DETAILED ACTION
This office action is in response to submission of application on 09/11/2023.
Claims 1-20 are presented for examination.
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 .
Information Disclosure Statement
The information disclosure statements (IDS) submitted on 02/28/2024, 08/20/2024, 10/23/2024, 01/08/2025, 03/19/2025, 01/23/2026, 02/24/2026, 04/28/2026, and 06/15/2026 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
Claim Objections
Claim 13 is objected to because of the following informalities: “The method of 1…” should read “The method of claim 1…” in line 1 of the claim. 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.
Claim 13 is 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 13 recites the limitation "the target state" in line 5. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, this limitation will be interpreted as referring to the target quantum state recited in parent claim 1.
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.
Claims 1, 3-4, 6, 13-15, 17, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over
Benedetti et al. (hereinafter Benedetti), “Adversarial quantum circuit learning for pure state approximation” (published 04/16/2019) in view of
Garcia-Escartin et al. (hereinafter Garcia), “The SWAP test and the Hong-Ou-Mandel effect are equivalent” (published 03/27/2013).
Regarding Claim 1,
Benedetti teaches A method for training a quantum generative adversarial network to learn a target quantum state, the method comprising: iteratively adjusting parameters of the quantum generative adversarial network until a value of a quantum generative adversarial network loss function converges, (Pg. 2, section I: “In this manuscript, we start from information theoretic arguments and derive an adversarial algorithm that learns to generate approximations to a target pure quantum state. We parametrize generator and discriminator circuits similarly to other variational approaches… Optimization is performed using an adaptive gradient descent method known as resilient backpropagation (Rprop)…” A generative adversarial network is trained to approximate (i.e. learn) a target quantum state. Algorithm 1 (pg. 6) shows the training process, wherein model parameters
w
are adjusted (see line 10) repeatedly for
i
iterations (see lines 1-2) until convergence (see line 11).)
wherein each iteration comprises:
performing an entangling operation on a discriminator network input of a discriminator network in the quantum generative adversarial network to measure a fidelity of the discriminator network input, wherein the discriminator network input comprises the target quantum state and a first quantum state output from a generator network in the quantum generative adversarial network, wherein the first quantum state approximates the target quantum state; and (Pg. 2, section II: “The task of the generator is to prepare a quantum state and fool the other player into thinking that it is the true target state… The discriminator has the task of distinguishing between the target state and the generated state. It is presented with the mixture
ρ
m
i
x
=
P
(
t
)
ρ
t
+
P
(
g
)
ρ
g
, where
P
(
t
)
and
P
(
g
)
are prior probabilities summing to one. … The discriminator performs a positive operator-valued measurement (POVM)
{
E
b
}
on the input…” Pg. 8, section II.D: “As described earlier, Helstrom [13] observed that the optimal POVM that distinguishes two states has the following particular form… Under the assumption of equal prior probabilities of 1/2, the above is minimized for a maximum overlap of the two states. Since the prior probabilities are hyperparameters, we can set them to ½ and use the swap test [33] to compute the overlap. This procedure effectively implements an optimal discriminator and provides a strong learning signal to the generator.” According to paragraphs 0010-0012 of the instant application, an entangling operation can be a swap test. Benedetti’s discriminator performs a swap test (i.e. entangling operation) on the target state and generated state (i.e. discriminator network input comprising the target quantum state and a first quantum state output) to measure overlap (i.e. fidelity), where the generated state (i.e. first quantum state output) is generated by the generator network to approximate the target state.)
