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 .
Response to Arguments
Remarks page 6-7, Applicant contends:
The cited combination at least fails to teach the amended claims.
Cheng’s sequential preparation is different than the recited sequential preparation.
Response:
Applicant’s arguments with respect to claim(s) 1 and 11 have been considered but are moot because the new ground of rejection contain elements that have not been previously examined or does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. The currently recited rejections will note aspects related the amended material.
Remarks page 8, Applicant contends:
Claims 6 and 16 are amended to overcome Iten and Vanchurin.
Response:
Applicant’s arguments with respect to claim(s) 6 and 16 have been considered but are moot because the new ground of rejection contain elements that have not been previously examined or does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. The currently recited responses to the claims will note aspects related the amended material.
Remarks page 9, Applicant contends:
New claims 21 and 22 further define the qubit efficiency and are not taught by the cited references.
Response:
Applicant’s arguments with respect to claim(s) 21 and 22 have been considered but are moot because the new ground of rejection contain elements that have not been previously examined or does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. The currently recited responses to the claims will note aspects related the amended material.
Claim Rejections - 35 USC § 112
Regarding 112(b):
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 6 and 16 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.
In regard to Claim 6:
Claim 6 recites “permuting ancilla basis states”. The recited element lacks antecedent basis, as one of ordinary skill in the art cannot determine whether the “ancilla basis states” refers to ancilla noted in claim 1 (such as “physical qubit with ancilla”) or are different ancilla with basis states. Amending the claim to clarify can resolve the issue.
In regard to claim 16:
Claim 16 is rejected for the same reason as claim 6.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1, 3-4, 8, 11, 13-14, 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Coyle et al (“The Born supremacy: quantum advantage and training of an Ising Born machine”), referred to as Coyle in this document, in further view of Cheng (“Tree Tensor Networks for Generative Modeling”), referred to as Cheng in this document, and in even further view of Tangpanitanon et al (“Expressibility and trainability of parameterized analog quantum systems for machine learning applications”), referred to as Tangpanitanon in this document, and in even further view of Huggins et al ("Towards quantum machine learning with tensor networks"), referred to as Huggins in this document.
Regarding Claim 1:
Coyle teaches:
encoding, by processing circuitry, classical data vectors into quantum data vectors by applying the classical data vectors to an encoding map to… the quantum data vectors being defined with respect to a plurality of quantum states
[Coyle Results page 3]: “In particular, we will adopt the following kernel37 in which the samples are encoded [encoding, by processing circuitry, classical data into a plurality of quantum states by applying the classical data to an encoding map to… the quantum data vectors being defined with respect to a plurality of quantum states] in a quantum state, |ϕ(x)〉, via a feature map, ϕ:x → |ϕ(x).”
Further evidence to help support the samples are classical data [Coyle Discussion page 9]: “However, this assumes that we have access only to classical samples from the distribution, and the possibility of gaining an advantage using quantum samples is unexplored in the context of distribution learning.”
Evidence of processing circuitry is taught by Coyle ([Coyle Figure 4 page 6]: “The probabilities given by the other bars are those achieved after training the model with either the MMD or SHD on the simulator or the physical Rigetti chip, on an average run.”) as supported by current application ([Current Application 0086]: “In other words, the processing circuitry 1310 may include one or more physical packages (e.g., chips) including materials, components or wires on a structural assembly (e.g., a baseboard).”).
training a quantum model comprising a tensor network structure based on the plurality of quantum states and the quantum data vectors
[Coyle Introduction page 1]: “Quantum circuit Born machines (QCBM) are a subclass of parameterized quantum circuits (PQCs) and are widely applicable (see ref. 19 for a review). PQCs consist of a quantum circuit which carries parameters that are updated during a training [training a quantum model comprising a tensor network structure based on the plurality of quantum states and the quantum data vectors] process (typically a classical optimization routine). The circuit is kept as shallow as possible so as to be suitable for NISQ devices.”
compiling, by the processing circuitry, the quantum model comprising the tensor network structure, into a quantum circuit by mapping virtual qubits of the quantum model onto hardware qubits of a quantum hardware device, the quantum circuit comprising a sequence of operations tailored for operation on the quantum hardware device
[Coyle Quantum Compiling page 8]: “The major objective in this area is to compile [compiling, by the processing circuitry, the quantum model comprising the tensor network structure, into a quantum circuit by mapping virtual qubits of the quantum model onto hardware qubits of a quantum hardware device,] a given target unitary, U, into one that consists exclusively of operations available to the native hardware [the quantum circuit comprising a sequence of operations tailored for operation on the quantum hardware device] of the quantum computer in question.”
