Research map / Label-efficient and robust learning / Recognition and long-tail
Extrapolating Expected Accuracies for Large Multi-Class Problems
Charles Zheng, Rakesh Achanta, Yuval Benjamini
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 |
|---|---|---|---|
| increase source classes k1 from 500 to 5000 at 2x extrapolation | ClassExReg · Gaussian simulation | max RMSE: 0.032 -> 0.009 | Improved |
| use ClassExReg instead of KDE-BCV | extrapolation method · Gaussian simulation (k1=500 to k2=1000) | max RMSE: 0.090 -> 0.032 | Improved |
| use ClassExReg instead of KDE-UCV | extrapolation method · Gaussian simulation (k1=5000 to k2=100000) | max RMSE: 0.086 -> 0.054 | Improved |
| use ClassExReg instead of KDE-UCV | extrapolation method · Gaussian simulation (k1=500 to k2=10000) | max RMSE: 0.045 vs 0.098 | Hurt |
| use ClassExReg instead of KDE-BCV | extrapolation method · Face recognition LFW (k1=100 to 1672 classes) | RMSE: 0.053 vs 0.113 | Hurt |
| increase k1 from 100 to 400 | ClassExReg · Face recognition LFW | RMSE: 0.113 -> 0.050 | Improved |
| use ClassExReg instead of KDE-BCV | extrapolation method · Telugu OCR Deep CNN (20 to 100 classes) | extrapolated accuracy: 0.9905 vs 0.7138 (true 0.9908) | Improved |
| use ClassExReg instead of KDE-BCV | extrapolation method · Telugu OCR Logistic (20 to 100 classes) | extrapolated accuracy: 0.8980 vs 0.8414 (true 0.8490) | Hurt |
| use ClassExReg instead of KDE methods | extrapolation method · Telugu OCR (20 to 400 classes) | extrapolated accuracy | Mixed |
| use ClassExReg for exponentiation bias correction instead of KDE | extrapolation method · Gaussian simulation | bias | Improved |
Closest papers by meaning
- Predicting Classification Accuracy When Adding New Unobserved Classes (ICLR 2021 · similarity 0.89)
- Demystifying the Optimal Performance of Multi-Class Classification (NeurIPS 2023 · similarity 0.88)
- Estimating Accuracy from Unlabeled Data: A Bayesian Approach (ICML 2016 · similarity 0.88)
- Theoretical Insights Into Multiclass Classification: A High-dimensional Asymptotic View (NeurIPS 2020 · similarity 0.88)
- A Data-Driven Measure of Relative Uncertainty for Misclassification Detection (ICLR 2024 · similarity 0.85)
- How many classifiers do we need? (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.
