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 1-25 have been examined.
Claim Objections
Claim 11 is objected to because of the following informalities: the term “zi“ in line 3 appears to be a typo of “zj.” 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 2, 5-12, 15, 17, 20 and 22-24 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.
The term “regular” in claim 2 is a relative term which renders the claim indefinite. The term “regular” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. For the purpose of further examination, the limitation will be disregarded.
Claim 5 is rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being incomplete for omitting essential structural cooperative relationships of elements, such omission amounting to a gap between the necessary structural connections. See MPEP § 2172.01. After reciting the decomposing “into two decomposition parts,” the following limitations are not integrated to provide any particular influence on the rest of the claim. That is, the limitations of P* ∈ ℝlxd, uk ∈ ℝl, and vk ∈ ℝd are provided as nomenclature elements without being integrated with the decomposing, or the other elements of parent claim 1. The claim does not require that any calculations are performed on these elements and therefore they have no limiting effect.
Claims 5-8, 10-12, 17 and 22-24 include numerous mathematical symbols, variables and operators relevant in the field of linear algebra. However, the use of variables must be presented in terms of definitions of what they represent. Without an associated definition, they cannot be definitely assessed. For example, claim 5 includes “uk ∈ ℝl, vk ∈ ℝd are task-specific vectors for each task k.” This could be presented as uk ∈ ℝl, vk ∈ ℝd , wherein k is an integer, and wherein uk and vk are task-specific vectors for each task k, and uk and vk are elements of l-dimensional and d-dimensional Euclidean space.” Also noted particularly in claim 6, the term “P*” is not defined and it is not clear. Clarification is required.
Claim 9 is rejected as being dependent upon a rejected base claim.
Claims 15 and 20 have limitations similar to claim 2 and are rejected for the same reason as claim 2 above.
Claim Rejections - 35 USC § 103
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.
Claim(s) 1, 3-4, 7, 9, 13-14, 16, 18-19, 21 and 24-25 is/are rejected under 35 U.S.C. 103 as being unpatentable over “Factorizing Knowledge in Neural Networks” by Yang (“Yang”) in view of U.S. Patent Application Publication 20230104662 by Fatemi et al. ("Fatemi") and “LoRA: Low-Rank Adaptation Of Large Language Models” by Hu et al. (“Hu”).
In regard to claim 1, Yang discloses:
1. A method comprising: See Yang, Fig. 1(d) and Fig. 2, broadly depicting methods.
decomposing, using a hardware processor, a source task … of each of a plurality of source tasks as a multiplication of a shared … [network] shared across the source tasks and a … task-specific … [network]; Yang p. 75, “Each factor network comprises a shared common-knowledge network (CKN) and a task-specific network (TSN).” Also Yang, p. 78, section 3.1: “Given a multi-task teacher model T that is able to predict K tasks simultaneously, KF aims to construct K factor networks
{
S
j
}
j
=
1
K
, each of which, again, tackles one task independently.” Also p. 78, section 3.2, “Specifically, a factor network Sj for the j-th task comprises two modular networks: a Common Knowledge Network (CKN) SC(·; ΘSC ) which is shared across all tasks, and a Task-specific Network (TSN) STj (·; ΘST ) which is task-exclusive.”
Yang does not expressly disclose: task prompt or shared prompt matrix. Fatemi teaches this. See Fatemi ¶ 0039, “To obtain an integrated word embedding matrix, gender-neutral prompt module 134 may concatenate Wx and Wp into matrix Wemb=Concat(Wx, Wp), which represents both new and frozen parameters of the embedding layer.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Fatemi’s prompt matrices with Yang’s task factor network in order to specialize a model without forgetting information of a base model, as essentially suggested by Fatemi (see ¶ 0019).
Yang and Fatemi do not expressly disclose: low-rank … matrix. This is taught by Hu. See Hu, p. 4, section 4.1: “For a pre-trained weight matrix W0 ∈ ℝdxk, we constrain its update by representing the latter with a low-rank decomposition W0 + ΔW + BA, where B ∈ ℝdxr, A ∈ ℝrxk , and the rank r
≪
min(d,k).” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Hu’s low-rank matrix with Yang’s networks in order to efficiently switch tasks while reducing storage and overhead and make training more efficient as suggested by Hu (see section 1, page 2).
