Research map / Label-efficient and robust learning / Recognition and long-tail
Breaking the Glass Ceiling for Embedding-Based Classifiers for Large Output Spaces
Chuan Guo, Ali Mousavi, Xiang Wu, Daniel N Holtmann-Rice, Satyen Kale, Sashank Reddi, Sanjiv Kumar
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 |
|---|---|---|---|
| add GLaS regularizer (ablation) | embedding-based model · AmazonCat-13K | P@1: 92.34 -> 94.21 | Improved |
| GLaS instead of spread-out regularizer (ablation) | embedding-based model · AmazonCat-13K | P@1: 93.34 -> 94.21 | Improved |
| GLaS instead of Gravity regularizer (ablation) | embedding-based model · AmazonCat-13K | P@1: 93.42 -> 94.21 | Improved |
| add GLaS regularizer (ablation) | embedding-based model · AmazonCat-13K | generalization gap: 6.89 -> 4.56 | Improved |
| add GLaS regularizer (ablation) | embedding-based model · Amazon-670K | test accuracy: 44.53 -> 46.32 | Improved |
| add input dropout (rho=0.8) (ablation) | embedding-based model · AmazonCat-13K | P@1: 93.39 -> 94.21 | Improved |
| use nonlinear ReLU embedding instead of linear (ablation) | embedding-based model · AmazonCat-13K | P@1: 91.77 -> 94.21 | Improved |
| increase embedding size from 256 to 1024 (ablation) | embedding-based model · AmazonCat-13K | P@1: 93.24 -> 94.21 | Improved |
| increase batch size from 1024 to 4096 (ablation) | embedding-based model · AmazonCat-13K | P@1: 94.04 -> 94.21 | No clear effect |
| regularization weight lambda 10 vs 1 (ablation) | embedding-based model · AmazonCat-13K | P@1: 93.68 -> 94.21 | Improved |
| GLaS embedding model instead of SLEEC | embedding-based extreme classification · Amazon-670K | PSP@1: 20.62 -> 38.94 | Improved |
| GLaS embedding model vs best other methods | extreme classification · Delicious-200K and EURLex-4K | P@1: 46.4 vs 47.85 (SLEEC); 77.5 vs 83.83 (PPD-Sparse) | Hurt |
Closest papers by meaning
- Adaptive Regularization for Large-Scale Sparse Feature Embedding Models (ICLR 2026 · similarity 0.87)
- Asymptotics of $\ell_2$ Regularized Network Embeddings (NeurIPS 2022 · similarity 0.86)
- Benign Overfitting and Grokking in ReLU Networks for XOR Cluster Data (ICLR 2024 · similarity 0.85)
- MaxSup: Overcoming Representation Collapse in Label Smoothing (NeurIPS 2025 · similarity 0.85)
- Residual Connections and Normalization Can Provably Prevent Oversmoothing in GNNs (ICLR 2025 · similarity 0.84)
- Curriculum-Enhanced Residual Soft An-Isotropic Normalization for Over-Smoothness in Deep GNNs (AAAI 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.
