Graph model and diffusion maps algorithm

[@bubbletea2223]

this channel is a hidden gem, thank you!

[@HauTiengWuMath]

Thank you! 😄

[@sophiaxiao5071]

Thank you for the lecture. In one of your lectures, you showed that Laplacian eigenmaps can be considered a special case of diffusion maps where t=0. I was told that DM supposedly is better for data that is nonuniformly sampled, but I am not sure what the implications of this statement are for common medical signals of interest (e.g., PPG, ECG, EEG). In what scenario (e.g., which kinds of signals, sampling rate, etc) would you prefer to use Laplacian eigenmaps over diffusion maps where $t eq 0$?

[@HauTiengWuMath]

For DM, if you consider the "alpha-normalization", you can alleviate the impact of nonuniform sampling with theoretical guarantee (It will be discussed in the theoretical session later). In my experience, using DM is better than using eigenmaps, particularly if you want to reserve as much as possible the metric information (it will also be discussed in the theoretical session later).