-2589. 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.
/GEORGE GIROUX/Primary Examiner, Art Unit 2128
Prosecution Insights