Robustness Verification of Tree-based Models

Authors: Hongge Chen, Huan Zhang, Si Si, Yang Li, Duane Boning, Cho-Jui Hsieh

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

Reproducibility Variable Result LLM Response
Research Type Experimental On RF/GBDT models trained on 10 datasets, our algorithm is hundreds of times faster than a previous approach that requires solving MILPs, and is able to give tight robustness verification bounds on large GBDTs with hundreds of deep trees.
Researcher Affiliation Collaboration Hongge Chen*,1 Huan Zhang*,2 Si Si3 Yang Li3 Duane Boning1 Cho-Jui Hsieh2,3 1Department of EECS, MIT 2Department of Computer Science, UCLA 3Google Research
Pseudocode Yes Algorithm 1: Enumerating all K-cliques on a K-partite graph with a known boxicity d; Algorithm 2: Multi-level verification framework
Open Source Code Yes Our code (XGBoost compatible) is available at https://github.com/chenhongge/treeVerification.
Open Datasets Yes We evaluate our proposed method for robustness verification of tree ensembles on two tasks: binary and multiclass classification on 9 public datasets including both small and large scale datasets... The datasets other than MNIST and Fashion MNIST are from LIBSVM [8].
Dataset Splits No The paper does not explicitly state the use of a distinct validation dataset split with specific percentages or counts. It mentions training and testing.
Hardware Specification Yes We run our experiments on Intel Xeon Platinum 8160 CPUs.
Software Dependencies No The paper mentions using the 'XGBoost framework [12]' but does not provide a specific version number for it or any other software dependencies.
Experiment Setup Yes The number of trees in GBDTs and parameters used in training GBDTs for different datasets are shown in Table 3 in the appendix.