In Figure 8.14(b), we have increased the noisiness of the third mixture component in the observations by letting images . The performance degradation of both algorithms is almost negligible.

Though DPMM-based approach seems a powerful tool for Bayesian nonparametric classification, arguably, its drawback is the complexity involved and the need to carefully select certain hyperparameters. In particular, there is the need for specifying the base prior and the parameter images . As we already discussed, it is computationally convenient to select the base prior to be conjugate for the assumed observation likelihood. However, in practice, this has to be properly justified. The choice of images seems to be somewhat arbitrary. However, without enough insight into the application scenario, it may be difficult to specify a suitable value for images . For this reason, sometimes, it is also modeled as having a certain prior distribution. Then, during the Gibbs sampling procedure, the posterior distribution of images is used ...

Get Signal Processing for Cognitive Radios now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.