performing a minimax optimization of the quantum generative adversarial network loss function to update the parameters of the quantum generative adversarial network, wherein the quantum generative adversarial network loss function is dependent on the measured fidelity of the discriminator network input. (Pg. 12, section IV: “In this work we proposed an adversarial algorithm and applied it to learn quantum circuits that can approximately generate and discriminate pure quantum states. We used information theoretic arguments to formalize the problem as a minimax game. The discriminator circuit maximizes the value function in order to better distinguish between the target and generated states. This can be thought of as learning to perform the Helstrom measurement [13]. In turn, the generator circuit minimizes the value function in order to deceive the discriminator. This can be thought of as minimizing the trace distance of the generated state to the target state. The desired outcome of this game is to obtain the best approximation to the target state for a given generator circuit layout.” Minimax optimization is performed on the value function (i.e. loss function) to update the parameters of the quantum generator and discriminator (i.e. the quantum generative adversarial network). The value function depends on the trace distance between the generated state and the target state (i.e. the fidelity of the discriminator network input).)
While Benedetti discloses measuring overlap between the target and generated states using a swap test, it also notes that “the swap test bears several disadvantages,” and “[a] potential solution is to find an efficient low-depth circuit implementing the swap test” (Benedetti, pg. 8, section II.D). Benedetti does not appear to explicitly disclose an efficient low-depth circuit implementing the swap test.
However, Garcia teaches an efficient low-depth circuit implementing the swap test (Pg. 4-5, section IV: “In this Section, we present a destructive SWAP test with no ancillas… we can proceed to simplify the SWAP test circuit… To reduce the number of gates, we notice that the ancillary qubit which carries the answer to the test is not affected by the tested qubits after the CCZ gate. We are not interested in the outcomes of the measurements on the qubits under test. We can just as well get rid of the last
H
and CNOT gates… we can just ignore the ancillary qubit and perform a SWAP test with the circuit in Figure 8.” The circuit for implementing the swap test is simplified and made efficient by reducing the number of gates and removing the ancillary qubit. Figure 3 (pg. 4) and figure 8 (pg. 5) diagram the circuits for a classical swap test using a CSWAP gate and the proposed simplified swap test, respectively. The prosed simplified swap test depicted in figure 8 has a lower circuit depth.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Benedetti and Garcia. Benedetti teaches using a generative adversarial network to learn to approximate a target quantum state, where overlap between the target and generated states can be determined by a swap test. Garcia teaches a simplified quantum circuit for implementing the swap test without requiring an ancillary qubit. One of ordinary skill would have motivation to combine Benedetti and Garcia because, according to Benedetti, the swap test “effectively implements an optimal discriminator and provides a strong learning signal to the generator,” however, “the swap test bears several disadvantages,” and “[a] potential solution is to find an efficient low-depth circuit implementing the swap test” (Benedetti, pg. 8, section II.D). Garcia provides an efficient low-depth circuit implementing the swap test, as described above.
Regarding Claim 3, Benedetti and Garcia teach The method of claim 1, as shown above.
Benedetti also teaches wherein each iteration further comprises: processing, by the generator network, an initial quantum state to output the first quantum state, the processing comprising applying a first quantum circuit to the initial quantum state, wherein i) the first quantum circuit is a parameterized quantum circuit and first quantum circuit parameters constitute parameters of the generator network included in the parameters of the quantum generative adversarial network. (Pg. 2, section II: “The task of the generator is to prepare a quantum state and fool the other player into thinking that it is the true target state. Thus, the generator is a unitary transformation G applied to some known initial state…” Pg. 4, section II.A: “[T]he generator’s unitary transformation shall be implemented by a parametrized quantum circuit applied to a known initial state.” The generator applies a parameterized quantum circuit to an initial quantum state to prepare the generated state (i.e. first quantum state). Its parameters are parameters of the generator of the quantum generative adversarial network.)
Regarding Claim 4, Benedetti and Garcia teach The method of claim 3, as shown above.
Benedetti also teaches wherein the first quantum circuit has a lower circuit depth than a quantum circuit used to produce the target quantum state. (Pg. 9, section III: “For the simulations we mock this scenario using circuits to prepare the target states… The complexity of our circuits is determined by the number of layers of gates… The generator is less complex than the target state, but it manages to produce a meaningful approximation in average.” The generator (i.e. first quantum circuit) is less complex (i.e. has a lower circuit depth) than the target state circuit (i.e. a quantum circuit used to produce the target quantum state).)