Coyle does not explicitly teach:
comprising a tensor network structure
comprising the tensor network structure
encode the classical data vectors as quantum data vectors in an unentangled product state in a quantum Hilbert space
the tensor network structure comprising a tensor network topology that captures matrix product states (MPSs);
performing a sequential preparation on each MPS of the tensor network structure to generate an MPS wavefunction for each respective MPS, wherein the sequential preparation comprises coupling a physical qubit with an ancilla using isometric operators derived from the MPS, measuring the physical qubit and returning the physical qubit to a fiducial state while leaving the ancilla unmeasured
wherein the mapping virtual qubits comprises reusing hardware qubits based on the MPS wavefunction
Cheng teaches:
Aspects of Cheng are left for support of aspects of tensor structure, but the additional limitations provided by the amendments results in numerous aspects taught here by Cheng being possibly more sensible of a combination using the teachings from Huggins later in this rejection (as Huggins teaches tensor network structures with MPS as well).
comprising a tensor network structure
comprising the tensor network structure
[Cheng Introduction page 1]: “the other one is the Born machine, where Born’s rule in quantum physics is borrowed to represent the joint probability distribution of data with the square amplitude of a wave function [13–16] and the wave function is represented by tensor networks [comprising a tensor network structure][comprising the tensor network structure].”
the tensor network structure comprising a tensor network topology that captures matrix product states (MPSs);
to generate an MPS wavefunction for reach respective MPS
[Cheng Introduction page 1]: “Tensor networks (TNs) were originally designed for efficiently representing quantum many-body wave functions [to generate an MPS wavefunction for reach respective MPS] [17,18], which, in general, are described by a high-order tensor with exponential parameters. A TN applies low-rank decompositions to the general tensor by discarding the vast majority of unrelated long-range information to break the so-called exponential wall of quantum many-body computation. Popular TNs include matrix product states (MPSs) [19] [the tensor network structure comprising a tensor network topology that captures matrix product states (MPSs)], tree tensor networks (TTNs) [20], the multiscale entanglement renormalization ansatz [21], projected entanglement pair states (PEPSs) [22], etc.”
performing a sequential preparation on each matrix product state of the tensor network structure to generate an MPS wavefunction for reach respective MPS
[Cheng C. Tree tensor network generative model page 4]: “It is technically easy to canonicalize a tensor in the TTN. For example, we can start from one end of the tree and use the QR decomposition of the tensor to push the noncanonical part of the tensor to the adjacent tensor. By repeating this step [performing a sequential preparation on each matrix product state of the tensor network structure to generate an MPS wavefunction for reach respective MPS], finally, one will push all noncanonical parts of the TTN to just one tensor, called the central tensor, and all other tensors are in one of the three canonical forms. Analogous to the mixed-canonical form of MPSs, we call this form the mixed-canonical form of the TTN.”
One of ordinary skill in the art, prior to the effective filing date, would have been motivated to combine Coyle and Cheng to incorporate tensor networks. Coyle and Cheng are of the same field of endeavor of quantum computing. One of ordinary skill in the art would have been motivated to combine Coyle and Cheng in order to incorporate tensor networks to take advantage of the tensor networks better modeling of some data, such as 2D data ([Cheng Introduction page 2]: “On the other hand, a TTN works as a new tensor network generative model, an extension of the recently proposed MPS Born machine [14]. Compared to MPSs, TTNs exhibit natural modeling on two-dimensional data such as natural images, and its more favorable in the growth of the correlation length of pixels.” where TTN stands for tree tensor network). Chen further notes that TNs (which MPS are noted to be a TN in the quote) are useful to deal with the exponential wall related to quantum many-body computations and to discard unrelated information ([Cheng Introduction page 1]: “Tensor networks (TNs) were originally designed for efficiently representing quantum many-body wave functions [17,18], which, in general, are described by a high-order tensor with exponential parameters. A TN applies low-rank decompositions to the general tensor by discarding the vast majority of unrelated long-range information to break the so-called exponential wall of quantum many-body computation. Popular TNs include matrix product states (MPSs) [19], tree tensor networks (TTNs) [20], the multiscale entanglement renormalization ansatz [21], projected entanglement pair states (PEPSs) [22], etc.”)
