Compressed unsupervised quantum state preparation with quantum autoencoders
A system and method include techniques for: generating, by a quantum autoencoder, based on a set of quantum states encoded in a set of qubits, a decoder circuit that acts on a subset of the set of qubits, a size of the subset being less than a size of the set; and generating a reduced-cost circuit, the reduced-cost circuit comprising: (1) a new parameterized quantum circuit acting only on the subset of the set of qubits, and (2) the decoder circuit.
1. A method for generating a reduced-cost circuit of a target circuit, the method comprising:
generating, by a quantum autoencoder, based on a set of quantum states in a set of qubits output from the target circuit, a decoder circuit that acts on a subset of the set of qubits, a size of the subset being less than a size of the set of qubits;
generating the reduced-cost circuit, the reduced-cost circuit comprising: (1) a new parameterized quantum circuit, the new parameterized quantum circuit having a set of parameters, acting only on the subset of the set of qubits, and (2) the decoder circuit,
wherein generating the reduced-cost circuit comprises:
training an encoder circuit and the decoder circuit to optimize the average fidelity of a plurality of training states, wherein training the encoder circuit comprises training a first and second subset of the set of qubits of the encoder circuit to reduce a latent space of the target circuit by at least one qubit; and
receiving the set of quantum states generated by at least one quantum circuit, having a depth D1 and a first cost function having a first cost value C1; and
training the reduced-cost circuit to act on the set of qubits as a generative model to reproduce the set of quantum states output from the target circuit, the reduced-cost circuit having a second depth D2 and being associated with a corresponding second cost function having a second cost value C2, wherein at least one of the following is true: (1) C2 is less than C1; and (2) D2 is less than D1.
2. The method of claim 1 , wherein C2 is less than C1 and D2 is less than D1.
3. The method of claim 1 , wherein C2 less than C1.
4. The method of claim 1 , wherein D2 is less than D1.
5. The method of claim 1 , wherein the first cost function calculates a first energy cost and wherein the second cost function calculates a second energy cost.
6. The method of claim 1 , wherein the first cost function represents a first function of a first number of gates within a circuit and wherein the second cost function represents a second function of a second number of gates within the circuit.
7. The method of claim 1 , wherein the first cost function represents a first fidelity F1 of a first output state of the decoder circuit compared to a reference state.
8. The method of claim 7 , wherein the second cost function represents a second fidelity F2 of a second output state of the decoder circuit compared to the reference state, wherein F1<F2.
9. A system for generating a reduced-cost circuit of a target circuit, the system comprising:
a quantum autoencoder (i) generating, based on a set of quantum states in a set of qubits output from the target circuit, a decoder circuit that acts on a subset of the set of qubits, a size of the subset being less than a size of the set of qubits; and
a reduced-cost circuit generator generating the reduced-cost circuit, the reduced-cost circuit comprising: (1) a new parameterized quantum circuit, the new parameterized quantum circuit having a set of parameters, acting only on the subset of the set of qubits, and (2) the decoder circuit,
wherein generating the reduced-cost circuit comprises:
training an encoder circuit and the decoder circuit to optimize the average fidelity of a plurality of training states, wherein training the encoder circuit comprises training a first and second subset of the set of qubits of the encoder circuit to reduce a latent space of the target circuit by at least one qubit; and
receiving the set of quantum states generated by at least one quantum circuit, having a depth D1 and a first cost function having a first cost value C1; and
wherein the system further comprises a reduced-cost circuit trainer training the reduced-cost circuit to act on the set of qubits as a generative model to reproduce the set of quantum states output from the target circuit, the reduced-cost circuit having a second depth D2 and being associated with a corresponding second cost function having a second cost value C2, wherein at least one of the following is true: (1) C2 is less than C1; and (2) D2 is less than D1.
10. The system of claim 9 , wherein C2 is less than C1 and D2 is less than D1.
11. The system of claim 9 , wherein C2 less than C1.
12. The system of claim 9 , wherein D2 is less than D1.
13. The system of claim 9 , wherein the first cost function calculates a first energy cost and wherein the second cost function calculates a second energy cost.
14. The system of claim 9 , wherein the first cost function represents a first function of a first number of gates within a circuit and wherein the second cost function represents a second function of a second number of gates within the circuit.
15. The system of claim 9 , wherein the first cost function represents a first fidelity F1 of a first output state of the decoder circuit compared to a reference state.
16. The system of claim 15 , wherein the second cost function represents a second fidelity F2 of a second output state of the decoder circuit compared to the reference state, wherein F1<F2.