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 .
Claims 1-20 are pending.
Claim Objections
Claim 1 is objected to because of the following informalities: Line 7 of claim 1 recites “the neural network model including” (emphasis added) is referred to which “a neural network model” recited in line 3 OR “a neural network model” recited in line 7.
Claims 2-8 further depend claim 1. Therefore, claims 2-8 are objected under the same rationale.
Claim 5 is objected to because of the following informalities: Line 12 recites “the plurality of device forward pass values”, which is exactly the same limitation preceding it. The Examiner thinks that it might be a typo error and it should be rewritten as “the plurality of device backward pass values”.
Claim 9 is objected to because of the following informalities:
Line 6 of claim 9 recites “a server” and “a neural network model”, which both should be rewritten as “the server” and “the neural network model” because both “a server” and “a neural network model” is already mentioned in line 3 of claim 9.
Line 9 recites “applying a device partition”, which should be rewritten as “applying the device partition”.
Line 10 recites “a plurality of layers”, which should be rewritten as “the plurality of layers”.
Line 10 recites “a server”, which should be rewritten as “the server”.
Claims 10-16 further depend claim 9. Therefore, claims 10-16 are objected under the same rationale.
Claim 13 is objected to because of the following informalities:
Line 4 recites “during training”, which should be rewritten as “during the training”.
Claim 17 is objected to because of the following informalities:
Line 6 recites “of a neural network model”, which should be rewritten as “of the neural network model”.
Claims 18-20 further depend claim 17. Therefore, claims 18-20 are objected under the same rationale.
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 4-6, 9, 14, 19 and 20 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 4 recites the limitation "the partition value” and “the mini-batch value" in line 7. There is insufficient antecedent basis for these limitations in the claim.
Claim 5 recites the limitations “the device and the server” in lines 19-20. There is insufficient antecedent basis for these limitations in the claim.
Claim 5 recites the limitations “the number of layers” in line 23 and “the plurality of mini-batches” in the last line. There is insufficient antecedent basis for these limitations in the claim.
Claim 6 recite the limitation “the weight values” in line 2. There is insufficient antecedent basis for the limitation in the claim.
Claim 9 recites “the server” in line 3. There is insufficient antecedent basis for the limitation in the claim.
Claim 14 recites “the weight values” in line 2 and “the set of gradient vectors” in lines 2-3. There is insufficient antecedent basis for these limitations in the claim.
Claim 19 recites “the activation” in line 4. There is insufficient antecedent basis for the limitation in the claim.
Claim 20 recites “the partition value” and “the mini-batch value” in the last line. There is insufficient antecedent basis for these limitations in the claim.
Allowable Subject Matter
The following is a statement of reasons for the indication of allowable subject matter:
The prior art of record includes Lamy-Poirier, US Patent Application Publication No. US 2022/0383084 A1, Ben-Itzhak et al. (Ben-Itzhak), US Patent Application Publication No. US 2022/0292342 A1 and Amid et al. (Amid), Us Patent Application Publication No. US 2022/0253713 A1.
Lamy-Poirier discloses a method including (i) assigning sequentially ordered layers of a machine learning model to a plurality of computer nodes, each of the layers being assigned to exactly one of the nodes; (ii) dividing training data into micro-batches; (iii) forward-propagating the micro-batches through the model, each node operating in parallel to generation respective activation states for the micro-batches with their assigned layers, and with the activation states being communicated between the nodes according to the layers’ sequential ordering; and (iv) backward-propagating the micro-batches through the model, each node operating in parallel to generate respective error states for the micro-batches with their assigned layers, wherein each of the nodes completes the backward-propagation of all micro-batches through a given layer prior to performing backward-propagation through any layer that precedes the given layer in the sequential ordering (Abstract and paragraphs [0003], [0164]). Lamy-Poirier further discloses two main forms of parallelism, which are data parallelism in which each device (e.g., computational node) performs the same computation on a subset of the data, and model parallelism in which the model and computation are split between devices, wherein in data parallelism, a batch of training samples is split between the devices, and each in dependently processing a single micro-batch, and then sharing the resulting gradients with each other to generate and overall gradient (paragraph [0129]). Lamy-Poirier further discloses in Figure 6 and paragraph [0150] an example schedule of computation and network use with data parallelism for non-layered gradient accumulation and layered gradient accumulation, wherein the letters represent the different layers or contiguous sets of layer of the model being trained, with the capitalization representing the forward (lowercase) and backward (uppercase) passes. Lamy-Poirier further discloses in paragraph [0152], the input batch was split into micro-batches in the “Layered Gradient Accumulation”, and all of the micro-batches are processed (forward and backward) for a given layer before proceeding to the next layer.
