Research map / Label-efficient and robust learning / Recognition and long-tail
The Extreme Value Machine
Ethan M. Rudd, Lalit P. Jain, Walter J. Scheirer, Terrance E. Boult
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 |
|---|---|---|---|
| reduce Letter training points kept to 10% via larger redundancy threshold (varsigma 0.008) (ablation) | EVM · Letter (closed set) | accuracy: 96% -> 92% | Hurt |
| keep only 50% or 40% of Letter training points via model reduction (ablation) | EVM · Letter (closed set) | accuracy: 96% base; 50% (varsigma = 0.492) and 40% (varsigma = 0.186) comparable | No clear effect |
| average over k>1 largest probabilities per class instead of k=1 (ablation) | EVM · Letter / OLETTER | accuracy: slight gains of 1-2% in accuracy | Improved |
| use cosine similarity instead of Euclidean distance for margins (ablation) | EVM · Open world ImageNet | open world accuracy / F1 | Improved |
| use cosine similarity instead of Euclidean distance (ablation) | NNO · Open world ImageNet | open world accuracy | Hurt |
| replace NNO with EVM for open world recognition | open world ImageNet protocol · Open world ImageNet | accuracy and F1 | Improved |
| compare EVM with W-SVM on OLETTER | EVM · OLETTER | F1-Measure | No clear effect |
| use EVM instead of SVM models for model size | EVM · OLETTER | vector ratio: an order of magnitude smaller than that of any of the SVM models | Improved |
| vary k in 1..10 and varsigma (ablation) | EVM · accuracy / F1 | accuracy or F1: less than 2% in accuracy or F1 | No clear effect |
| use incremental KDE classifier instead of EVM | open world ImageNet · Open world ImageNet | scalability: fewer than 40 samples enrolled after 24 hours; approximately 18 years | Hurt |
Closest papers by meaning
- Class-Incremental Learning: Survey and Performance Evaluation on Image Classification (TPAMI 2022 · similarity 0.87)
- Class-Incremental Learning With Strong Pre-Trained Models (CVPR 2022 · similarity 0.86)
- IL2M: Class Incremental Learning With Dual Memory (ICCV 2019 · similarity 0.86)
- Class-Incremental Learning with CLIP: Adaptive Representation Adjustment and Parameter Fusion (ECCV 2024 · similarity 0.85)
- Prediction Error-based Classification for Class-Incremental Learning (ICLR 2024 · similarity 0.83)
- Compress to One Point: Neural Collapse for Pre-Trained Model-Based Class-Incremental Learning (AAAI 2025 · similarity 0.83)
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.