Yang and Hu also teaches:
performing, using the hardware processor, prompt distillation to transfer multitask knowledge to the shared prompt matrix by distilling knowledge from the source task prompts; Yang, section 3.2, “For each input sample, SC is adopted to extract the task-agnostic feature z:”
performing, using the hardware processor, low-rank multiplicative updates to the shared prompt matrix to transfer the multitask knowledge to one or more target tasks; and Yang, section 3.2, “On the contrary, STj learns the task-related knowledge tj from the input x, which together with z is processed by a task head Hj to make the final prediction.” Also see Hu, section 4.1, discussing low-rank multiplicative updates.
using the hardware processor, carrying out the one or more target tasks in accordance with the transferred knowledge. See Yang, section 4, describing execution of tasks.
In regard to claim 3, Yang et al. also teaches:
3. The method of claim 1, wherein each source task is represented as a prompt matrix. See Yang and Fatemi as cited above.
In regard to claim 4, Yang also discloses:
4. The method of claim 1, wherein the decomposition enables efficient knowledge sharing across source tasks while still allowing each source task to maintain corresponding parameters for encoding task-specific knowledge. Yang, p. 75, “given a pretrained teacher, KF decomposes it into several factor networks, each of which masters one specific knowledge factorized from the teacher, while remaining disentangled with respect to others.”
In regard to claim 7, Yang also discloses:
7. The method of claim 1, further comprising: obtaining a teacher prompt Pkt for a k-th source task using prompt tuning; and performing multitask training on the source tasks to jointly learn the single shared soft prompt via a knowledge distillation loss function. p. 78, section 3.2, “We accordingly define a structure factorization objective L(j)sf to enforce each single-task factor network to imitate the teacher’s prediction while minimizing the supervised loss:”
In regard to claim 9, Yang also discloses:
9. The method of claim 7, wherein a knowledge distillation loss is configured to transfer cross-task knowledge into the shared prompt matrix. Yang, p. 78, section 3.2, “We accordingly define a structure factorization objective L(j)sf to enforce each single-task factor network to imitate the teacher’s prediction while minimizing the supervised loss:”
In regard to claim 13, Yang does not expressly disclose:
13. The method of claim 1, wherein the one or more target tasks comprise natural language processing. This is taught by Hu. See Hu, p. 1, “natural language processing.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Hu’s NLP with Yang’s factorization in order to provide NLP functionality as suggested by Hu.
In regard to claim 14, Yang discloses:
14. A computer program product, comprising: one or more tangible computer-readable storage media and program instructions stored on at least one of the one or more tangible computer-readable storage media, the program instructions executable by a processor, the program instructions comprising: See Yang, section 4, describing experimentation using the “KF” computer program product. Note that such experiments require computer implementation including storage media, instructions, and processor execution in order to produce results. Also see Fatemi, Fig. 1, elements 110, 120 and 130, depicting a computer program product.
All further limitations of claim 14 have been addressed in the above rejection of claim 1.
In regard to claims 16 and 18, parent claim 14 is addressed above.
All further limitations of claims 16 and 18 have been addressed in the above rejections of claims 3 and 13, respectively..
In regard to claim 19, Yang discloses:
19. A system comprising: a memory; and at least one processor, coupled to said memory, and operative to perform operations comprising: See Yang, section 4, describing experimentation using the “KF” computer program product. Note that such experiments require computer system implementation including storage media, instructions, and processor execution in order to produce results. Also see Fatemi, Fig. 1, elements 110, 120 and 130, depicting a computer system.
All further limitations of claim 19 have been addressed in the above rejection of claim 1.
In regard to claims 21 and 24-25, parent claim 19 is addressed above.
All further limitations of claims 21 and 24-25 have been addressed in the above rejections of claims 4, 7 and 13, respectively.
Claim(s) 2, 5, 15, 17, 20 and 22 is/are rejected under 35 U.S.C. 103 as being unpatentable over Yang in view of Fatemi and Hu as applied above, and further in view of U.S. Patent Application Publication 20230419027 by Pang et al. ("Pang").
In regard to claim 2, Yang et al. does not expressly teach:
2. The method of claim 1, wherein the decomposition is learned by performing the prompt distillation obtained from [] prompt tuning. This is taught by Pang. See Pang, ¶ 0015-0016, e.g. “prompt-tuning.” “This vector modulates a frozen model, via soft prompts which are a simple prompt transformation (the prompt generator in FIG. 3) of the basis skill vector, to generate an answer for the instance. The latent space and prompt transformation are learned end-to-end in upstream pre-training on source tasks.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Pang’s prompt tuning with Yang’s decomposition in order to improve parameter efficiency as suggested by Pang (see ¶ 0016).
