Research map / Probabilistic and scientific modeling / Variational and generative sampling
Likelihood over Estimation: Robust Quadratic Discriminant Analysis for Heavy-Tailed Distributions with Theory and Evidence
Niranjana Ambadi, eugene pinsky
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 |
|---|---|---|---|
| replace Gaussian likelihood with surrogate alpha-stable likelihood (mean + Ledoit-Wolf) | QDA · Credit Card Fraud | PR-AUC: 0.488 -> 0.739 | Improved |
| replace Gaussian likelihood with surrogate alpha-stable likelihood (mean + Ledoit-Wolf) | QDA · HTRU2 | recall at 95% precision: 0.707 -> 0.796 | Improved |
| replace Gaussian likelihood with stable likelihood | QDA · Ionosphere | accuracy (%): 87.46 -> 91.74 | Improved |
| use Tyler's M-estimator for dispersion instead of Ledoit-Wolf shrinkage (ablation) | Stable-QDA · HTRU2 | accuracy (%): 97.83 -> 93.85 | Hurt |
| use Tyler's M-estimator for dispersion instead of Ledoit-Wolf shrinkage (ablation) | Stable-QDA · Credit Card Fraud | PR-AUC: 0.739 -> 0.437 | Hurt |
| use Tyler's M-estimator on Ionosphere (ablation) | Stable-QDA · Ionosphere | accuracy (%): 91.74 -> 92.03 | No clear effect |
| use spatial median instead of sample mean with Ledoit-Wolf (ablation) | Stable-QDA · HTRU2 | accuracy (%): 97.83 -> 97.81 | No clear effect |
| use a fixed alpha 1.5 instead of the true tail index (ablation) | Stable-QDA · Synthetic stable data | accuracy (%): matches oracle within 0.1% | No clear effect |
| estimate alpha with McCulloch's method instead of fixed 1.5 (ablation) | Stable-QDA · Synthetic stable data | accuracy (%): 76.4 -> 76.4 average | No clear effect |
| use per-class alpha instead of shared alpha (ablation) | Stable-QDA · Synthetic stable data (alpha 1.0 and 1.8) | accuracy (%): 77.1 -> 74.1 | Hurt |
| use Gaussian QDA on training data contaminated with 20% outliers | Gaussian QDA · Stable alpha 1.8 base data | accuracy (%): 91.0 -> 73.6 | Hurt |
| use Tyler's M-estimator under adversarial outliers (ablation) | Stable-QDA · Gaussian base, 10% adversarial outliers | accuracy (%): 93.1 -> 90.8 | Hurt |
Closest papers by meaning
- Outlier-Robust Sparse Mean Estimation for Heavy-Tailed Distributions (NeurIPS 2022 · similarity 0.87)
- Robust and Heavy-Tailed Mean Estimation Made Simple, via Regret Minimization (NeurIPS 2020 · similarity 0.86)
- Sub-Gaussian Confidence Intervals for Heavy-Tailed Data: Characterizing the Limits of Inference (NeurIPS 2026 · similarity 0.85)
- Long-Tailed Out-of-Distribution Detection: Prioritizing Attention to Tail (AAAI 2025 · similarity 0.84)
- Near-Optimal Streaming Heavy-Tailed Statistical Estimation with Clipped SGD (NeurIPS 2024 · similarity 0.84)
- A Separation in Heavy-Tailed Sampling: Gaussian vs. Stable Oracles for Proximal Samplers (NeurIPS 2024 · 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.