Tangpanitanon teaches:
encode the classical data vectors as quantum data vectors in an unentangled product state in a quantum Hilbert space
[Tangpanitanon Introduction page 1]: “A hint to answer this question lies in the ability of NISQ devices to efficiently explore Hilbert space [encode the classical data vectors as quantum data vectors in an unentangled product state in a quantum Hilbert space]. For example, in quantum chemistry, NISQ devices can produce highly-entangled variational ansatzes, such as unitary coupled clusters, that cannot be efficiently represented on a classical computer [18].”
[Tangpanitanon II. Driven Analog Quantum Systems and Their Statistics page 3]: “The parameters {Θi, m} are `varied' by randomly drawing them from a uniform distribution in the range [0;W] where W is the disorder strength. This allows us to vary the parameters without changing the phase of the system. The dimension of the Hilbert space is N = 2L. The initial state |ψ_0> is prepared as a product state [in an unentangled product state] where each spin points along the +z direction. This simple model has been implemented in various quantum platforms, including Rydberg atoms [51], trapped ions [52] and superconducting circuits [16].”
One of ordinary skill in the art, prior to the effective filing date, would have been motivated to combine Modified Coyle and Tangpanitanon to incorporate encoding into Hilbert space. Coyle and Tangpanitanon are of the same field of endeavor of quantum computing. One of ordinary skill in the art would have been motivated to combine Coyle and Tangpanitanon in order to incorporate aspects of NISQ devices that cannot be efficiently represented on a classical computer ([Tangpanitanon Introduction page 1]: “A hint to answer this question lies in the ability of NISQ devices to efficiently explore Hilbert space. For example, in quantum chemistry, NISQ devices can produce highly-entangled variational ansatzes, such as unitary coupled clusters, that cannot be efficiently represented on a classical computer [18].”).
Huggins teaches:
training a quantum model comprising a tensor network structure based on the plurality of quantum states and the quantum data vectors, the tensor network structure comprising a tensor network topology that captures matrix product states (MPSs)
[Huggins 2 Learning with tensor network quantum circuits page 4]: “In the context of machine learning, we use these states to represent the trainable ‘weight’ parameters [training a quantum model comprising a tensor network structure based on the plurality of quantum states and the quantum data vectors] of a particular family of models, defined by a feature mapΦ, with an example given below. For the types of feature maps we will choose, working in the space of all possible weights (full Hilbert space) corresponds to extremely powerful and expressive models which would likely overfit most data sets [11–13].”
[Continuing the quote from above in Huggins]: “Using a finite-bond-dimension tensor network to represent the weights instead limits this expressivity, but the limitation can be rapidly lifted by increasing the bond dimension. Currently, the limit for practical bond dimensions is in the tens of thousands for MPS [the tensor network structure comprising a tensor network topology that captures matrix product states (MPSs)] tensor networks, and many hundreds or perhaps thousands for more general tree tensor networks. By implementing tensor network circuits on quantum hardware instead, one could in principle go far beyond these classical limitations on model expressivity.”
performing a sequential preparation on each MPS of the tensor network structure to generate an MPS wavefunction for each respective MPS, wherein the sequential preparation comprises coupling a physical qubit with an ancilla using isometric operators derived from the MPS, measuring the physical qubit and returning the physical qubit to a fiducial state while leaving the ancilla unmeasured
[Huggins Figure 5 page 6]: “Generative tree tensor network model architecture, showing a case with V=2 qubits connecting each subtree. To sample from the model, qubits are prepared in a reference computational basis state <0| (left-hand side of circuit). Then 2V qubits are entangled via unitary operations [performing a sequential preparation on each MPS of the tensor network structure to generate an MPS wavefunction for each respective MPS, wherein the sequential preparation comprises coupling a physical qubit with an ancilla using isometric operators derived from the MPS] at each layer of the tree as shown. The qubits are measured at the points in the circuit labeled by square symbols (right-hand side of circuit), and the results of these measurements provides the output of the model. While all qubits could be entangled before being measured, we discuss in section 4 the possibility performing opportunistic measurements to reduce the physical qubit overhead.”
