Quantum advantage in learning from experiments