The algorithm for EM is pretty similar to the K-means algorithm. It keeps optimizing both parameters in an iterative manner. Let's go over these steps:
- First, it initializes the Gaussian distribution's positions and prior distribution parameter,
.
- Expectation: Then, it estimates the expected value of z by estimating
.
- Maximization: Next, it updates the mean of the Gaussian distribution to the center of the new clusters that are led by the new cluster assignment.
- Repeat steps 2-3 until the cluster assignment remains unchanged or ...