In regard to claim 5, Yang also discloses:
5. The method of claim 1, wherein the decomposing of the source task prompt decomposes a k-th task into two decomposition parts, Yang p. 75, “Each factor network comprises a shared common-knowledge network (CKN) and a task-specific network (TSN)
Yang does not expressly disclose: where P* ∈ ℝlxd denotes an initial shared task prompt across all source tasks and uk ∈ ℝl, vk ∈ ℝd are task-specific vectors for each task k, the task-specific vectors forming a rank-one matrix Wk = uk ∙ vkT, which has a same dimension as the initial shared task prompt. This is taught by Pang. See ¶ 0015, “Specifically, a shared latent space is assumed, among all source and target tasks, where each vector in the space captures a basis skill to do a particular task. Given an instance (from either a source task or a target task), it is first encoded into an instance representation vector and then queries the latent space, which yields a skill vector for this instance.” Note that the specific claimed matrix and vector elements are provided in terms of definitions, and do not combine to further limit the claim. In this sense, a dot product of Pang’s instance representation vector and skill vector forms a rank-one matrix taken from the same latent space of the initial shared task prompt by nature of the mathematical operations described. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Pang’s task vector analysis with Yang’s decomposition in order to save memory and time as suggested by Pang (see ¶ 0016).
In regard to claims 15 and 17, parent claim 14 is addressed above.
All further limitations of claims 15 and 17 have been addressed in the above rejections of claims 2 and 5, respectively..
In regard to claims 20 and 22, parent claim 14 is addressed above.
All further limitations of claims 20 and 22 have been addressed in the above rejections of claims 2 and 5, respectively.
Claim(s) 6 and 23 is/are rejected under 35 U.S.C. 103 as being unpatentable over Yang in view of Fatemi, Hu and Pang as applied above, and further in view of U.S. Patent Application Publication 20200193228 by Lu et al. ("Lu") and U.S. Patent Application Publication 20160019456 by Annapureddy et al. ("Annapureddy").
In regard to claim 6, Yang as modified above also teaches:
6. The method of claim 5, further comprising parametrizing a final shared task prompt
P
^
for a k-th source task as: … wherein a parameterization of prompt decomposition captures general information of a corresponding source task S by weights shared across the source tasks and weights Wk encode task-specific knowledge in a low-rank subspace. Yang, Fig. 1(d), depicting knowledge factorization. Also see section 1 on p. 4, “Our approach decomposes a pretrained teacher into factor networks that are task-wise disentangled.”
Yang does not expressly disclose:
P
^
k =
P
*
∘
W
k
=
P
*
∘
(
u
k
∙
v
k
T
)
where ∘ denotes a Hadamard product (element-wise product) between two matrices,
However, Lu teaches parametrization using Hadamard product. See Lu ¶ 0089, “In addition, using Hadamard product may also reduce computing burden, thereby improving the execution efficiency of the method.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Lu’s Hadamard product with Yang’s factorizing in order to improve execution efficiency as suggested by Lu.
Also, Annapureddy teaches matrix decomposition. See ¶ 0074, “In one example, suppose Wi=UiViT where Ui is a column vector and ViT is a row vector. Then the 2-D convolution operation X*Wi may be decomposed by first convolving each column of matrix X with the column vector Ui and then convolving each row of the resulting matrix with the row vector ViT.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Annapureddy’s decomposition with the parametrization of Yang and Lu in order to simplify computation as suggested by Annapureddy.
In regard to claim 23, parent claim 22 is addressed above.
All further limitations of claim 23 have been addressed in the above rejection of claim 6.
Claim(s) 8 is/are rejected under 35 U.S.C. 103 as being unpatentable over Yang in view of Fatemi and Hu as applied above, and further in view of Lu and Annapureddy.
In regard to claim 8, Yang and Hu also teach:
8. The method of claim 7, further comprising: randomly initializing a corresponding student prompt as … where all student prompts share P* and have corresponding task-specific vectors.
See Yang, Fig. 2, depicting student “Task-Specific” networks. Also see section 3.2:
Specifically, a factor network Sj for the j-th task comprises two modular networks: a Common Knowledge Network (CKN) SC (·; ΘSC ) which is shared across all tasks, and a Task-specific Network (TSN) STj (·; ΘST ) which is task-exclusive.
Also see Hu, section 4.1 on p. 4, “We illustrate our reparametrization in Figure 1. We use a random Gaussian initialization for A and zero for B, so ∆W = BA is zero at the beginning of training. We then scale ∆Wx by α , where α is a constant in . When optimizing with Adam, tuning is roughly the same as tuning the learning.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Hu’s initialization with Yang’s networks in order to assist with parameter tuning as suggested by Hu.
