Quantum score matching with applications to learning thermal states
The paper sets up quantum score matching as a trainable way to learn Gibbs states without extra thermal-state preparation.
Dong and Leng extend score matching from classical probability models to quantum density operators, where noncommutativity makes the score and training objective harder to define. They claim end-to-end theoretical guarantees and optimal sample complexity in the high-temperature regime for bounded-locality Hamiltonians. Simulations suggest the method tolerates inaccurate gradient estimates under limited measurements. IBM hardware tests, run without error mitigation or correction, reduced relative Hamiltonian-parameter error from 64% to about 10%. ArXiv · AI/CL/LG's note
Dong and Leng extend score matching from classical probability models to quantum density operators, where noncommutativity makes the score and training objective harder to define. They claim end-to-end theoretical guarantees and optimal sample complexity in the high-temperature regime for bounded-locality Hamiltonians. Simulations suggest the method tolerates inaccurate gradient estimates under limited measurements. IBM hardware tests, run without error mitigation or correction, reduced relative Hamiltonian-parameter error from 64% to about 10%. ArXiv · AI/CL/LG's note
score 5