Add to Chrome
Add to Firefox
Add to EdgeKeep or toss? A nonparametric score to evaluate solutions for noisy ICA
Jan 16, 2024
In this paper, we propose a non-parametric score to evaluate the quality of the solution to an iterative algorithm for Independent Component Analysis (ICA) with arbitrary Gaussian noise. The novelty of this score stems from the fact that it just assumes a finite second moment of the data and uses the characteristic function to evaluate the quality of the estimated mixing matrix without any knowledge of the parameters of the noise distribution. We also provide a new characteristic function-based contrast function for ICA and propose a fixed point iteration to optimize the corresponding objective function. Finally, we propose a theoretical framework to obtain sufficient conditions for the local and global optima of a family of contrast functions for ICA. This framework uses quasi-orthogonalization inherently, and our results extend the classical analysis of cumulant-based objective functions to noisy ICA. We demonstrate the efficacy of our algorithms via experimental results on simulated datasets.
