Course Outline and Administrative Issues

Adaptive Information Processing (5SSB0)

AIP Logistic Issues

Course comes in Two Parts

Part I: Linear Gaussian Models and the EM Algorithm
Part II: Model Complexity Control and the MDL Principle
  • $4 \times 4=16$ lecture hours, mostly in March
  • Instructor: Tjalling Tjalkens, rm. FLUX-7.101
  • email [email protected]

Why Take This Class?

  • Suppose you need to develop an algorithm for a complex DSP task. This is what you'll do:

    1. Choose a set of candidate algorithms $y=H_k(x;\theta)$ where $k \in \{1,2,\ldots,K\}$ and $\theta \in \Theta_k$; (you think that) there's an algorithm $H_{k^*}(x;\theta^*)$ that performs according to your liking.
    2. You collect a set of examples $D=\{(x_1,y_1),(x_2,y_2),\ldots,(x_N,y_N)\}$ that are consistent with the correct algorithm behavior.
  • Using the methods from this class, you will be able to design a suitable algorithm through:
    1. model selection, i.e., find $k^*$ (mostly discussed in part 2 of this class)
    2. parameter estimation, i.e., find $\theta^*$ (mostly in part 1 of this class)

Materials

  • Background reading; covers about the same stuff as (mandatory) slides.
  • Contains much more material; great for future study and reference.

Exam Guide

  • Tested material consists of these lecture notes, reading assignments (as assigned in the first cell/slide of each lecture notebook) and exercises (see class website).
  • Slides that are not required for the exam are indicated by the word (OPTIONAL) in the header.
  • Very strong advice: Make old exams!
  • You are not allowed to use books nor bring printed or handwritten formula sheets to the exam. Difficult-to-remember formulas are supplied at the exam sheet (see old exams).
  • You may use a simple pocket calculator, but no smartphones (only arithmetic assistance is allowed.)
  • Further exam instructions for part-2 from Tjalling.
In [1]:
open("../../styles/aipstyle.html") do f
    display("text/html", readstring(f))
end