DETAILED ACTION
This action is in response to the original filing of 3-31-2023. Claims 1-20 are pending and have been considered below:
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 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 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.
Claims 1, 3-12 and 14-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over “Learning to learn with quantum neural networks via classical neural networks” Pages 1-12, 7-11-2019 Verdon et al. (“Verdon”) in view of Del Bimbo et al. (“Del” 20230058527 A1).
Claim 1: Verdon discloses a method, executable by a processor of an electronic device(Section 2-A and Figure 1;CPU),
comprising:
receiving a dataset associated with a machine learning task (Section 3-C; provides 1000 instances)
preparing an input quantum state based on the received dataset(Section 2A-B and Figures 1-2; states are set up);
preparing a Variational Quantum Circuit (VQC) on a quantum computer to function as a Quantum Neural Network (QNN)
(Section 1, 2A, Paragraph 2 and Section 3 Paragraph 2; variational quantum algorithm utilized);
executing, for a current time-step of a set of time-steps, operations comprising: reading content of a state buffer to determine whether the state buffer is empty or includes past information on parameters of the QNN (Figure 1, CPU; Figure 2, Section 2A, Paragraph 2, Section 2B, Paragraphs 4-5; LSTM use buffer like mechanisms, keeps some data in memory);
selecting values for the parameters of the QNN based on the content (Section 2B, Paragraph 5; parameters mapped);
preparing an input for an optimizer network based on the selected values of the parameters of the QNN (Figures 1-2 and Section 2B, Paragraph 5; black box optimizer);
computing an output by applying the optimizer network on the input (Figures 1-2 and Section 2A and Section 2B, Paragraph 5; black box optimizer);
updating the values of the parameters of the QNN based on the output(Figures 1-2 and Section 2A and Section 2B, Paragraph 5; parameters updated);
and passing the input quantum state and the updated values of the parameters to the QNN to obtain a current value of a cost function of the QNN from the quantum computer (Figure 2 (loss value evaluated) and Section 2B, Paragraphs 4-5; parameters updated);
updating the state buffer based on the current value of the cost function and the updated values of the parameters(Figure 2 (loss value evaluated) and Section 2B, Paragraphs 4-5; LSTM holds data in memory and parameters updated);
and training the QNN on the machine learning task by repeating the execution of the operations using the updated values of the parameters, (Section 2A, 2B, 3C and Page 4 Section 1; provides efficient parameter update related to cost function).
To further capture repeating until the current value of the cost function is below a cost threshold, Del is provided.
Del discloses a quantum functionality where a cost function is computed on a loop until the result is below a desired threshold (Paragraphs 14-20). Therefore it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to apply a known technique to a known device ready for improvement and provide the looped computation for attaining the threshold. This functionality could be incorporated with the cost function determination disclosed in Verdon. One would have been motivated to provide the functionality because once convergence is achieved, the model is optimized which provides the best response.
Claim 3: Verdon and Del discloses a method according to claim 1, wherein the selection of the values for the parameters of the QNN is performed by sampling past values of the parameters of the QNN included in the state buffer, and wherein the sampling is performed based on the determination that the state buffer includes the past information on the parameters of the QNN (Verdon: Figure 1-2 (previous iteration parameters feed into QPU), Section 2B Paragraphs 4-5 (LSTM use buffer like mechanisms, keeps some data in memory) and Page 8, Paragraph 2).
Claim 4: Verdon and Del disclose a method according to claim 1, further comprising: retrieving past values of the cost function obtained for two consecutive time-steps that precede the current time-step of the set of time-steps; computing a difference between the past values; constructing an input vector by concatenating the difference and the selected values of the parameters; and preparing the input for the optimizer network based on an application of an exponential scaling to the input vector and a normalization factor.
(Verdon: Section 3C, Paragraph 3 (normalization), Page 8; Additionally, gradient-based training required backpropagation through time for the temporal hybrid quantum-classical computational graph, which added further linearly-scaling overhead and Page 3, Column 2, Paragraph 2 “a recurrent neural network is a network where, for each item in a sequence, the network accepts an input vector, produces an output vector, and potentially keeps some data in memory for use with subsequent items. The computational graph of a RNN usually consists of many copies of the network, each sharing the same set of parameters, and each representing a time step. The recurrent connections, which can be interpreted as self-connections representing the data ow over time, can be represented as a connections between copies of the network representing subsequent time steps.”)
Claim 5: Verdon and Del disclose a method according to claim 1, wherein the optimizer network is a Long-Short Term Memory (LSTM) network that is associated with meta-parameters and is configured to maintain a hidden state of the LSTM network. (Verdon: Page 3; Figure 2 (hidden state) and Page 3-4; Paragraph 1)
“The meta-learning neural network architecture used in this paper is depicted in Figure 2, there, an LSTM 4 recurrent neural network is used to recursively propose updates to the QNN parameters, thereby acting as the classical black-box optimizer for quantum-classical optimization. At a given iteration, the RNN receives as input the previous QNN querys estimated cost function expectation yt ˜ p(y θt), where yt is the estimate of Ht, as well as the parameters for which the QNN was evaluated θt. The RNN at this time step also receives information stored in its internal hidden state from the previous time step ht. The RNN itself has trainable parameters , and hence it applies the parameterized mapping
which generates a new suggestion for the QNN parameters as well as a new hidden state.
Claim 6: Verdon and Del disclose a method according to claim 5, wherein the input for the optimizer network further includes the hidden state of the LSTM network (Verdon: Page 3; Figure 2 (hidden state) and Page 3-4; Paragraph 1).
