Research map / Probabilistic and scientific modeling / Variational and generative sampling
Detection and Localization of Changes in Conditional Distributions
Lizhen Nie, Dan L Nicolae
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 KCE (kernel conditional expectation change) instead of DXY, which treats (x,y) jointly | KCE · Experiment A, 7 regression-function changes | l1 localization error: DXY 1.7±0.3 vs KCE 2.1±0.3 on 5x; KCE 2.7±0.5 vs DXY 6.4±1.6 on cos(x) | Mixed |
| use covariate-aware KCE instead of response-only DY | KCE · Experiment A, f1=5x | l1 localization error: 322.5±55.6 -> 2.1±0.3 | Improved |
| use KCE instead of fixed-design CP method [31] (designed for equi-spaced covariates) | KCE · Experiment A, f1=cos(x) | l1 localization error: 239.1±41.1 -> 2.7±0.5 | Improved |
| use KCD instead of DXY | KCD · Experiment B, v1=x^2 | l1 localization error: 38.4±11.7 -> 12.6±3.0 | Improved |
| use KCD instead of DY | KCD · Experiment B, v1=cos(x) | l1 localization error: 146.8±43.7 -> 3.0±0.9 | Improved |
| use KCD instead of DXY | KCD · Experiment B, v1=max(0,1-x) | l1 localization error: 2.6±0.6 vs 1.8±0.3 | No clear effect |
| use KCD instead of DXY | KCD · Experiment C, Y=R^p, lambda=0.8 | l1 localization error: 9.4±3.2 vs 11.6±3.3 | No clear effect |
| use KCD instead of DXY, weaker signal | KCD · Experiment C, Y=R^p, lambda=0.6 | l1 localization error: 69.8±28.7 -> 38.2±12.9 | Improved |
| use KCD instead of DXY, weak signal on distribution space | KCD · Experiment C, Y=P, lambda=0.5 | l1 localization error: 104.7±31.7 -> 49.1±12.3 | Improved |
| use KCD instead of DXY | KCD · Experiment C, Y=P, lambda=0.8 | l1 localization error: 7.0±1.7 vs 9.2±2.3 | Hurt |
| reduce change magnitude lambda from 1.0 to 0.6 | KCD · Experiment C, Y=R^p | l1 localization error: 6.7±2.1 -> 38.2±12.9 | Hurt |
| use KCD instead of DXY for matrix-valued Y | KCD · Experiment C, Y=R^pxp sin(Xi)sin(Xj) | l1 localization error: 5.9±1.4 -> 1.4±0.6 | Improved |
Closest papers by meaning
- A nonparametric method for gradual change problems with statistical guarantees (NeurIPS 2021 · similarity 0.87)
- Change point detection and inference in multivariate non-parametric models under mixing conditions (NeurIPS 2023 · similarity 0.87)
- Inference on the Change Point under a High Dimensional Covariance Shift (NeurIPS 2024 · similarity 0.86)
- Statistically and Computationally Efficient Change Point Localization in Regression Settings (JMLR 2021 · similarity 0.86)
- Beyond Euclidean Summaries: Online Change Point Detection for Distribution-Valued Data (ICML 2026 · similarity 0.85)
- Feature Shift Detection: Localizing Which Features Have Shifted via Conditional Distribution Tests (NeurIPS 2020 · 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.
