ISIPTA school day 4

Oulah, les fonctions transcendantales le lundi matin ça déchire la tête. Les Gamma, Beta, BeBi, Diri, DiMn, factorielle ascendante et autres coefficients binomiaux généralisés n'ont plus de secret pour nous. Les processus de Markov non plus.

Vocabulary warning: statistical "inference" is about learning from data. Do not confuse with logical "inference" which is about reasoning.

There are 2 ways to say that the future will look like the past. One is to assume a random sampling model, the other is to assume exchangeability.

In a probability tree, nodes represent values of variables. In a bayesian or credal network, nodes represent the variables themselves.

An imprecise Markov chain can be seen as a credal net, where the graph is a chain (1 parent, 1 child), under the "epistemic irrelevance" notion of independance.

Ideas: Do experimental economics to assess empirically the hyperparameter s. Use real-world data, for example course choice by students are gambles...
What is the expected value of future information in an imprecise model ?
Make a javascript interactive page for the IDM and IDMM (people type in the counts, they get lower and upper probabilities, expectations and expected frequencies for s=1 and s=2, and n'=1 / custom. ).