Core Concepts
We present an efficient agnostic tomography algorithm that can learn a stabilizer product state approximation to an unknown quantum state, even if the unknown state is far from any stabilizer product state.
Abstract
The key insights and highlights of the content are:
The authors define a quantum learning task called agnostic tomography, where the goal is to output a succinct description of a state that approximates an unknown state ρ at least as well as any state in a given class C.
Agnostic tomography is more challenging than ordinary quantum tomography, as the learning algorithm must be robust to perturbations of ρ and cannot exploit the structure of states in C.
The authors present an efficient agnostic tomography algorithm for the class of n-qubit stabilizer product states. Assuming ρ has fidelity at least τ with a stabilizer product state, the algorithm runs in time nO(1+log(1/τ))/ε^2, which is quasipolynomial in all parameters and polynomial if τ is a constant.
The key ideas behind the algorithm are: (i) leveraging the structure of stabilizer product states to identify their stabilizer group from fewer Bell difference samples compared to general stabilizer states, and (ii) an entropy counting argument to show that a small number of samples suffices to learn the stabilizer group up to a few unassigned qubits.
The authors discuss open problems, such as improving the runtime further and extending the techniques to more general classes of quantum states.