Research map / Learning theory and algorithms / Prediction and decision losses
Adapting and Evaluating Influence-Estimation Methods for Gradient-Boosted Decision Trees
Jonathan Brophy, Zayd Hammoudeh, Daniel Lowd
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use GBDT models instead of LR/DT/KNN/SVM/RF/MLP | tabular prediction · 22 tabular data sets | test AUC/accuracy/MSE: Concrete MSE 50.6 -> 18.8 | Improved |
| use BoostIn instead of Random for multi-test removal | influence estimation for GBDT · multiple test instances, removing examples | loss increase relative to random: 2.2x | Improved |
| use BoostIn instead of LOO for removing a group of examples | influence estimation for GBDT · multiple test instances, removing examples | average rank | Improved |
| use LeafRefit instead of BoostIn for single-test removal | influence estimation for GBDT · single test instance, removal | average rank | Improved |
| use LeafRefit/TREX/TreeSim instead of BoostIn for targeted label edits | influence estimation for GBDT · single test instance, targeted label edits | average rank | Improved |
| add fixed-structure assumption (LeafRefit) to LOO (ablation) | LOO · single test instance, removal | test loss increase | Improved |
| reestimate influence after each removal instead of fixed ordering (ablation) | all influence methods · one-at-a-time removal, LGB | test loss increase | Improved |
| use LeafInfluence/LeafRefit instead of LOO for setup time | influence estimation · runtime, SDS data sets | setup time | Hurt |
| use BoostIn instead of Loss baseline for fixing mislabelled examples | influence estimation · fix mislabelled, 40% flipped labels | average rank | Improved |
| use BoostIn instead of LeafInfSP for adding label noise | influence estimation · multiple test instances, label noise | average rank | Improved |
Closest papers by meaning
- Finding Influential Training Samples for Gradient Boosted Decision Trees (ICML 2018 · similarity 0.90)
- LightGBM: A Highly Efficient Gradient Boosting Decision Tree (NeurIPS 2017 · similarity 0.87)
- Most Influential Subset Selection: Challenges, Promises, and Beyond (NeurIPS 2024 · similarity 0.85)
- Neural Networks for Learnable and Scalable Influence Estimation of Instruction Fine-Tuning Data (NeurIPS 2025 · similarity 0.84)
- DeepSN: A Sheaf Neural Framework for Influence Maximization (AAAI 2025 · similarity 0.84)
- Towards Robust Influence Functions with Flat Validation Minima (ICML 2025 · similarity 0.84)
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.
