AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

Robust Watermarking on Gradient Boosting Decision Trees

Jun Woo Chung, Yingjie Lao, Weijie Zhao

AAAI 2026 · Learning theory and algorithms › Prediction and decision losses · approach: Tree ensemble optimization · 0 citations (OpenAlex)

Read the paper

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
choose watermark samples with Cluster Center Flip instead of randomGBDT watermarking · Dcand=Dtrain avg, ratio 0.01watermark effectiveness: 0.819 -> 0.980Improved
choose watermark samples with Confidence Flip instead of randomGBDT watermarking · Dcand!=Dtrain avg, ratio 0.1watermark effectiveness: 0.800 -> 0.947Improved
choose watermark samples with Cluster Center Flip instead of randomGBDT watermarking · Dcand!=Dtrain avg, ratio 0.1adjusted model accuracy: 0.714 -> 0.851Improved
choose Cluster Center Flip instead of random for robustnessGBDT watermarking · Dcand!=Dtrain avg, ratio 0.1fine-tuning robustness: 0.684 -> 0.918Improved
choose Confidence Flip (max-distance) instead of random for robustnessGBDT watermarking · Dcand=Dtrain avg, ratio 0.1fine-tuning robustness: 0.934 -> 0.989Improved
use Outlier Flip at high watermark ratioGBDT watermarking · Dcand!=Dtrain avg, ratio 0.1watermark effectiveness: 0.863 vs 0.947 for Confidence FlipHurt
use Outlier Flip at low watermark ratioGBDT watermarking · Dcand=Dtrain avg, ratio 0.001watermark effectiveness: 0.896 vs 0.694 randomImproved
use Wrong Prediction Flip when Dcand=DtrainGBDT watermarking · Dcand=DtrainapplicabilityHurt
add in-place fine-tuning instead of adding treesGBDT watermarking · designrobustness to tree pruningImproved
use maximum-distance selection instead of lowest-confidence selection (ablation)GBDT watermarking · Dcand=Dtrain avg, ratio 0.1watermark effectiveness: 0.999 vs 0.999 (Confidence Flip)No clear effect
duplicate watermark samples by factor 5 when Dcand=DtrainGBDT watermarking · Dcand=Dtrainwatermark effectivenessImproved

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