Visual Analogy Imagine you have a map with several locations (data points). Instead of pinpointing each location exactly, you draw a circle (distribution) around each point. The size of the circle depends on how certain you are about the point’s location:
Mean (μ): The center of the circle. Standard Deviation (σ): The radius of the circle. When you sample a point, you randomly pick a point within the circle:
Random Noise (ε): Determines the exact position within the circle. Reparameterization: Ensures you can adjust the circle’s position and size (mean and standard deviation) during training.