Yang does not expressly disclose:
P
k
S
^
=
P
*
∘
u
k
∙
v
k
T
However, Lu teaches parametrization using Hadamard product. See Lu ¶ 0089, “In addition, using Hadamard product may also reduce computing burden, thereby improving the execution efficiency of the method.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Lu’s Hadamard product with Yang’s factorizing in order to improve execution efficiency as suggested by Lu.
Also, Annapureddy teaches matrix decomposition. See ¶ 0074, “In one example, suppose Wi=UiViT where Ui is a column vector and ViT is a row vector. Then the 2-D convolution operation X*Wi may be decomposed by first convolving each column of matrix X with the column vector Ui and then convolving each row of the resulting matrix with the row vector ViT.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Annapureddy’s decomposition with the parametrization of Yang and Lu in order to simplify computation as suggested by Annapureddy.
Claim(s) 10 is/are rejected under 35 U.S.C. 103 as being unpatentable over Yang in view of Fatemi and Hu as applied above, and further in view of U.S. Patent Application Publication 20220335303 by Haidar et al. ("Haidar").
In regard to claim 10, Yang does not expressly disclose:
10. The method of claim 9, wherein a first distillation loss matches output probability distributions of student models and teacher models through minimizing a corresponding KL-Divergence,
ℒLogits =
∑
k
∑
i
∈
S
k
K
L
(
P
y
i
P
k
t
;
x
i
,
P
y
i
[
P
k
S
^
;
x
i
]
)
.
This is taught by Haidar. See Haidar ¶ 0079, “KL divergence is used to measure the difference between the student predicted logits 306 and the teacher predicted logits 310. Minimizing the KL divergence, and therefore the KD loss 314, by adjusting a plurality of the values of the learnable parameters of the student model 234 should result in the student model 234 learning to output student inference data 34 that is close to (i.e. similar to) the teacher inference data 24 output by the teacher model 20.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use KL divergence of Haidar with Yang’s distillation in order to result in a student model learning to output student inference data that is close to (i.e. similar to) teacher inference data output by the teacher model, as suggested by Haidar (see ¶ 0079).
Claim(s) 11 is/are rejected under 35 U.S.C. 103 as being unpatentable over Yang in view of Fatemi, Hu and Haidar as applied above, and further in view of U.S. Patent Application Publication 20220230095 by Stergioudis ("Stergioudis") and U.S. Patent 11200497 to Yan et al. ("Yan").
In regard to claim 11, Yang does not expressly disclose the claimed limitations.
However, the limitations are further taught by Stergioudis and Yan as follows:
11. The method of claim 10, further comprising controlling, using a temperature T, a smoothness of an output distribution for both the teacher model and the student model as
1
z
e
x
p
(
z
j
T
)
[where zj] is a logit score for class j and Z is a normalization factor and This is taught by Stergioudis. See ¶ 0085-0086, “For example, the probability pi of class i can be determined from the logits z using the following smoothing technique 224 indicated by Function 1
PNG
media_image1.png
99
260
media_image1.png
Greyscale
where T is the temperature parameter.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use the smoothing technique of Stergioudis with Yang’s probability distribution in order to provide more information during distillation as suggested by Yang (see ¶ 0086).
wherein an additional mean squared loss on hidden states of the teacher model is defined as:
L
H
i
d
d
e
n
=
∑
k
∑
i
∈
S
k
(
H
k
i
S
-
H
k
i
t
)
2
where
H
k
i
S
,
H
k
i
t
denotes hidden states of teacher and student networks, respectively, comprising a sequence of hidden vectors for an i-th input. Yan col. 9 lines 19-22, “a loss function based on a mean-square error of a difference between hidden representations of one or more layers of the student network and the teacher network;” Also col. 9, lines 33-40, “Lhid=ΣMSE (HiS, HiT) refers to the difference between hidden representations of the student network and the teacher network, and Lprd=−softmax(zT)*log_softmax(zS/temp) represents the soft cross-entropy loss between the logits of the student network and the teacher network (temp is the temperature parameter in the context of knowledge distillation, zT is the prediction logits of the teacher network, and zS is the prediction logits of the student network).” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Yan’s loss function with Yang’s network in order to train a neural network as essentially suggested by Yan and known to those of ordinary skill in the art.