Regarding Claim 6, Benedetti and Garcia teach The method of claim 1, as shown above.
Garcia also teaches wherein the entangling operation comprises an ancilla-free swap test. (Pg. 4, section IV: “In this Section, we present a destructive SWAP test with no ancillas.”)
Regarding Claim 13, Benedetti and Garcia teach The method of claim 1, as shown above.
Benedetti also teaches wherein iteratively adjusting the parameters of the quantum generative adversarial network until a value of the quantum generative adversarial network loss function converges produces trained generator network and discriminator network parameters, and wherein the method further comprises generating the target state using the generator network and according to the trained generator network parameters. (Pg. 12, section IV: “The discriminator circuit maximizes the value function in order to better distinguish between the target and generated states… In turn, the generator circuit minimizes the value function in order to deceive the discriminator. This can be thought of as minimizing the trace distance of the generated state to the target state. The desired outcome of this game is to obtain the best approximation to the target state for a given generator circuit layout.” Parameters are adjusted to optimize the value function for (i.e. train) the discriminator and generator. The generator is then used to generate an approximation of the target state.)
Regarding Claim 14, Benedetti and Garcia teach The method of claim 1, as shown above.
Benedetti also teaches wherein performing a minimax optimization of the quantum generative adversarial network loss function to update the parameters of the quantum generative adversarial network comprises performing multiple circuit evaluations to compute gradients of the parameters of the quantum generative adversarial network. (Pg. 4-5, section II.A: “To start with, we need to compute the gradient of the value function with respect to the parameters… Hence, for each parameter
l
, we are required to execute the circuit compositions
D
G
l
+
and
D
G
l
-
on initial state
|
0
⊗
n
+
1
and measure the ancilla qubit. Because these auxiliary circuits have depth similar to that of the original circuit, estimation of the gradient is efficient… Finally, all parameters are updated by gradient descent/ascent…” Circuit evaluations are executed to compute gradients of the parameters.)
Claims 15, 17, and 19 are system claims containing substantially the same elements as method claims 1, 6, and 3, respectively. Benedetti and Garcia teach the elements of claims 1, 6, and 3, as shown above.
Benedetti also teaches A quantum generative adversarial network system implemented by one or more quantum computers (Pg. 12, section IV: “In this work we proposed an adversarial algorithm and applied it to learn quantum circuits that can approximately generate and discriminate pure quantum states. We used information theoretic arguments to formalize the problem as a minimax game… We demonstrated how to perform such a minimax game in near-term quantum devices, i.e., NISQ computers [18], and we discussed long-term implementations on universal quantum computers.”)
Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over Benedetti in view of Garcia, and further in view of
Heusel et al. (hereinafter Heusel), “GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium” (published 01/12/2018).
Regarding Claim 2, Benedetti and Garcia teach The method of claim 1, as shown above.
Benedetti and Garcia do not appear to explicitly disclose the remaining features of claim 2.
However, Heusel teaches wherein the value of the quantum generative adversarial network loss converges to a Nash equilibrium. (Pg. 2, para. 2: “We prove that GANs [generative adversarial networks] converge to a local Nash equilibrium when trained by a two time-scale update rule (TTUR), i.e., when discriminator and generator have separate learning rates.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Benedetti, Garcia, and Heusel. Benedetti teaches using a generative adversarial network to learn to approximate a target quantum state, where overlap between the target and generated states can be determined by a swap test. Garcia teaches a simplified quantum circuit for implementing the swap test without requiring an ancillary qubit. Heusel teaches training a generative adversarial network with separate learning rates for the discriminator and generator to facilitate convergence to a Nash equilibrium. One of ordinary skill would have motivation to combine Benedetti, Garcia, and Heusel because “training GANs is a game and its solution is a Nash equilibrium” (Heusel, pg. 1, para. 1).
