Supplementary Documents for “ Semi-Crowdsource Clustering : Generalizing Crowd Labeling by Robust Distance Metric Learning ”

where T (A′) denotes the space spanned by the elements of the form uky and xv> k , for 1 ≤ k ≤ r, Ω(A′) denotes the space of matrices that have the same support toA′, ‖ ·‖ denotes the spectral norm and ‖ · ‖∞ denotes the largest entry in magnitude. Lemma 1. Let A∗ ∈ RN×N be a similarity matrix of rank r obeying the incoherence properties (A1) and (A2), with μ = max(μ0, μ1). Suppose we observe m1 entries of A∗ recorded in Ã