[Huggins 4.2 Qubit-efficient matrix product models page 12]: “To implement the generative MPS [support for MPS relation] algorithm shown in figure 11(b), one prepares all qubits to a reference state |0>V+1 and after entangling [which as noted above in figure 5 description involves unitary operators which are considered isometric using the quote from Iten noted below] the qubits, one measures [measuring the physical qubit] and records a single qubit to generate the first output value. This qubit is reset to the state ∣0> [and returning the physical qubit to a fiducial state while leaving the ancilla unmeasured] and all the qubits are then acted on by another (V+1) qubit unitary. A single qubit is again measured to generate the second output value, and the algorithm continues until N outputs have been generated.”
Aspects of being sequential are supported by the steps noted to go a qubit at a time in the above quotes. MPS relation is shown by Huggins indicating the aspects are implementing MPS algorithm, thus the elements are from/for MPS. Support for ancilla qubits being present is provided in [Huggins 2.1 Discriminative algorithm page 4]: “At each step of implementing the model on a quantum computer, one takes V of the qubits resulting from one of the unitary operations of the previous step, or subtree, and V from another subtree and acts on them with another parameterized unitary transformation (possibly together with some ancilla qubits—not shown).”. Aspects of leaving the ancilla unmeasured is indicated by the system only measuring the one qubit, thus the other qubits, such as the ancilla, would be unmeasured.
The interpretation of the unitary operations being considered isometric operations is supported by the description of isometries from Iten, which indicates the use of ancilla qubits in a fixed state like described in Huggins being acted on by a unitary ([Iten Introduction page 1]: "Physically, isometries can be thought of as the introduction of ancilla qubits in a fixed state (conventionally |0>) followed by a general unitary on the system and ancilla qubits. However, because its action only has to be specified when the ancilla systems start in state |0>, there is a great deal of freedom when constructing the general unitary.").
wherein the mapping virtual qubits comprises reusing hardware qubits based on the MPS wavefunction
[Huggins 4 Implementation on near-term devices page 9-10]: “A key advantage of carrying out machine learning tasks with models equivalent to tree or matrix product tensor networks is that they could be implemented using a very small number of physical qubits… Below we will first discuss how the number of qubits needed to implement either a discriminative or generative tree tensor network model can be made to scale only logarithmically in both the data dimension and in the bond dimension of the network. Then we will discuss the special case of MPS tensor networks, which can be implemented with a number of physical qubits that is independent [wherein the mapping virtual qubits comprises reusing hardware qubits based on the MPS wavefunction] of the input or output data dimension. Another key advantage of using tensor network models on near-term devices could be their robustness to noise, which will certainly be present in any near-term hardware. To explore the noise resilience of our models, we present a numerical experiment where we evaluate the model trained in section 3 with random errors, and observe whether it can still produce useful results.”
Further support for the idea of reusing qubits is taught [Huggins 4.1 Qubit efficient tree network models page 10]: “Then V of the qubits can be measured and reused, but the other V qubits must remain entangled. So only V new qubits must be introduced to process more inputs.”
One of ordinary skill in the art, prior to the effective filing date would have been motivated to combine Coyle and Huggins. Coyle and Huggins are in the same field of endeavor of quantum computing. One of ordinary skill in the art would have been motivated to combine Coyle and Huggins in order to take advantage of reusing qubits in order to reduce the required qubits required for algorithms ([Huggins 4 Implementation on near-term devices page 9-10]: “A key advantage of carrying out machine learning tasks with models equivalent to tree or matrix product tensor networks is that they could be implemented using a very small number of physical qubits… Below we will first discuss how the number of qubits needed to implement either a discriminative or generative tree tensor network model can be made to scale only logarithmically in both the data dimension and in the bond dimension of the network.”).
Huggins also notes the motivation for the tensor network aspects as such implementations for circuits are noted to provide efficient qubit schemes ([Huggins Abstract]: “Tensor network circuits can also provide qubit-efficient schemes in which, depending on the architecture, the number of physical qubits required scales only logarithmically with, or independently of the input or output data sizes”).