Claims 5, 7, 16, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Benedetti in view of Garcia, and further in view of
Heya et al. (hereinafter Heya), “Variational Quantum Gate Optimization” (published 10/30/2018).
Regarding Claim 5, Benedetti and Garcia teach The method of claim 1, as shown above.
Garcia also teaches wherein the entangling operation comprises a [parameterized] entangling operation that approximates a swap test. (Pg. 4-5, section IV: “In this Section, we present a destructive SWAP test with no ancillas… we can proceed to simplify the SWAP test circuit… To reduce the number of gates, we notice that the ancillary qubit which carries the answer to the test is not affected by the tested qubits after the CCZ gate. We are not interested in the outcomes of the measurements on the qubits under test. We can just as well get rid of the last
H
and CNOT gates… we can just ignore the ancillary qubit and perform a SWAP test with the circuit in Figure 8.” A swap test is implemented on a circuit which is simplified (i.e. approximated) by reducing the number of gates and removing the ancillary qubit.)
Benedetti and Garcia do not appear to explicitly disclose that the quantum circuit performing the entangling operation is parameterized.
However, Heya teaches quantum circuits which are parameterized. (Pg. 1, Abstract: “We propose a gate optimization method, which we call variational quantum gate optimization (VQGO). VQGO is a method to construct a target multi-qubit gate by optimizing a parametrized quantum circuit which consists of tunable single-qubit gates with high fidelities and fixed multi-qubit gates with limited controlabilities.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Benedetti, Garcia, and Heya. Benedetti teaches using a generative adversarial network to learn to approximate a target quantum state, where overlap between the target and generated states can be determined by a swap test. Garcia teaches a simplified quantum circuit for implementing the swap test without requiring an ancillary qubit. Heya teaches optimizing quantum gates using parameterized quantum circuits. One of ordinary skill would have motivation to combine Benedetti, Garcia, and Heya because “In near-term quantum devices, gate fidelities are finally limited by the coherent time. Therefore, for implementing a given unitary operation on a set of qubits, it is strongly demanded to find an efficient gate construction… We show that even in the presence of crosstalk VQGO can achieve higher fidelity of the CNOT gate compared with a conventional method” (Heya, pg. 1, para. 3-4).
Regarding Claim 7, Benedetti and Garcia teach The method of claim 6, as shown above.
Garcia also teaches wherein the ancilla-free swap test approximates an exact swap test and comprises a second quantum circuit, (Pg. 4-5, section IV: “In this Section, we present a destructive SWAP test with no ancillas… we can proceed to simplify the SWAP test circuit… To reduce the number of gates, we notice that the ancillary qubit which carries the answer to the test is not affected by the tested qubits after the CCZ gate. We are not interested in the outcomes of the measurements on the qubits under test. We can just as well get rid of the last
H
and CNOT gates… we can just ignore the ancillary qubit and perform a SWAP test with the circuit in Figure 8.” The ancilla-free swap test is implemented on a quantum circuit (i.e. a second quantum circuit) which is simplified (i.e. approximated) by reducing the number of gates and removing the ancillary qubit.)
Benedetti and Garcia do not appear to explicitly disclose the remaining features of claim 7.
However, Heya teaches wherein the second quantum circuit is a parameterized quantum circuit and second quantum circuit parameters constitute parameters of the discriminator network included in the parameters of the quantum generative adversarial network. (Pg. 1, Abstract: “We propose a gate optimization method, which we call variational quantum gate optimization (VQGO). VQGO is a method to construct a target multi-qubit gate by optimizing a parametrized quantum circuit which consists of tunable single-qubit gates with high fidelities and fixed multi-qubit gates with limited controlabilities.” Pg. 2, para. 2: “We approximate the target gate
U
t
a
r
g
e
t
using the parameterized quantum circuit
U
(
θ
)
by optimizing the parameters
θ
iteratively… In each iteration, the parameters are updated from
θ
l
to
θ
l
+
1
so as to minimize the cost
h
(
θ
)
with gradient-free or gradient-based optimizers…” The quantum circuit (i.e. second quantum circuit) is a parameterized quantum circuit, and its parameters are optimized iteratively using gradients (i.e. the parameters are trainable and thus included in the parameters of the quantum generative adversarial network).)
