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 1-20 remain pending and are ready for examination.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 12/17/2024, was filed. The submission is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
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 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.
Claims 1, 6-13 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by "NISP: Pruning Networks using Neuron Importance Score Propagation" by Yu et al. 2018 (hereinafter 'Yu').
Yu teaches the claims below:
1. A method comprising:
determining critical neurons in a neural network (page 1, wherein determine an importance score for neurons);
pruning the critical neurons from the neural network, wherein the pruning generates a pruned neural network with the critical neurons removed (see fig.1, page 1 and page 3, wherein ranking neurons in the final response layer and propagates those importance scores throughout the network and removing neurons with lower importance); and
continuing a training process using the pruned neural network (see fig.1, page 1 and page 3, wherein the CNN is pruned by removing neurons with least importance, and then fine-tuned (continuing a training process) to retain its predictive power).
6. The method of claim 1, wherein the pruning of the critical neurons is implemented globally (see page 1 and 7, wherein global measurement of neuron importance across different layers of a CNN).
7. The method of claim 1, wherein the pruning of the critical neurons is implemented layer-wise (see page 7, wherein layer-by-layer pruning corresponds to layer-wise).
8. A critical neuron pruning system comprising:
a memory; and
a processor that is configured to execute machine readable instructions stored in the memory for causing the processor to:
determine critical neurons in a neural network (page 1, wherein determine an importance score for neurons);
prune the critical neurons from the neural network, wherein the pruning generates a pruned neural network with the critical neurons removed (see fig.1, page 1 and page 3, wherein ranking neurons in the final response layer and propagates those importance scores throughout the network and removing neurons with lower importance); and
continue a training process using the pruned neural network (see fig.1, page 1 and page 3, wherein the CNN is pruned by removing neurons with least importance, and then fine-tuned (continuing a training process) to retain its predictive power).
9. The critical neuron pruning system of claim 8, wherein the system receives a pruning rate parameter (see page 3 and fig.2 wherein pre-defined pruning ratio is the pruning rate parameter).
10. The critical neuron pruning system of claim 9, wherein the system determines a number of critical neurons for a layer of the network based on the pruning rate parameter and a total number of neurons in the layer (see fig.1, page 1 and page 3 and 6, wherein ranking neurons in the final response layer and propagates those importance scores throughout the network and removing neurons with lower importance).
11. The critical neuron pruning system of claim 8, wherein the pruning comprises removing the number of critical neurons from the layer and weights associated with the critical neurons from the layer (see fig.1, page 1 and page 3 and 6, wherein ranking neurons in the final response layer and propagates those importance scores throughout the network and removing neurons with lower importance).
12. The critical neuron pruning system of claim 8, wherein the critical neurons are determined based on features associated with each neuron comprising forward propagation- based features and back propagation-based features (see page 2, 5 and 9).
13. The critical neuron pruning system of claim 8, wherein the weight pruning is implemented layer-wise (see page 7, wherein layer-by-layer pruning corresponds to layer-wise).
Claim 14 is rejected under 35 U.S.C. 102(a)(1) as being anticipated by "Gradual Channel Pruning while Training using Feature Relevance Scores for Convolutional Neural Networks" by Aketi et al. 2020 (hereinafter Aketi).
Aketi teaches the claims below:
14. A method, comprising:
receiving a normalized feature vector corresponding to each neuron in a layer of a neural network (see page 5, wherein Algorithm 1 shows the pseudo code for computing the feature-relevance score of layer l. After determining the relevance scores at each node in the network using LRP, we compute the relevance score of every feature map fi at layer l by averaging the scores of all nodes corresponding to fi. The relevance vector of a feature map fi is obtained by taking class-wise average over relevance scores of all training samples and forms the ith column of feature relevance matrix (FMl). The weighted average of rows of FMl with class-wise accuracy returns feature relevance scores (FSl) for any layers l. The computed feature relevance scores are then utilized to determine the least important channels that can be pruned.);
calculating a score for each neuron in the layer, wherein the score is based on the normalized feature vector and features collected during training of the neural network (page. 2, Feature relevance score is computed based on normalized class-wise accuracy of the model and class-wise relevance score of each channel."; page. 4, Normalize rows of feature-relevance matrix. see also page. 5); and
determining a set of neurons having a score lower than a threshold, wherein the neurons in the set of neurons are identified as critical neurons (page. 3, wherein The pruning techniques which prune at the level of individual weights or nodes can bei categorized under unstructured pruning methods. Some of the recent works such as [7, 8, 18, 23] have proposed unstructured pruning. The authors of [8] proposed iterative weight pruning where the weights with magnitude less than a given threshold are removed."; see also page 5).
Claim Rejections - 35 USC § 103
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 of this title, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 2-5 are rejected under 35 U.S.C. 103 as being unpatentable over Yu in view of “Towards Efficient Neuromorphic Hardware: Unsupervised Adaptive Neuron Pruning” Guo 2020 (Hereinafter “Guo”).