Regarding Claim 3:
The method of claim 1 is taught by Coyle, Cheng, Tangpanitanon, and Huggins.
Tangpanitanon teaches:
wherein the classical data vectors are encoded into the quantum Hilbert space, the quantum Hilbert space being orthonormal
[Tangpanitanon Introduction page 1]: “A hint to answer this question lies in the ability of NISQ devices to efficiently explore Hilbert space [wherein the classical data vectors are encoded into the quantum Hilbert space]. For example, in quantum chemistry, NISQ devices can produce highly-entangled variational ansatzes, such as unitary coupled clusters, that cannot be efficiently represented on a classical computer [18].”
[Tangpanitanon Section 2 Driven Analog Quantum Systems and Their Statistics page 2]: “For the thermalized dynamics, the statistics of Hˆ ave(θm) follows the Gaussian orthogonal ensemble (GOE) [47]. This is the ensemble of matrices whose entries are independent normal random variables subjected to the orthogonality [the quantum Hilbert space being orthonormal] constraint.”
The motivation to combine with Tangpanitanon is the same as for claim 2.
Regarding Claim 4:
The method of claim 1 is taught by Coyle, Cheng, Tangpanitanon, and Huggins.
Cheng teaches:
wherein the training the quantum model comprises encoding the quantum data vectors into a wavefunction that is structured as a Born machine
[Cheng Introduction page 1]: “the other one is the Born machine [wherein the training the quantum model comprises encoding the quantum data vectors into a wavefunction that is structured as a Born machine], where Born’s rule in quantum physics is borrowed to represent the joint probability distribution of data with the square amplitude of a wave function [13–16] and the wave function is represented by tensor networks.”
One of ordinary skill in the art, prior to the effective filing date, would have been motivated to combine Coyle and Cheng to incorporate the use of a Born Machine. Coyle and Cheng are of the same field of endeavor of quantum computing. One of ordinary skill in the art would have been motivated to combine Coyle and Cheng in order to incorporate the expressiveness of the born machine compared to classical probability functions ([Cheng II. Models and Algorithms page 3]: “The reason we choose the quantum-inspired Born machine instead of directly modeling a joint probability is based on a belief that the Born machine representation is more expressive than classical probability functions”).
Regarding Claim 8:
The method of claim 1 is taught by Coyle, Cheng, Tangpanitanon, and Huggins.
Coyle teaches:
wherein the quantum hardware device comprises a noisy intermediate-scale quantum (NISQ) computing device
[Coyle Introduction page 1]: “In spite of this, NISQ devices [wherein the quantum hardware device comprises a noisy intermediate-scale quantum (NISQ) computing device] could provide efficient solutions to other problems that cannot be solved in polynomial time by classical means. Showing this to be true is referred to as a demonstration of quantum computational supremacy with the first such experimental realization occurring recently.”
Regarding Claim 11:
This claim is analogous to claim 1.
Regarding Claim 13:
The method of claim 11 is taught by Coyle, Cheng, Tangpanitanon, and Huggins.
This claim is analogous to claim 3.
Regarding Claim 14:
The method of claim 11 is taught by Coyle, Cheng, Tangpanitanon, and Huggins.
This claim is analogous to claim 4.
Regarding Claim 18:
The method of claim 11 is taught by Coyle, Cheng, Tangpanitanon, and Huggins.
This claim is analogous to claim 8.
Claims 7 and 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Coyle et al (“The Born supremacy: quantum advantage and training of an Ising Born machine”), referred to as Coyle in this document, in further view of Cheng (“Tree Tensor Networks for Generative Modeling”), referred to as Cheng in this document, and in even further view of Tangpanitanon et al (“Expressibility and trainability of parameterized analog quantum systems for machine learning applications”), referred to as Tangpanitanon in this document, and in even further view of Huggins et al ("Towards quantum machine learning with tensor networks"), referred to as Huggins in this document, and in even further view of Iten et al (“Quantum circuits for isometries”), referred to as Iten in this document.
Regarding Claim 7:
The method of claim 1 is taught by Coyle, Cheng, Tangpanitanon, and Huggins.
