Philippe Brouillard (@_kurowasan) profile picture

Philippe Brouillard

@_kurowasan
Joined July 2016
  • Tweets7
  • Following126
  • Followers25
1/ We’re presenting “Typing assumptions improve identification in causal discovery” today at the CausalAssumptions workshop @icmlconf. TLDR: causal discovery -- variable types can help! Paper: arxiv.org/abs/2107.10703 @ceer2k @alex_lacoste_ @seblachap @alexandredrouin
Image from Philippe Brouillard (@_kurowasan)
2
12
2
33
more replies
5/ We show theoretically and empirically that, in random graphs with a fixed number of types, the equivalence class converges exponentially fast to a singleton as more variables are observed (identification).
Image from Philippe Brouillard (@_kurowasan)
1
6/ We’re actively looking for real-world datasets to apply this work and we’d be happy to hear your suggestions! Key requirement: 1) variables can be segmented into categories 2) categories dictate the plausibility of causal relationships.
I just voted for Crucifix (English subtitles) awardeo.tv/video/80514481/57… on @Awardeocompany
Load more