Research map / Learning theory and algorithms / Prediction and decision losses
Measures of distortion for machine learning
Leena Chennuru Vankadara, Ulrike von Luxburg
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 sigma-distortion instead of average distortion as the measure | Isomap embeddings of Gaussian data · Embedding dimension vs distortion | distortion decreases with embedding dimension | Improved |
| use sigma-distortion or epsilon-distortion instead of worst-case, average, normalized average, k-local | Isomap embeddings of gamma and beta data · Original dimension 10 to 100, fixed embedding dimension | distortion increases with original dimension | Improved |
| use sigma-distortion instead of other measures under additive noise | Mixture-of-Gaussians data with noise in R20 · Variance of noise vs distortion | distortion increases with noise variance | Improved |
| increase variance of additive noise | SVM and kNN on Isomap embeddings · Noisy Gaussian mixture classification | classification loss | Hurt |
| use sigma-distortion or epsilon-distortion as indicator instead of worst-case, average, k-local | Embeddings from PCA, GPLVM, Isomap, MVU, SPE · Distortion vs classification loss (kNN and SVM) | monotone relation with classification loss | Improved |
| use other embedding algorithms (PCA, MVU) instead of Isomap | noise experiments · Noise vs classification loss | classification loss trend | No clear effect |
| use sigma-distortion instead of other measures regarding variability across repetitions | Isomap embeddings · Original dimension vs distortion error bars | std over 10 repetitions | Improved |
Closest papers by meaning
- Dimensionality reduction: theoretical perspective on practical measures (NeurIPS 2019 · similarity 0.87)
- Kernel Distribution Embeddings: Universal Kernels, Characteristic Kernels and Kernel Metrics on Distributions (JMLR 2018 · similarity 0.82)
- On the Power and Limits of Distance-Based Learning (ICML 2016 · similarity 0.82)
- Simpson’s Bias in NLP Training (AAAI 2021 · similarity 0.82)
- Consistency Analysis for Binary Classification Revisited (ICML 2017 · similarity 0.82)
- Learning Low-Dimensional Metrics (NeurIPS 2017 · similarity 0.81)
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.