Modified Coyle does not explicitly teach:
wherein compiling the quantum model comprises implementing greedy heuristics for determining gate sequences that match a target isometry and transforming the target isometry into operations of the quantum circuit
Iten teaches:
wherein compiling the quantum model comprises implementing greedy heuristics for determining gate sequences that match a target isometry and transforming the target isometry into operations of the quantum circuit
[Iten IV. DECOMPOSITION SCHEMES FOR ISOMETRIES C. Column-by-column decomposition page 5]: “In essence, the idea is to find a sequence of unitary operations that when applied to 𝑉 successively brings it closer to 𝐼2𝑛×2𝑚. We will do this in a column-by-column fashion, first choosing a sequence of quantum gates [wherein compiling the quantum model comprises implementing greedy heuristics for determining gate sequences that match a target isometry], corresponding to a unitary 𝐺0 that gets the first column right, i.e., 𝐺0𝑉|0⟩𝑚=𝐼2𝑛×2𝑚|0⟩𝑚=|0⟩𝑛, then using 𝐺1 to get the second column right without affecting the first, i.e., 𝐺1𝐺0𝑉|1⟩𝑚=𝐼2𝑛×2𝑚|1⟩𝑚=|1⟩𝑛 and 𝐺1𝐺0𝑉|0⟩𝑚=𝐺1|0⟩𝑛=|0⟩𝑛, and so on (up to the 2𝑚th column). In other words, 𝐺𝑘 gets the (𝑘+1)th column right and acts trivially on the first 𝑘 columns of 𝐼2𝑛×2𝑚.”
Quote for improved context. [Iten IV. DECOMPOSITION SCHEMES FOR ISOMETRIES C. Column-by-column decomposition page 5]: “In this section we introduce a circuit topology corresponding to a column-by-column decomposition of an arbitrary isometry, i.e., we decompose any isometry into single-qubit and CNOT gates proceeding one column at a time.”
[Iten III Lower Bound page 3]: “We can think of this isometry in terms of a unitary operation [and transforming the target isometry into operations of the quantum circuit] on 𝑛 qubits, 𝑛−𝑚 of which always start in a fixed state, which we take to be |0⟩.2 Without any cnot gates, all we can do is apply single-qubit unitaries individually to each of these 𝑛 qubits.”
One of ordinary skill in the art, prior to the effective filing date, would have been motivated to combine Modified Coyle and Iten to incorporate some greedy heuristics related to isometry. Coyle and Iten are of the same field of endeavor of quantum computing. One of ordinary skill in the art would have been motivated to combine Coyle and Iten in order to incorporate the efficiency related to the implementation ([Iten IV. DECOMPOSITION SCHEMES FOR ISOMETRIES C. Column-by-column decomposition page 7]: “Remark 3. In some physical realizations it is difficult to implement cnot gates between nonadjacent qubits. The decomposition in this section can be adapted to the gate library containing only nearest-neighbor cnot and single-qubit gates in a relatively efficient way.”).
Regarding Claim 17:
The method of claim 11 is taught by Coyle, Cheng, Tangpanitanon, and Huggins
This claim is analogous to claim 7.
Claims 9, 10, 19, and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Coyle et al (“The Born supremacy: quantum advantage and training of an Ising Born machine”), referred to as Coyle in this document, in further view of Cheng (“Tree Tensor Networks for Generative Modeling”), referred to as Cheng in this document, and in even further view of Tangpanitanon et al (“Expressibility and trainability of parameterized analog quantum systems for machine learning applications”), referred to as Tangpanitanon in this document, and in even further view of Huggins et al ("Towards quantum machine learning with tensor networks"), referred to as Huggins in this document, and in further view of Soeken et al (“Compiling Permutations for Superconducting QPUs”), referred to as Soeken in this document.
Regarding Claim 9:
The method of claim 1 is taught by Coyle, Cheng, Tangpanitanon, and Huggins
Modified Coyle does not explicitly teach:
wherein the quantum hardware device comprises a plurality of qubits in a qubit topology comprising single-qubit rotations and entangling gates between pairs of qubits
Soeken teaches:
wherein the quantum hardware device comprises a plurality of qubits in a qubit topology comprising single-qubit rotations and entangling gates between pairs of qubits
[Soeken B. Qubits and quantum gates page 2]: “Since single qubit states correspond to points on the Bloch sphere [12], quantum operations on a single qubit [wherein the quantum hardware device comprises a plurality of qubits in a qubit topology comprising single-qubit rotations] correspond to rotations.”
