Information and coding theory (2014-2015)

Last modification: 24 April 2015 15:25

This is not the good page: See new page

Evaluations

Feedbacks during Q&A sessions.

Informations

  • Project 4 (optional) available! (statement, data)
  • Follow the guidelines to submit your project. The deadline is the deadline: delays are not acceptable. The mail subject and the name of the archive have to be correct.
  • Bayes Net Toolbox for Matlab (required for projects 1, 2 and 3): Download, Installation and Documentation.

Agenda

3/02 Theory Cours cancelled
10/02 Exercise session 1 Information measures (statement)
17/02 - Carnival holidays (Congé de carnaval)
24/02 Theory Project 1: statement, data
03/03 Theory

Q&A


Project 1
10/03 Theory

Q&A


Project 1
15/0 Deadline Deadline for project 1
17/03 Theory

Q&A


Project 2: statement, data
24/03 Theory

Q&A


Project 2
29/03 Deadline Deadline for project 2
31/03 Theory

Feedback
Project 3: statement, data

Project 1
14/04 Theory

Q&A


Project 3
21/04 Theory

Q&A


Projects 3 & 4
22/04 Deadline Deadline for project 3
28/04 Theory

Q&A

Feedbacks


Project 4: statement, data

Projects 2 & 3
3/05 Deadline Deadline for project 4 (optional)
5/05 Theory

Exercise session 2


Summary (statement)

Frequently asked questions (F.A.Q.)

It is strongly recommanded to participate to Q&A sessions and to read the F.A.Q. section before any mail question.

    Project 1

  • For questions 1 to 4, you have to explain what you do to compute the measures. Copy the code in the report is not an explanation (so don't do it). I am looking for an explanation of the choices you have made to correctly implement your function. However, it is not necessary to say "There is a loop to consider all lines of the array, [...]". Instead, try to be consistent with the subject of the course. Moreover, it is not useful to repeat at each question the same thing.
  • Project 2

  • For question 4, let assume that we analyse the node 3, and we are looking nodes 1 and 2 as candidates. If we observe that conditions to detect a v-structure are met, we identify nodes 1 and 2 as parents. However it is not the case, what should be done? Can I or should I, based on data, deduce which nodes among 1 and 2 is the parent of 3?
    The oracle is the only source you can use.
    - If a node is not a summit of a v-structure, you have to analyse the next node.
    - If a node is a summit of a v-structure, you can give a orientation to links between parents and summit. Then you have to propagate the orientation information in the rest of the graph (where it can be done).
    For this part of the project, data from first project can not be used. Moreover, except the v-structure detection which involves independence relationships, it is extremely complicated to find causal relationships between two nodes (node 1 --> node 3 or vice versa) from observed data.
  • Can we use ready-made Matlab functions?
    If it is not explicity asked to implement the function in the statement, you can use such functions. However, pay attention to the algorithm implemented in theses functions and verify it has the same properties and output as the algorithm presented in the theoretical lesson. Be also careful to give the right explanations that correspond to the algorithms/functions you use. Do not forget to give your sources.
  • What means "independences" and "independence set"?
    "Independences" means the relationships of independence between two (or more) nodes. And the "independence" set means all these independence relationships.