Research map / Learning theory and algorithms / Clustering
Regularized K-means Through Hard-Thresholding
Jakob Raymaekers, Ruben H. Zamar
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 hard-thresholding penalty instead of ridge penalty (ablation) | regularized K-means · Simulation n=80 p=1000 K=4, mu=0.6 | ARI: 0.44 -> 0.86 | Improved |
| use hard-thresholding penalty instead of lasso penalty (ablation) | regularized K-means · Simulation n=80 p=1000 K=4, mu=0.6 | ARI: 0.69 -> 0.86 | Improved |
| use hard-thresholding penalty instead of group-lasso (ablation) | regularized K-means · Simulation n=80 p=1000 K=4, mu=0.6 | ARI: 0.8 -> 0.86 | Improved |
| add any penalty on cluster centers (HT vs classical K-means) (ablation) | K-means · Simulation n=80 p=1000 K=4, mu=0.6 | ARI: 0.35 -> 0.86 | Improved |
| regularization at weakly separated clusters (ablation) | regularized K-means · Simulation n=80 p=1000 K=4, mu=0.4 | ARI: 0.08 -> 0.15 | No clear effect |
| use HT K-means instead of Sparse K-means | K-means with feature selection · Simulation n=80 p=1000 K=4, mu=0.6 | ARI: 0.66 -> 0.8 | Improved |
| use HT K-means instead of Reg K-means | K-means with feature selection · Simulation n=80 p=1000 K=4, mu=0.6 | ARI: 0.36 -> 0.8 | Improved |
| select lambda with AIC instead of stability criteria (ablation) | HT K-means · Simulation n=80 p=1000 K=4, mu=0.6 | ARI: 0.47 -> 0.8 | Improved |
| select lambda with AIC instead of BIC (ablation) | HT K-means · p=1000, mu=0.5 | ARI: 0.21 -> 0.26 | Improved |
| select lambda with BIC instead of AIC (number of variables) (ablation) | HT K-means · p=1000, mu=0.8 | selected variables (true 50): 90.42 -> 50.01 | Improved |
| use AIC instead of gap method for lambda (ablation) | HT K-means · p=1000, mu=0.5 | ARI: 0.26 -> 0.28 (gap1 is higher) | Hurt |
| use AIC/BIC instead of stability for computation time (ablation) | lambda selection · p=1000, mu=0.6 | seconds: 917.88 -> 0.23 | Improved |
Closest papers by meaning
- K-means clustering using random matrix sparsification (ICML 2018 · similarity 0.85)
- Robust k-means: a Theoretical Revisit (NeurIPS 2016 · similarity 0.85)
- Fast k-means with accurate bounds (ICML 2016 · similarity 0.85)
- The Complexity of k-Means Clustering when Little is Known (ICML 2022 · similarity 0.84)
- Near-Linear Time Approximation Algorithms for k-means with Outliers (ICML 2024 · similarity 0.84)
- Universal Weak Coreset (AAAI 2024 · similarity 0.82)
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.
