Sample Efficient Active Learning of Causal Trees

Authors: Kristjan Greenewald, Dmitriy Katz, Karthikeyan Shanmugam, Sara Magliacane, Murat Kocaoglu, Enric Boix Adsera, Guy Bresler

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We consider several experimental settings and for each setting we simulate 200 random trees of n nodes. Here we present a subset of the results, more are described in Appendix I. We generate an undirected tree with three different strategies: a) sampling uniformly from the space of undirected trees, b) generating power-law trees, and c) generating high degree d = n/2 random graphs and then creating an undirected version of the BFS tree.
Researcher Affiliation Collaboration Kristjan Greenewald IBM Research MIT-IBM Watson AI Lab kristjan.h.greenewald@ibm.com Dmitriy Katz IBM Research MIT-IBM Watson AI Lab dkatzrog@us.ibm.com Karthikeyan Shanmugam IBM Research MIT-IBM Watson AI Lab karthikeyan.shanmugam2@ibm.com Sara Magliacane IBM Research MIT-IBM Watson AI Lab sara.magliacane@ibm.com Murat Kocaoglu IBM Research MIT-IBM Watson AI Lab murat@ibm.com Enric Boix-Adser a MIT MIT-IBM Watson AI Lab eboix@mit.edu Guy Bresler MIT MIT-IBM Watson AI Lab guy@mit.edu
Pseudocode Yes Algorithm 1 Central Node Algorithm input Observational tree G0. Confidence parameter δ. 1: t 0. 2: q0(i) 1 n, 8i = 1, . . . , n. 3: while maxi qt(i) 1 δ do 4: t t + 1. 5: Identify central node index vc(t) of G with respect to qt 1 (Algorithm 4). 6: Intervene on node vc(t) and observe x1, . . . , xn. 7: Update posterior distribution qt as given in Lemma 1. 8: end while output argmaxi qt(i) as the estimated root node of G0.
Open Source Code No The paper does not contain any explicit statement about releasing source code or a link to a code repository.
Open Datasets No We consider several experimental settings and for each setting we simulate 200 random trees of n nodes. We generate an undirected tree with three different strategies: a) sampling uniformly from the space of undirected trees, b) generating power-law trees, and c) generating high degree d = n/2 random graphs and then creating an undirected version of the BFS tree. Once we have a tree, we pick the root node uniformly at random. In this section we focus on binary random variables, where each variable is a function of its parent: if XP ai = 0, then Xi Bern ( ), else Xi Bern (1 ), where for each variable we sample uniformly from [δ, 0.5 δ]. The root node is distributed as Xr Bern (0.5). We show similar results with discrete variables in Appendix I.
Dataset Splits No The paper describes generating random trees and performing simulations but does not specify any train/validation/test dataset splits, cross-validation, or random seeds for data partitioning.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running the experiments. It only mentions 'simulated experiments'.
Software Dependencies No The paper mentions 'simulated experiments' but does not list any specific software dependencies with version numbers (e.g., programming languages, libraries, or solvers).
Experiment Setup Yes We consider several experimental settings and for each setting we simulate 200 random trees of n nodes. ... We generate an undirected tree with three different strategies: a) sampling uniformly from the space of undirected trees, b) generating power-law trees, and c) generating high degree d = n/2 random graphs and then creating an undirected version of the BFS tree. Once we have a tree, we pick the root node uniformly at random. In this section we focus on binary random variables, where each variable is a function of its parent: if XP ai = 0, then Xi Bern ( ), else Xi Bern (1 ), where for each variable we sample uniformly from [δ, 0.5 δ]. The root node is distributed as Xr Bern (0.5).