Claim(s) 12 is/are rejected under 35 U.S.C. 103 as being unpatentable over Yang in view of Fatemi and Hu as applied above, and further in view of Yan and U.S. Patent Application Publication 20240070394 by Peng et al. ("Peng").
In regard to claim 12, Yang also discloses:
12. The method of claim 9, further comprising training a student source prompt for obtaining a single soft task prompt to be transferred to a target task using a total loss function defined by: ℒTotal = ℒPLM + λ(ℒLogits + ℒHidden), wherein … λ is a weight to balance an impact of distillation loss terms. Yang, p. 78, section 3.2:
PNG
media_image2.png
52
269
media_image2.png
Greyscale
… For example, L(j) sup may take the form of L2 norm for regression and cross-entropy for classification, while L(j) kt may take the form of soft-target.”
Yang does not expressly disclose: ℒHidden … wherein
L
P
L
M
=
∑
k
L
P
L
M
k
represents aggregated task losses for source tasks, and … However, Yan teaches loss in terms of logits and hidden representations. See Yan col. 9 lines 19-22, “a loss function based on a mean-square error of a difference between hidden representations of one or more layers of the student network and the teacher network;” Also col. 9, lines 24-25, “For example, the distillation loss function may be defined as Ldistil=Lemb+Latt+Lhid+Lprd”. Also col. 9, lines 33-40, “Lhid=ΣMSE (HiS, HiT) refers to the difference between hidden representations of the student network and the teacher network, and Lprd=−softmax(zT)*log_softmax(zS/temp) represents the soft cross-entropy loss between the logits of the student network and the teacher network (temp is the temperature parameter in the context of knowledge distillation, zT is the prediction logits of the teacher network, and zS is the prediction logits of the student network).” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Yan’s loss function with Yang’s network in order to train a neural network as essentially suggested by Yan and known to those of ordinary skill in the art.
Peng teaches calculation of PLM loss. Peng ¶ 0025, “After receiving the first training input sequence, the PLM 100 generates a first output logit, lx,1, 114. Next, the softmax decoder 104 generates a first predicted source output, Y, 118 from the first output logit, lx,1, 114. Next, the first predicted source output 118 and a corresponding source output may be compared through a loss function. The soft prompt 110a-b is updated based on the loss function via backpropagation while the PLM 100 remains frozen.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Pang’s PLM loss with Yang’s distillation loss calculation in order to train a language prompt model suggested by Peng (e.g. ¶ 0020).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
U.S. Patent Application Publication 20200026977 by Lee et al. ("Lee") Lee ¶ 0075, “Decomposition may be a compression method for reducing the size of the weight parameter by performing approximated decomposition of a weight matrix, which is a collection of weight parameters, or a tensor. Knowledge Distillation may be a compression method for generating and training a Student model smaller than an original model by setting the original model as a Teacher model. The Student model may be generated through the above-described Pruning, Decomposition, or Quantization.”
U.S. Patent Application Publication 20240362487 by ElKordy et al. ("ElKordy") ElKordy ¶ 0059, “Parameter efficient tuning may generally be achieved by representing the updated weight matrix as a product of two low-rank decomposition matrices, matrix A and matrix B (also referred to as LoRA blocks), and optimizing/training the parameter weights of the decomposition matrices A and B. Fine-tuning the parameters of the low-rank decomposition matrices reduces the number of parameters to be tuned during the second stage of the training.”
U.S. Patent Application Publication 20240256865 by Jain et al. ("Jain") Jain ¶ 0009, “For example, the system can use a transformer that approximates attention via low-rank decomposition of the attention matrix.” ¶ 0117, “FIG. 3 shows that N2 is the sum of a product between a feature mapping of the key and a transposed value, multiplied by a discount factor, over previous training stages.” Also ¶ 0028, “The system can also be configured to pre-train or prompt-tune models”
U.S. Patent Application Publication 20240404505 by Nguyen et al. ("Nguyen") Nguyen ¶ 0100 “One method of rank decomposition may include rank-1 factorization. In rank-1 factorization, a matrix is factorized into the outer product of a column vector and a row vector. … In some examples, the adaptive weights 421 may be expressed as the outer product of two vectors”
Any inquiry concerning this communication or earlier communications from the examiner should be directed to James D Rutten whose telephone number is (571)272-3703. The examiner can normally be reached M-F 9:00-5:30 ET.
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, Li B Zhen can be reached at (571)272-3768. 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.
/James D. Rutten/Primary Examiner, Art Unit 2121
Prosecution Insights