[Soeken B. Qubits and quantum gates page 2]: “A quantum computer consists of an array of qubits, which in contrast to classical bits, can be in a superposition state and can be entangled [12]… A quantum state can be transformed into another quantum state by applying quantum gates, which are represented by 2×2 unitary matrices… For example, there are no two independent qubit states
PNG
media_image1.png
27
33
media_image1.png
Greyscale
and
PNG
media_image2.png
30
39
media_image2.png
Greyscale
such that
PNG
media_image3.png
107
200
media_image3.png
Greyscale
, the state that is in the perfect superposition between the classical states 00 and 11. This phenomena is called entanglement [and entangling gates between pairs of qubits].”
One of ordinary skill in the art, prior to the effective filing date, would have been motivated to combine modified Coyle and Soeken to incorporate single-qubit rotations and entangling gates in the manner Soeken mentions. Coyle and Soeken are in the same field of endeavor of quantum computing. One of ordinary skill in the art would have been motivated to combine Coyle and Soeken in order to be able to reduce the gate count and depth compared to other art methods ([Soeken Introduction page 1]“In an experimental evaluation, we show that our proposed approach leads to quantum circuits with lower quantum gates and lower depth compared to state-of-the-art generic compilation techniques. For Rigetti QPUs we can reduce gate count and gate depth up to 59% and 53%, respectively, and for IBM QPUs we can reduce gate count and gate depth up to 56% and 53%, respectively.”).
Regarding Claim 10:
The method of claim 1 is taught by Coyle, Cheng, Tangpanitanon, and Huggins
Modified Coyle does not explicitly teach:
wherein the quantum circuit comprises a plurality of gates; and wherein compiling the quantum model comprises minimizing a number of entangled gates within the plurality of gates
Soeken teaches:
wherein the quantum circuit comprises a plurality of gates; and wherein compiling the quantum model comprises minimizing a number of entangled gates within the plurality of gates
[Soeken Introduction page 1]“In an experimental evaluation, we show that our proposed approach leads to quantum circuits with lower quantum gates [wherein the quantum circuit comprises a plurality of gates] and lower depth compared to state-of-the-art generic compilation techniques [and wherein compiling the quantum model comprises minimizing a number of entangled gates within the plurality of gates]. For Rigetti QPUs we can reduce gate count and gate depth up to 59% and 53%, respectively, and for IBM QPUs we can reduce gate count and gate depth up to 56% and 53%, respectively.”
By reducing the gate count as a whole, that would include reducing the number of entangled gates.
Motivation is the same as claim 9.
Regarding Claim 19:
The method of claim 11 is taught by Coyle, Cheng, Tangpanitanon, and Huggins
This claim is analogous to claim 9.
Regarding Claim 20:
The method of claim 11 is taught by Coyle, Cheng, Tangpanitanon, and Huggins
This claim is analogous to claim 10.
Allowable Subject Matter
Claims 6, 16 are rejected under 112(b), but would be allowable if amended to correct the 112(b) rejection and rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Claims 21, 22 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
The following is an examiner's statement of reasons for allowable subject matter:
The prior art of record teaches limitations as noted by the previous office action.
However, the claims 6 and 21 in the application are deemed to be directed to a nonobvious improvement over the prior art of record.
In regards to claim 6:
Wherein implementing the diagonal gauge comprises permuting ancilla basis states derived from a polar decomposition to ensure that qubits preferentially remain in their same state
Prior art of record does not appear to teach the diagonal gauge as described in claim 6. Aspects related to diagonal gauge are noted in the previous office action in the rejection of claim 6.
The closest prior art of record:
Daskin et al ("An ancilla-based quantum simulation framework for non-unitary matrices") notes aspects of polar decomposition in quantum circuits ([Daskin 2.1.2 Estimation of U]: “We shall describe the closeness of U to a unitary matrix via polar decomposition which is known to produce the closest unitary matrix to a given arbitrary matrix (e.g., see [23]): In the polar decomposition, , while is the closest unitary to U, is an Hermitian positive definite matrix.”) and notes support for some relation to ancilla qubits ([Daskin Abstract]: “The success probability in an ancilla-based circuit generally decreases exponentially in the number of qubits consisted in the ancilla”).
