AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

Fast Provably Robust Decision Trees and Boosting

Jun-Qi Guo, Ming-Zhuo Teng, Wei Gao, Zhi-Hua Zhou

ICML 2022 · Adversarial robustness and security › Certified and empirical robustness · approach: Neural network robustness verification

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Robust split optimization on adversarial 0/1 loss with leaf-aware trade-off (FPRDT) vs. local-only robust splitting (RIGDT-h, GROOT)robust decision trees · 18 datasetsadversarial accuracy (win/tie/loss): FPRDT wins 18/0/0 vs RIGDT-h and GROOT, average .7802 vs .2922 and .7233 (decision tree/GROOT)Improved
Robust boosting via PRAdaBoost (upper bound on adversarial exponential loss, FPRDT base learners) vs. standard AdaBoosttree ensembles · 18 datasetsadversarial accuracy: .7918 vs .0943 averageImproved
PRAdaBoost vs PRBoosting (same upper bound, different base learners)tree ensembles · 18 datasetsadversarial accuracyNo clear effect
FPRDT vs ROCT (global optimization) on the 18 datasetsrobust decision trees · 18 datasetsadversarial accuracy (win/tie/loss): 16/1/1; average .7802 vs .7342Improved
FPRDT running time vs provably robust methodsrobust decision trees · 18 datasetsrunning time: smallest running time among provably robust methodsImproved
PRAdaBoost vs PRBoosting running timetree ensembles · 18 datasetsrunning time: about 10 times fasterImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy