Research map / Probabilistic and scientific modeling / Variational and generative sampling
Adversarial Monte Carlo Meta-Learning of Optimal Prediction Procedures
Alex Luedtke, Incheoul Chung, Oleg Sofrygin
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 |
|---|---|---|---|
| make Module 1 equivariant to feature permutations only, dropping observation-permutation invariance (ablation) | AMC100 · FLAM sparse, n=100 | MSE: 7-fold, 6-fold, 5-fold, and 4-fold increase in MSE across scenarios 1-4 | Hurt |
| add adversarial meta-training instead of lasso/OLS (boundary prior, sparse signal, n=100) | AMC100 · Linear regression sparse boundary n=100 | MSE: 0.06 -> 0.02 | Improved |
| use AMC instead of lasso under the interior prior (sparse signal, n=100) | AMC100 · Linear regression sparse interior n=100 | MSE: 0.06 -> 0.11 | Hurt |
| meta-train at n_mt=500 instead of 100 (interior sparse, n=100) (ablation) | AMC · Linear regression sparse interior n=100 | MSE: 0.11 -> 0.07 | Improved |
| meta-train at n_mt=500 instead of 100 (interior sparse, n=500) (ablation) | AMC · Linear regression sparse interior n=500 | MSE: 0.09 -> 0.04 | Improved |
| use AMC with n_mt=n in dense linear regression | AMC · Linear regression dense signal | MSE: 0.13 -> 0.10 (OLS vs AMC100, boundary n=100) | Improved |
| use AMC instead of FLAM (sparse, scenario 2, n=100) | AMC100 · FLAM sparse scenario 2 n=100 | MSE: 0.47 -> 0.18 | Improved |
| use AMC500 instead of FLAM (dense, scenario 4, n=100) | AMC500 · FLAM dense scenario 4 n=100 | MSE: 0.76 -> 0.36 | Improved |
| meta-train at n_mt=100 instead of 500 (dense FLAM, scenario 1, n=500) (ablation) | AMC · FLAM dense scenario 1 n=500 | MSE: 0.15 -> 0.91 | Hurt |
| use AMC500 instead of FLAM (sparse scenario 1, n=100) | AMC500 · FLAM sparse scenario 1 n=100 | MSE: 0.44 -> 0.48 | Hurt |
| stack the four AMC estimators with NNLS | Stacked AMC · UCI fish, airfoil, yacht datasets | MSE | Improved |
| use lasso instead of single AMC linear estimators | lasso · UCI real datasets | MSE | Mixed |
Closest papers by meaning
- A representation-learning game for classes of prediction tasks (ICLR 2024 · similarity 0.87)
- Blessings and Curses of Covariate Shifts: Adversarial Learning Dynamics, Directional Convergence, and Equilibria (JMLR 2024 · similarity 0.86)
- Meta-Learning Adversarial Bandit Algorithms (NeurIPS 2023 · similarity 0.86)
- Model-based Adversarial Meta-Reinforcement Learning (NeurIPS 2020 · similarity 0.86)
- Adversaries With Incentives: A Strategic Alternative to Adversarial Robustness (ICLR 2025 · similarity 0.86)
- Learning Not to Regret (AAAI 2024 · similarity 0.85)
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.
