Fig. 3
From: Demonstration of quantum advantage in machine learning
Learning error probability \(\overline{p}\) averaged over all the n-bit oracles k, for different n and solvers. a n = 2, b n = 3. Making use of the analog measurements \(\{{V}_{{D}_{1}},\ldots {V}_{{D}_{n}},{V}_{A}\}\) (squares) improves over the digital solvers in Fig. 2 (circles) for both classical (empty symbols) and quantum (solid symbols) learning. The analog solver in Q proves to be the most efficient solution. Moreover, the gap between Q and C grows with n. The same dataset is used in Figs 2 and 3, with D 3 ignored in the analysis for n = 2. See Supplementary Information for the p(N) corresponding to each 3-bit k
