Research map / Learning theory and algorithms / Prediction and decision losses
RieszNet and ForestRiesz: Automatic Debiased Machine Learning with Neural Nets and Random Forests
Victor Chernozhukov, Whitney Newey, Vı́ctor M Quintas-Martı́nez, Vasilis Syrgkanis
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 the doubly robust moment rather than the direct plug-in moment | RieszNet · IHDP ATE | MAE: 0.123 -> 0.110 | Improved |
| Use the doubly robust moment rather than the direct plug-in moment | ForestRiesz · IHDP ATE | MAE: 0.197 -> 0.126 | Improved |
| Use RieszNet DR instead of Dragonnet | IHDP ATE · IHDP ATE | MAE: 0.146 ± 0.010 -> 0.110 ± 0.003 | Improved |
| Use ForestRiesz DR instead of RF plug-in | IHDP ATE · IHDP ATE | MAE: 0.389 ± 0.024 -> 0.126 ± 0.004 | Improved |
| Add the post-processing TMLE adjustment to DR | ForestRiesz · BHP complex f with linear and non-linear confounders | RMSE: 0.377 -> 0.327 | Improved |
| Add the post-processing TMLE adjustment to DR | ForestRiesz · BHP complex f with linear and non-linear confounders | coverage: 0.909 -> 0.953 | Improved |
| Replace shared multitasking networks with separate regression and Riesz networks (ablation) | RieszNet · IHDP ATE | RMSE: 0.147 -> 0.411 | Hurt |
| Freeze shared layers after training on the Riesz loss (ablation) | RieszNet · IHDP ATE | RMSE: 0.147 -> 1.221 | Hurt |
| Replace end-to-end TMLE adjustment training with post-processing (ablation) | RieszNet · IHDP ATE | bias: -0.044 -> -0.088 | Hurt |
| Remove cross-fitting (ablation) | ForestRiesz DR · BHP complex f with linear and non-linear confounders | coverage: 0.909 -> 0.894 | Hurt |
| Remove cross-fitting while retaining multitasking (ablation) | ForestRiesz DR · BHP complex f with linear and non-linear confounders | coverage: 0.909 -> 0.884 | Hurt |
| Remove multitasking while retaining simple cross-fitting (ablation) | ForestRiesz DR · BHP complex f with linear and non-linear confounders | RMSE: 0.377 -> 0.365 | Hurt |
Closest papers by meaning
- Localized Debiased Machine Learning: Efficient Inference on Quantile Treatment Effects and Beyond (JMLR 2024 · similarity 0.87)
- DeNetDM: Debiasing by Network Depth Modulation (NeurIPS 2024 · similarity 0.84)
- Generalization in Kernel Regression Under Realistic Assumptions (ICML 2024 · similarity 0.84)
- Generalized Random Forests Using Fixed-Point Trees (ICML 2025 · similarity 0.83)
- SEBRA : Debiasing through Self-Guided Bias Ranking (ICLR 2025 · similarity 0.83)
- Kernel Debiased Plug-in Estimation: Simultaneous, Automated Debiasing without Influence Functions for Many Target Parameters (ICML 2024 · similarity 0.83)
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.
