Research map / Learning theory and algorithms / Prediction and decision losses
Feature Selection with Annealing for Computer Vision and Big Data Learning
Adrian Barbu, Yiyuan She, Liangjing Ding, Gary Gramajo
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 FSA instead of L1-penalized logistic regression for feature selection | linear classifier · Simulated classification (clean, N=1000, M=1000, k=10) | training time (s): 434 -> 0.06 | Improved |
| use FSA instead of Logitboost for feature selection | linear classifier · Simulated classification (clean, N=300, M=1000, k=10) | AUC: .950 -> .992 | Improved |
| use FSA instead of L1 logistic regression with few observations | linear classifier · Simulated classification (clean, N=300, M=1000, k=10) | AUC: .915 -> .992 | Improved |
| use Lorenz loss instead of logistic loss in FSA under label noise | FSA · Simulated classification (noisy labels, N=1000, M=1000, k=10) | percent correctly detected: 92.5 -> 98.5 | Improved |
| use FSA instead of Logitboost on equally correlated large data | FSA with Lorenz loss · Large correlated data (M=10^5, k=500) | training time: at least 10 times faster | Improved |
| use FSA instead of lasso in regression | linear regression · Simulated regression (N=1000, M=1000, k=30) | RMSE: 2.79 -> 1.02 | Improved |
| use FSA instead of simulated annealing wrapper | linear regression · Simulated regression (N=1000, M=1000, k=30) | time (s): 2730 -> 0.18 | Improved |
| add piecewise linear bins (2 bins) to FSA (ablation) | FSA on Gisette · Gisette (test) | test error (%): 1.83 -> 1.88 | Hurt |
| train on training plus validation sets | linear FSA · Dexter (test) | test error (%): 10.3 -> 8.55 | Improved |
| use all features (likelihood plus prior) in FSARank instead of likelihood only (ablation) | FSARank · Hopkins 155 (all sequences combined, test) | misclassification rate (%): 0.90 -> 0.44 | Improved |
| use all features instead of prior features only in FSARank (ablation) | FSARank · Hopkins 155 (all sequences combined, test) | misclassification rate (%): 2.52 -> 0.44 | Improved |
| use FSARank instead of RankBoost with 40 vs about 100 features | motion segmentation selection · Hopkins 155 (all sequences combined, test) | misclassification rate (%): 1.54 -> 0.44 | Improved |
Closest papers by meaning
- Heuristic Search Algorithm for Dimensionality Reduction Optimally Combining Feature Selection and Feature Extraction (AAAI 2019 · similarity 0.85)
- Best Subset Selection: Optimal Pursuit for Feature Selection and Elimination (ICML 2025 · similarity 0.85)
- Learning Sparse Classifiers: Continuous and Mixed Integer Optimization Perspectives (JMLR 2021 · similarity 0.85)
- On the Efficiency of ERM in Feature Learning (NeurIPS 2024 · similarity 0.84)
- SAND: One-Shot Feature Selection with Additive Noise Distortion (ICML 2025 · similarity 0.83)
- Differentially Private High-dimensional Variable Selection via Integer Programming (NeurIPS 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.