Claims 16 and 18 are system claims containing substantially the same elements as method claims 5 and 7, respectively. Benedetti, Garcia, and Heya teach the elements of claims 5 and 7, as shown above.
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Benedetti in view of Garcia, and further in view of
Fontana et al. (hereinafter Fontana), “Evaluating the noise resilience of variational quantum algorithms” (published 11/23/2020).
Regarding Claim 8, Benedetti and Garcia teach The method of claim 1, as shown above.
Benedetti and Garcia do not appear to explicitly disclose the remaining features of claim 8.
However, Fontana teaches wherein the quantum generative adversarial network loss function comprises one minus the measured fidelity of the discriminator network input. (Pg. 3, section II.C: “We investigate… target state optimisation, where the infidelity with respect to a random target state is minimized… The optimisation procedure is therefore modified to maximizing the fidelity (see Eq. (6)) with a target state
ρ
T
. Equivalently, the problem can be formulated as a minimization of the infidelity, defined as
R
∶
=
1
-
F
, and hence the cost function is
C
(
θ
)
=
R
(
ρ
T
,
ρ
(
θ
)
)
∶
=
1
-
F
(
ρ
T
,
ρ
(
θ
)
)
.” The cost function (i.e. loss function) comprises one minus the fidelity of the generated and target quantum states.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Benedetti, Garcia, and Fontana. Benedetti teaches using a generative adversarial network to learn to approximate a target quantum state, where overlap between the target and generated states can be determined by a swap test. Garcia teaches a simplified quantum circuit for implementing the swap test without requiring an ancillary qubit. Fontana teaches learning to prepare a quantum state by minimizing its infidelity to a target quantum state, measured by one minus the fidelity. One of ordinary skill would have motivation to combine Benedetti, Garcia, and Fontana because Benedetti measures the overlap/fidelity between generated and target state, and Fontana’s one-minus-fidelity cost function presents a natural conversion of the overlap measure to a minimizable error term.
Claims 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over Benedetti in view of Garcia, and further in view of
Stein et al. (hereinafter Stein), “QuGAN: A Generative Adversarial Network Through Quantum States” (published 10/18/2020).
Regarding Claim 9, Benedetti and Garcia teach The method of claim 1, as shown above.
Benedetti and Garcia do not appear to explicitly disclose the remaining features of claim 9.
However, Stein teaches wherein performing the minimax optimization of the quantum generative adversarial network loss function comprises:
fixing generator network parameters to values determined at a previous iteration and maximizing the quantum generative adversarial network loss function with respect to discriminator network parameters to determine updated values of the discriminator network parameters for the iteration; and (Pg. 2, para. 2: “The discriminator’s output is a probability, and its goal is to maximize
D
(
x
)
and to minimize
D
(
G
(
p
z
,
θ
d
)
. The generator’s goal is to maximize
D
(
G
(
p
z
,
θ
d
)
. If we label the real data as 1 and fake data as 0, this minmax cost function is described in Equation 1.” Pg. 4, para. 4: “Optimization of a GAN is characterized by the loss function described in Equation 1… Using this loss function, we can update the defining parameters of the networks similarly to a classical GAN. This is characterized by measuring the loss of the Discriminator with respect to both the Generator and the fake data, then updating its state to improve its performance. Following this the generator analyzes how well it performs in the Discriminator and updates its state to improve performance… we train our Discriminator more than our generator and do not train them at the same time.” Updated parameters of the discriminator network are determined by maximizing the loss. The discriminator and generator are not trained at the same time (i.e. the generator parameters are fixed during discriminator training). This sequence can also be seen in algorithm 1 (pg. 5).)