Ben-Itzhak discloses a client can receive from a server a copy of a neural network form a server including N layers, the client can compute a client gradient comprising gradient values for the N layers, determine a partial client gradient comprising gradient values for a first K out of the N layers, and determine an output of the K-th layer of the copy, the output being a result of processing performed by the first K layers on the one or more data instances, and the client can then transmit the partial client gradient along with the output of the K-th layer of the neural network (which is generated in response to input training data instances X) and the training labels for X (i.e., Y or alternatively a loss vector instead of Y) to the server (Abstract and paragraph [0008]). Ben-Itzhak further discloses client then computes a loss vector (error) for using a loss function that takes f(X) and the training labels of X as input, uses backpropagation to compute a gradient for the entirety of the copy of neural network based on the loss vector (paragraphs [0013], [0019]).
Amid discloses a system implemented as computer programs on one or more computers in one or more locations that trains a neural network that processes network inputs to generate network outputs, wherein the system trains the neural network using layer-wise losses, so that weight updates for the layers of the neural network can be computed in parallel for each of the layers in the neural network (paragraph [0004]). Amid further discloses the neural network includes multiple layers that each have respective weights, and each of the multiple layers is configured to receive a layer input and apply to the respective weights for the layer to the layer input to generate a pre-activation for the layer, and each of the multiple layers is then configured to apply a transfer function of the layer to the pre-activation to generate a post-activation, i.e., the layer output of the layer, and then provide the post-activation to the one or more other layers of the neural network that are conjured to receive input from the layer according to the neural network architecture (paragraphs [0035]-[0037]). Amid further discloses to train the neural network, the training system repeatedly updates the weights of the multiple layers using the training data at different training steps to minimize a task loss function (paragraph [0039]). Amid further discloses the system obtains a batch that includes one or more training inputs and a respective label for each training input, wherein the label for each training input identifies a target output for the training input that should be generated by performing the particular machine learning task on the training input (paragraph [0050]). Amid further discloses the system performs a forward pass through the neural network to generate a respective training network output for each training input in the batch, and the system performs a backward pass through the neural network using, for each training input, the training output for the training input and the label for the training input to determine, for each layer of the neural network and for each training input, an estimated target for the neural network layer (paragraphs [0051]-[0052]).
The primary reason for the allowance of the claims in this case, is the inclusion of the specific details “during each of a first plurality of consecutive time periods, operations of applying the server partition to each activation among a preceding set of activations to obtain a current set of output instances, the preceding set of activation received during a preceding time period among the first plurality of consecutive time periods”, and during each of a second plurality of consecutive time periods, operations of transmitting, to the computation device, a preceding set of gradient vectors of the layer bordering the device partition computed during a preceding time period among any of the first plurality of consecutive time periods and the second plurality of consecutive time period and a preceding set of loss values of the loss function obtained in the preceding time period among any of the first plurality of consecutive time periods and the second plurality of consecutive time periods”, as are now included in all the independent claims, in combination with the other elements recited, which is not found in the prior art of record.
However, Claims 1-20 are only allowed if claims are rewritten to overcome the claimed objections and the rejection under 112.
Any comments considered necessary by applicant must be submitted no later than the payment of the issue fee and, to avoid processing delays, should preferably accompany the issue fee. Such submissions should be clearly labeled “Comments on Statement of Reasons for Allowance.”
Conclusion
Any inquiry concerning this communication should be directed to CHAU T NGUYEN at telephone number (571)272-4092. The examiner can normally be reached on M-F from 8am to 5pm (PT).
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) Form at https://www.uspto.gov/patents/uspto-automated-interview-request-air-form.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Cesar Paula, can be reached at telephone number 5712724128. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from Patent Center and the Private Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from Patent Center or Private PAIR. Status information for unpublished applications is available through Patent Center and Private PAIR for authorized users only. Should you have questions about access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
/CHAU T NGUYEN/Primary Examiner, Art Unit 2145