However the prior art of record does not provide indication of using ancilla basis states from polar decomposition to implement diagonal gauge. Daskin et al and previous notes of elements related to diagonal gauge give teachings of related aspects, but do not teach the limitations within the current claim 6.
In regards to claim 21:
Wherein the sequential preparation requires a single physical qubit independent of a length of the classical data vectors, and wherein the ancilla comprises log_2(X) qubits for a bond dimension X of the MPS
The previously recited art discussed aspects related to sequential preparation, such as by Cheng, and the currently recited rejection for claim 1 notes teachings for sequential preparation, but the prior art of record does not appear to teach claim 21.
The closest prior art of record teaches:
[Huggins 4 Implementation on near-term devices page 11]: ”Below we will first discuss how the number of qubits needed to implement either a discriminative or generative tree tensor network model can be made to scale only logarithmically in both the data dimension and in the bond dimension of the network [related to and wherein the ancilla comprises log_2(X) qubits for a bond dimension X of the MPS]. Then we will discuss the special case of MPS tensor networks, which can be implemented with a number of physical qubits that is independent [related to wherein the sequential preparation requires a single physical qubit independent of a length of the classical data vectors] of the input or output data dimension.”
Huggins notes qubits can be made to scale logarithmically in the data dimension and the bond dimension of the network. This scaling logarithmically appears to indicate aspects related to the “the ancilla comprises log_2(X) qubits for a bond dimension X of the MPS”, but Huggins does not appear to limit the qubits being addressed to only being ancilla, nor does Huggins appear to limit solely to the bond dimension (as Huggins notes both the data dimension and bond dimension). As a result, Huggins appears to note aspects related to the understanding and use of qubit scaling off of the dimensions of aspects related to MPS, but does not appear to note the scaling in the same way as the current invention. Huggins notes the physical qubits are independent of the input or output dimension, which appears to indicate aspects of “wherein the sequential preparation requires a single physical qubit independent of a length of the classical data vectors”, for at least a single physical qubit is mentioned and the independence to size of some dimension is noted (where the input or output can be considered akin to classical data vectors as the input or output of a discriminative or generative model would not be representing a quantum element as would be representing the input or answer for the model).
Huggins is noted as having the motivation of utilizing smaller physical qubits, which is similar to one of the motivations of the current application ([Huggins 4 Implementation on near-term devices page 10]: “A key advantage of carrying out machine learning tasks with models equivalent to tree or matrix product tensor networks is that they could be implemented using a very small number of physical qubits. The key requirement is that the hardware must allow the measurement of individual physical qubits without further disturbing the state of the other qubits, a capability also required for certain approaches to quantum error correction [43].”).
As a result of the prior art of record not teaching all of the elements of the recited claims 6 and 21, those claims are deemed directed to non-obvious subject matter and not rejected under 103.
Claims 16 and 22 are considered analogous to claims 6 and 21 respectively, thus are considered allowable subject matter for the same reasons given for claim 6 and 21. Any claims dependent upon allowable subject matter claims or made dependent upon allowable subject matter claims are considered allowable under 102/103 for being dependent upon an allowable subject matter claim.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Stojevic et al (US 20210081804 A1) is considered relevant art due to the noting of tensor networks, isometrics, Hilber spaces, matrix product states, and some other aspects of quantum computing in the reference.
Carignan-Dugas et al (“A polar decomposition for quantum channels (with applications to bounding error propagation in quantum circuits)”) is considered relevant art as the reference covers aspects noted in the current application’s specification, such as polar decomposition, orthogonal, and diagonalizable.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to CHRISTOPHER D DEVORE whose telephone number is (703)756-1234. The examiner can normally be reached Monday-Friday 7:30 am - 5 pm 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, Michael J Huntley can be reached at (303) 297-4307. 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.
/C.D.D./Examiner, Art Unit 2129
/MICHAEL J HUNTLEY/Supervisory Patent Examiner, Art Unit 2129
Prosecution Insights