Yu fails to explicitly disclose the limitations below. Guo discloses these limitation as mapped below:
2. The method of claim 1, further comprising determining an early stop point during the training of the neural network (see sec. 2.2, wherein Before delving into the details of neuron pruning, we need to find out when to start the pruning process. If the neuron pruning process starts too early, important neurons that have a profound impact on the output could be mistakenly pruned away, which will deteriorate network performance. On the other hand, if it starts too late, the network might not have enough training cycles to compensate for the accuracy loss. Thus, it is crucial to determine when to start the pruning process. To find this critical point, we have observed how the network dynamics evolve with time by monitoring firing activities and weight updates of output neurons, as shown in Figure 2. Firing activity directly reflects the response of neurons to input patterns. As neurons start to fire regularly after training over 30,000 images, the network has learned the major input features and starts to adjust for small details. This is also confirmed by the statistics of weight updates over time, as shown in Figure 2b. The synchronization happens due to the STDP learning rule that links firing activity and weight updates together).
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the system of Yu to include the missing limitation, as taught by LIM, as it would allow the system to determine when to start the pruning process for accuracy loss (see sec. 2.2).
3. The method of claim 2, wherein the early stop point comprises a subset of a training data set (see sec. 2, wherein dividing a whole dataset into multiple batches. Thus, the pruning/start point corresponds to 30000of the 60000 training samples).
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the system of Yu to include the missing limitation, as taught by LIM, as it would allow the system to determine when to start the pruning process for accuracy loss (see sec. 2.2).
4. The method of claim 3, wherein initial phases of the training process train the neural network using the subset of training data before the early stop point (see sec. 2 to sec. 2.2, wherein training the unpruned SNN on the first 30000 training images. Monitor neuro activity and weight updates during that pre-pruning phase and then the network has learned major input features after over those images).
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the system of Yu to include the missing limitation, as taught by LIM, as it would allow the system to determine when to start the pruning process for accuracy loss (see sec. 2.2).
5. The method of claim 3, wherein continuing the training process comprises training the neural network using a remaining subset of training data after the early stop point (see sec. 2 to sec. 2.2, wherein the training dataset has 60000 images, and neuron pruning starts at 30000 images. After that point, pruning occurs after successive batches. Algorithm 1 states pruning process starts after 30000 training images followed by continuing loop over the batches).
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the system of Yu to include the missing limitation, as taught by LIM, as it would allow the system to determine when to start the pruning process for accuracy loss (see sec. 2.2).
Claims 9-14 and 16-21 are rejected under the same rationale as claims 2-7.
Claims 15-20 are rejected under 35 U.S.C. 103 as being unpatentable over Aketi in view of Yu.
Aketi fails to explicitly disclose the limitations below. Yu discloses these limitation as mapped below:
15. The method of claim 14, further comprising: receiving a pruning rate parameter (Yu, see page 3 and fig.2 wherein pre-defined pruning ratio is the pruning rate parameter).
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the system of Aketi to include the missing limitation, as taught by Yu, as it would allow the system to achieve significant acceleration and compression with negligible accuracy loss (Yu, see Abstract).
16. The method of claim 15, further comprising: determining a number of critical neurons for a layer of the network based on the pruning rate parameter and a total number of neurons in the layer (see fig.1, page 1 and page 3 and 6, wherein ranking neurons in the final response layer and propagates those importance scores throughout the network and removing neurons with lower importance).
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the system of Aketi to include the missing limitation, as taught by Yu, as it would allow the system to achieve significant acceleration and compression with negligible accuracy loss (Yu, see Abstract).
17. The method of claim 14, further comprising: pruning the critical neurons from the neural network, wherein the pruning generates a pruned neural network with the critical neurons removed (see fig.1, page 1 and page 3, wherein ranking neurons in the final response layer and propagates those importance scores throughout the network and removing neurons with lower importance).
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the system of Aketi to include the missing limitation, as taught by Yu, as it would allow the system to achieve significant acceleration and compression with negligible accuracy loss (Yu, see Abstract).
18. The method of claim 17, wherein the pruning comprises removing the number of critical neurons from the layer and weights associated with the critical neurons from the layer (see fig.1, page 1 and page 3 and 6, wherein ranking neurons in the final response layer and propagates those importance scores throughout the network and removing neurons with lower importance).
19. The method of claim 17, wherein the pruning of the critical neurons is implemented layer-wise (see page 7, wherein layer-by-layer pruning corresponds to layer-wise).
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the system of Aketi to include the missing limitation, as taught by Yu, as it would allow the system to achieve significant acceleration and compression with negligible accuracy loss (Yu, see Abstract).
20. The method of claim 17, wherein the pruning of the critical neurons is implemented globally (see page 1 and 7, wherein global measurement of neuron importance across different layers of a CNN).
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the system of Aketi to include the missing limitation, as taught by Yu, as it would allow the system to achieve significant acceleration and compression with negligible accuracy loss (Yu, see Abstract).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MAHER N ALGIBHAH whose telephone number is (571)272-0718. The examiner can normally be reached on Monday-Thursday.
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, Aleksandr Kerzhner can be reached on (571) 270-1760. The fax phone number for the organization where this application or proceeding is assigned is 571-273-1264.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/MAHER N ALGIBHAH/Primary Examiner , Art Unit 2165