Claim 7: Verdon and Del disclose a method according to claim 1, wherein the values of the parameters of the QNN are updated by: transforming the output by applying a non-linear activation function on the output of the optimizer network (Verdon: Section 3-C; squashed function acts as a non-linear activation function);
multiplying the transformed output with a value of a learning rate parameter to generate update values for the current time-step; and adding the update values to the selected values of the parameters of the QNN in the current time-step (Verdon: Page 3, Section 2B, Column 2, Paragraph 2-Page 4 Paragraph 1; inputs received at a time step).
Claim 8: Verdon and Del disclose a method according to claim 1, wherein the state buffer further includes past values of the cost function for one or more time-steps which precede the current time-step, and the past information corresponds to past values of the parameters of the QNN for one or more time-steps which precede the current time-step (Verdon: Figure 2, values accessed from memory, also provides one or more time-steps and Page 9, Paragraphs 2 and 4-5; cost function evaluation and parameter comparison).
Claim 9: Verdon and Del disclose a method according to claim 1, further comprising comparing the current value of the cost function for the current time-step with a past value of the cost function for a time-step that precedes the current time-step of the set of time-steps, wherein the past value is included in a memory of the electronic device, and the state buffer is updated to include the current value of the cost function and the updated values of the parameters based on the comparison (Verdon: Figure 1-2 and Page 5, Column 1 (cost function), Section 2B Paragraphs 4-5 (LSTM use buffer like mechanisms, keeps some data in memory)).
Claim 10: Verdon and Del disclose a method according to claim 1, further comprising: initializing meta-parameters of the optimizer network with meta-parameter values before the execution of the operations; evaluating a meta-loss function for the optimizer network after an end of the set of time-steps based on a value of the cost function obtained for each time- step of the set of time-steps (Verdon: Page 4, Section 1; meta loss function to optimize the parameters over time interval (time step) and Page 5, Section 1; loss function obtained over time steps); and updating the meta-parameter values based on the meta-loss function).
Claim 11: Verdon and Del disclose a method according to claim 1, wherein the VQC is a parametrized quantum circuit that includes a set of quantum gates in a specific arrangement, and the set of quantum gates represents a set of operations to be performed on a set of qubits of the quantum computer (Verdon: Figure 1 (arrangement), Page 2, Section 2A, Paragraph 2 and Page 6, Paragraph 1; gates scale linearly).
Claims 12 and 20 are similar in scope to claim 1 and therefore rejected under the same rationale.
Claim 14 is similar in scope to claim 3 and therefore rejected under the same rationale.
Claim 15 is similar in scope to claim 4 and therefore rejected under the same rationale.
Claim 16 is similar in scope to claim 5 and therefore rejected under the same rationale.
Claim 17 is similar in scope to claim 7 and therefore rejected under the same rationale.
Claim 18 is similar in scope to claim 9 and therefore rejected under the same rationale.
Claim 19 is similar in scope to claim 10 and therefore rejected under the same rationale.
Claims 2 and 13 is/are rejected under 35 U.S.C. 103 as being unpatentable over “Learning to learn with quantum neural networks via classical neural networks” Pages 1-12, 7-11-2019 Verdon et al. (“Verdon”) and Del Bimbo et al. (“Del” 20230058527 A1)in further view of Pan et al. (“Pan” 20210248087 A1).
Claim 2: Verdon and Del disclose a method according to claim 1, but may not explicitly disclose each element further comprising: initializing the parameters of the QNN with initial values before the execution of the operations; and instantiating the state buffer based on the initialization of the parameters, wherein the state buffer is a double-sided queue that is empty when the parameters of the QNN are initialized with the initial values. Pan is provided because it discloses a buffer functionality which utilizes a double ended queue (Paragraph 33, deque). The deque could be utilized within the memory storage capabilities provided by the modified Verdon. Therefore it would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to apply a known technique to a known device ready for improvement and incorporate the deque utility. One would have been motivated to provide the functionality because it offers flexibility and efficiency of operations at both ends of a sequence.
Claim 13 is similar in scope to claim 2 and therefore rejected under the same rationale.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
20220222548 A1 [0045]
META LEARNING VIA LOSS BECHTLE ET AL 1-19-2021
Applicant is required under 37 C.F.R. § 1.111(c) to consider these references fully when responding to this action.
It is noted that any citation to specific pages, columns, lines, or figures in the prior art references and any interpretation of the references should not be considered to be limiting in any way. A reference is relevant for all it contains and may be relied upon for all that it would have reasonably suggested to one having ordinary skill in the art. In re Heck, 699 F.2d 1331, 1332-33, 216 U.S.P.Q. 1038, 1039 (Fed. Cir. 1983) (quoting In re Lemelson, 397 F.2d 1006, 1009, 158 U.S.P.Q. 275, 277 (C.C.P.A. 1968)).
In the interests of compact prosecution, Applicant is invited to contact the examiner via electronic media pursuant to USPTO policy outlined MPEP § 502.03. All electronic communication must be authorized in writing. Applicant may wish to file an Internet Communications Authorization Form PTO/SB/439. Applicant may wish to request an interview using the Interview Practice website: http://www.uspto.gov/patent/laws-and-regulations/interview-practice.
Applicant is reminded Internet e-mail may not be used for communication for matters under 35 U.S.C. § 132 or which otherwise require a signature. A reply to an Office action may NOT be communicated by Applicant to the USPTO via Internet e-mail. If such a reply is submitted by Applicant via Internet e-mail, a paper copy will be placed in the appropriate patent application file with an indication that the reply is NOT ENTERED. See MPEP § 502.03(II).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SHERROD KEATON whose telephone number is 571-270-1697. The examiner can normally be reached 9:30am to 5:00pm.
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 MICHELLE BECHTOLD can be reached at 571-431-0762. 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.
/SHERROD L KEATON/ Primary Examiner, Art Unit 2148
11-22-2025