Research map / Reinforcement learning / Policy gradient and actor-critic
TabNAS: Rejection Sampling for Neural Architecture Search on Tabular Datasets
Chengrun Yang, Gabriel Bender, Hanxiao Liu, Pieter-Jan Kindermans, Madeleine Udell, Yifeng Lu, Quoc V Le, Da Huang
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 |
|---|---|---|---|
| replace Abs Reward with rejection-based reward (ablation) | RL weight-sharing NAS · Criteo 3-layer | validation loss: 0.4461 -> 0.4455 | Improved |
| replace Abs Reward with rejection-based reward (ablation) | RL weight-sharing NAS · Criteo 4-layer | validation loss: 0.4452 -> 0.4449 | Improved |
| replace Abs Reward with rejection-based reward (ablation) | RL weight-sharing NAS · Criteo 5-layer | validation loss: 0.4448 -> 0.4445 | Improved |
| replace Abs Reward with rejection-based reward (ablation) | RL weight-sharing NAS · Aloi | balanced error: 0.0662 -> 0.0669 | No clear effect |
| replace Abs Reward with rejection-based reward (ablation) | RL weight-sharing NAS · Criteo 4-layer | hyperparameter settings finding reference: 3 out of 338 | Improved |
| replace MnasNet reward with rejection-based reward (ablation) | RL weight-sharing NAS · tabular search spaces | fraction of trials within 5% of target size: fewer than 1% | Improved |
| use larger number of Monte-Carlo samples in rejection estimate (ablation) | TabNAS · Criteo 3-layer | finding reference architecture: no fewer than 2,048 | Improved |
| use bottleneck architecture | feedforward network · Aloi | balanced error: 0.0662 -> 0.0669 | No clear effect |
| vary beta in Abs Reward (small versus large magnitude) (ablation) | Abs Reward NAS · Volkert 4-layer / toy example | feasibility and quality | Mixed |
| replace Abs Reward with rejection-based reward (ablation) | RL weight-sharing NAS · NATS-Bench size search space | accuracy | Improved |
| train SuperNet without updating the RL controller (ablation) | SuperNet one-shot weights · Criteo 3-layer | correlation of one-shot and stand-alone loss: 0.96 | No clear effect |
Closest papers by meaning
- pTNAS: Progressive Neural Architecture Search for Tabular Data (ICML 2026 · similarity 0.91)
- Surrogate NAS Benchmarks: Going Beyond the Limited Search Spaces of Tabular NAS Benchmarks (ICLR 2022 · similarity 0.89)
- Bridging the Gap between Sample-based and One-shot Neural Architecture Search with BONAS (NeurIPS 2020 · similarity 0.88)
- NAS-Bench-201: Extending the Scope of Reproducible Neural Architecture Search (ICLR 2020 · similarity 0.88)
- Revisiting Deep Learning Models for Tabular Data (NeurIPS 2021 · similarity 0.87)
- Insights from the Use of Previously Unseen Neural Architecture Search Datasets (CVPR 2024 · similarity 0.87)
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.
