Research map / Architectures and efficiency / RNNs and normalization
On Embeddings for Numerical Features in Tabular Deep Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 numerical-feature embeddings using periodic activations plus linear and ReLU layers | MLP · Gesture Phase | accuracy: 0.665 -> 0.700 | Helped |
| Add numerical-feature embeddings using periodic activations plus linear and ReLU layers | MLP · House 16H | RMSE: 3.109 -> 2.975 | Helped |
| Add target-aware piecewise linear embeddings followed by linear and ReLU layers | ResNet · California Housing | RMSE: 0.483 -> 0.425 | Helped |
| Add quantile piecewise linear embeddings followed by linear and ReLU layers | MLP · Facebook Comments Volume | RMSE: 5.616 -> 5.144 | Helped |
| Add linear and ReLU numerical-feature embedding layers | MLP · California Housing | RMSE: 0.486 -> 0.463 | Helped |
| Replace piecewise linear encoding with binary thermometer encoding (ablation) | MLP-Q · Churn Modelling | accuracy: 0.854 -> 0.815 | Hurt |
| Replace piecewise linear encoding with one-blob encoding (ablation) | MLP-Q · Gesture Phase | accuracy: 0.653 -> 0.613 | Hurt |
| Replace per-feature learned periodic embeddings with original Fourier features that mix features and use fixed coefficients (ablation) | MLP · Gesture Phase | accuracy: 0.632 -> 0.612 | Hurt |
| Add fixed-coefficient Periodic features while keeping original features | XGBoost · California Housing | RMSE: 0.436 -> 0.441 | Hurt |
| Add fixed-coefficient Periodic features while keeping original features | XGBoost · Higgs Small | accuracy: 0.724 -> 0.724 | No clear effect |
| Use PLE representations without traditional quantile preprocessing (ablation) | MLP-Q · California Housing | RMSE: 0.464 -> 0.463 | No clear effect |
| Use PLE representations without traditional quantile preprocessing (ablation) | MLP-T · California Housing | RMSE: 0.447 -> 0.447 | No clear effect |
Closest papers by meaning
- TabR: Tabular Deep Learning Meets Nearest Neighbors (ICLR 2024 · similarity 0.89)
- TabM: Advancing tabular deep learning with parameter-efficient ensembling (ICLR 2025 · similarity 0.89)
- Hybrid Autoencoders for Tabular Data: Leveraging Model-Based Augmentation in Low-Label Settings (NeurIPS 2025 · similarity 0.87)
- TabPrep: Closing the Feature Engineering Gap in Tabular Benchmarks (NeurIPS 2026 · similarity 0.87)
- Making Pre-trained Language Models Great on Tabular Prediction (ICLR 2024 · similarity 0.86)
- TabReD: Analyzing Pitfalls and Filling the Gaps in Tabular Deep Learning Benchmarks (ICLR 2025 · similarity 0.86)
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.