1: \begin{abstract}

2: % summary

3: We study the trajectory of iterations and the convergence rates of the Expectation-Maximization (EM) algorithm for two-component Mixed Linear Regression (2MLR).

4: % background

5: The fundamental goal of MLR is to learn the regression models from unlabeled observations.

6: The EM algorithm finds extensive applications in solving the mixture of linear regressions.

7: % previous works

8: Recent results have established the super-linear convergence of EM for 2MLR in the noiseless and high SNR settings under some assumptions and its global convergence rate with random initialization has been affirmed.

9: % open problem

10: However, the exponent of convergence has not been theoretically estimated and the geometric properties of the trajectory of EM iterations are not well-understood. 

11: % contributions

12: In this paper, first, using Bessel functions we provide explicit closed-form expressions for the EM updates under all SNR regimes. Then, 

13: in the noiseless setting, we completely characterize the behavior of EM iterations by deriving a recurrence relation at the population level 

14: and notably show that all the iterations lie on a certain cycloid.

15: Based on this new trajectory-based analysis, we exhibit the theoretical estimate for the exponent of super-linear convergence and further improve the statistical error bound at the finite-sample level.

16: % novelty

17: Our analysis provides a new framework for studying the behavior of EM for Mixed Linear Regression.

18: \end{abstract}

19: