Supplementary Material: Multiple Clustering Views from Multiple Uncertain Experts

For variational inference in our approach, we use the following parameter settings 1. G, the number of components in truncated Dirichlet Process, is set to be M/2, where M is the total number of experts. In this way, we try to enforce the constraint that on average, there should be at least two experts in each view. In all experiments, the number of expert views recovered by our approach is smaller than G = M/2. Therefore, the value of G we use is large enough to discover the true number of expert views. 2. For the parameters of prior distributions: • p(αm), p(βm): set the parameters of prior Beta distributions to be (10, 1) to incorporate the prior knowledge that each expert’s accuracy parameters should be far away from 0.5 (random guess) and close to 1. The choice can be illustrated from Figure 1. Under this setting, there is very small probability that the accuracy parameters can be close to 0.5.