Problem Set 4 (26.5 - 2.6.08)

In this exercise we want to implement the SFA algorithm (see lecture 5). More information on SFA can for example be found in the paper Slow feature analysis: Unsupervised learning of invariances. (Wiskott, L. and Sejnowski, T.J.).

Exercise 1

  1. Implement the SFA algorithm. As usual, the algorithm should be encapsulated in an object oriented way (e.g. as an object with a train and an execute method). Instead of writing your own implementation you can also use the MDP toolkit.
  2. Now create a linear mixture of some slow sine signals and fast noise signals. Use SFA to extract the original slow signals. Visualize the combined signals and the extracted slow features (i.e. the slow sinals).
  3. Write a simple unittest to check that the algorithm runs without raising any exceptions.