fixing the discriminator network parameters to the updated values of the discriminator network parameters for the iteration and minimizing the quantum generative adversarial network loss function with respect to generator network parameters to determine updated values of the generator network parameters for the iteration. (See the portions of pg. 2 and 4, as well as algorithm 1 on pg. 5. Updated parameters of the generator network are determined by minimizing the loss. The discriminator and generator are not trained at the same time (i.e. the discriminator parameters are fixed during generator training).)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Benedetti, Garcia, and Stein. Benedetti teaches using a generative adversarial network to learn to approximate a target quantum state, where overlap between the target and generated states can be determined by a swap test. Garcia teaches a simplified quantum circuit for implementing the swap test without requiring an ancillary qubit. Stein teaches a quantum generative adversarial network with swap test-based state comparison, where the generator and discriminator parameters are fixed and trained in an alternating pattern. One of ordinary skill would have motivation to combine Benedetti, Garcia, and Stein because the alternating fix-and-optimize algorithm is a standard method for GAN training, and Stein explicitly teaches implementing this algorithm for a quantum GAN with swap test-based state comparison: “we can update the defining parameters of the networks similarly to a classical GAN” (Stein, pg. 4, para. 4).
Regarding Claim 10, Benedetti and Garcia teach The method of claim 1, as shown above.
wherein performing the minimax optimization of the quantum generative adversarial network loss function comprises:
Benedetti also teaches fixing the discriminator network parameters to values corresponding to a perfect swap test and minimizing the quantum generative adversarial network loss function with respect to generator network parameters to determine initial updated values of the generator network parameters for the iteration; (Pg. 8, section II.D: “Since the prior probabilities are hyperparameters, we can set them to ½ and use the swap test [33] to compute the overlap. This procedure effectively implements an optimal discriminator and provides a strong learning signal to the generator.” Pg. 12, section IV: “In turn, the generator circuit minimizes the value function in order to deceive the discriminator.” Updated parameters of the generator are determined by fixing the discriminator parameters to perform an optimal swap test and minimizing the value (i.e. loss) function.)
Benedetti and Garcia do not appear to explicitly disclose the remaining features of claim 10.
However, Stein teaches fixing generator network parameters to the initial updated values and maximizing the quantum generative adversarial network loss function with respect to discriminator network parameters to determine updated values of the discriminator network parameters for the iteration; and (Pg. 2, para. 2: “The discriminator’s output is a probability, and its goal is to maximize
D
(
x
)
and to minimize
D
(
G
(
p
z
,
θ
d
)
. The generator’s goal is to maximize
D
(
G
(
p
z
,
θ
d
)
. If we label the real data as 1 and fake data as 0, this minmax cost function is described in Equation 1.” Pg. 4, para. 4: “Optimization of a GAN is characterized by the loss function described in Equation 1… Using this loss function, we can update the defining parameters of the networks similarly to a classical GAN. This is characterized by measuring the loss of the Discriminator with respect to both the Generator and the fake data, then updating its state to improve its performance. Following this the generator analyzes how well it performs in the Discriminator and updates its state to improve performance… we train our Discriminator more than our generator and do not train them at the same time.” Updated parameters of the discriminator network are determined by maximizing the loss. The discriminator and generator are not trained at the same time (i.e. the generator parameters are fixed during discriminator training). This sequence can also be seen in algorithm 1 (pg. 5).)
fixing the discriminator network parameters to the updated values of the discriminator network parameters for the iteration and minimizing the quantum generative adversarial network loss function with respect to generator network parameters to determine updated values of the generator network parameters for the iteration. (See the portions of pg. 2 and 4, as well as algorithm 1 on pg. 5. Updated parameters of the generator network are determined by minimizing the loss. The discriminator and generator are not trained at the same time (i.e. the discriminator parameters are fixed during generator training).)
