Unsupervised state learning from pairs of states

Suppose you receive a sequence of qubits where each qubit is guaranteed to be in one of two pure states, but you do not know what those states are. Your task is to either determine the states or to construct a POVM (Positive Operator Valued Measure) that will discriminate them. This can be viewed as a quantum analog of unsupervised learning. A problem is that without more information, all that can be determined is the density matrix of the sequence, and, in general, density matrices can be decomposed into pure states in many different ways. To solve the problem additional information, either classical or quantum, is required. We show that if an additional copy of each qubit is supplied, that is, one receives pairs of qubits, both in the same state, rather than single qubits, the task can be accomplished. We then simulate numerically the measurement of a sequence of qubit pairs and show that the unknown states and their respective probabilities of occurrence can be found with high accuracy.