Claims 11-12 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Benedetti in view of Garcia, and further in view of
Kerenidis, U.S. Patent Application Publication US-20210319351-A1 (filed 08/06/2020).
Regarding Claim 11, Benedetti and Garcia teach The method of claim 1, as shown above.
Benedetti also teaches wherein the method further comprises generating, by the generator network and according to trained generator network parameters, the target quantum state (Pg. 2, section I: “In this manuscript, we start from information theoretic arguments and derive an adversarial algorithm that learns to generate approximations to a target pure quantum state.” The generator, using its learned parameters, generates an approximation of the target state.)
Benedetti and Garcia do not appear to explicitly disclose the remaining features of claim 11.
However, Kerenidis teaches wherein the target quantum state comprises a superposition state and (0098: “Because quantum processors 601 operate on qubits, the ability of qubits to exist in superpositions of 0 and 1 allows for greatly enhanced performance for certain computational tasks.”)
Generating a quantum state to approximate a quantum random access memory. (0003: “Many quantum machine learning and optimization algorithms load classical data into quantum states in order to use quantum procedures for tasks like classification, clustering, or solving linear systems. This makes these algorithms not near-term, since the proposals for such loaders, also called Quantum Random Access Memory (QRAM), are large and complex circuits both in the number of qubits and the depth of the quantum circuit…” 0016: “From the classical data point
(
x
1
,
x
2
,
.
.
.
,
x
n
)
, we will describe a specific implementation of a circuit that can efficiently create the quantum state that encodes this classical data…” A quantum state is generated to encode classical data (i.e. approximate a quantum random access memory).)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Benedetti, Garcia, and Kerenidis. Benedetti teaches using a generative adversarial network to learn to approximate a target quantum state, where overlap between the target and generated states can be determined by a swap test. Garcia teaches a simplified quantum circuit for implementing the swap test without requiring an ancillary qubit. Kerenidis teaches generating quantum states approximating QRAM to encode classical data for downstream use. One of ordinary skill would have motivation to combine Benedetti, Garcia, and Kerenidis because using quantum states to load and encode classical data “reduces the computational resources of the circuit, e.g., number of qubits, depth of quantum circuit, and type of gates in the circuit” (Kerenidis, 0002), and this benefits downstream tasks: “We show how to use the quantum data loader described in Part 1 in order to perform a number of fundamental procedures that are useful among others in machine learning and optimization, including applications in distance estimation, inner product estimation, linear algebra, classification, clustering, neural networks, and many more” (Kerenidis, 0042).
Regarding Claim 12, Benedetti, Garcia, and Kerenidis teach The method of claim 11, as shown above.
Kerenidis also teaches further comprising training a quantum neural network using the generated target quantum state. (0080-0081: “The inner product estimation method we presented above can be used to provide applications in neural networks. We describe here one of the many possible embodiments of this application. In one aspect, one can use a hybrid classical-quantum algorithm based on the well-known feed-forward and back-propagation algorithm. There, the quantum inner product estimation method described above with respect to FIGS. 2 and 3 can be used both for multiplying the matrices of data points and weights during the evaluation, and during the backpropagation algorithm (which may be gradient descent algorithm) where again matrix-matrix multiplication is used.” The data loader quantum state (i.e. generated target quantum state) can be used for quantum neural network backpropagation (i.e. training).)
Claim 20 is a system claim containing substantially the same elements as method claim 11. Benedetti, Garcia, and Kerenidis teach the elements of claim 11, as shown above.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BENJAMIN M ROHD whose telephone number is (571)272-6445. The examiner can normally be reached Mon-Thurs 8:00-6:00 EST.
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, Viker Lamardo can be reached at (571) 270-5871. 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.
/B.M.R./Examiner, Art Unit 2147
/VIKER A LAMARDO/Supervisory Patent Examiner, Art Unit 2147
Prosecution Insights