Research map / Architectures and efficiency / Neural architecture search
Dynamic Ensemble of Low-Fidelity Experts: Mitigating NAS “Cold-Start”
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 dynamic low-fidelity expert ensembling to predictor training | Predictor-based NAS with GATES on NAS-Bench-201 · NAS-Bench-201, 1% training samples | Kendall's Tau: 0.7332$_{(0.0110)}$ -> 0.8244$_{(0.0081)}$ | Helped |
| Add dynamic low-fidelity expert ensembling to predictor training | Predictor-based NAS with GATES on NAS-Bench-301 · NAS-Bench-301, 1% training samples | Kendall's Tau: 0.4160$_{(0.0450)}$ -> 0.5529$_{(0.0135)}$ | Helped |
| Add dynamic low-fidelity expert ensembling to predictor training | LSTM predictor on NAS-Bench-201 · NAS-Bench-201, 1% training samples | Kendall's Tau: 0.5692$_{(0.0087)}$ -> 0.7835$_{(0.0062)}$ | Helped |
| Add dynamic low-fidelity expert ensembling to predictor training | LSTM predictor on NAS-Bench-301 · NAS-Bench-301, 1% training samples | Kendall's Tau: 0.4757$_{(0.0150)}$ -> 0.4805$_{(0.0083)}$ | No clear effect |
| Add dynamic low-fidelity expert ensembling to predictor training | LSTM predictor on NDS-ResNet · NDS-ResNet, 1% training samples | Kendall's Tau: 0.2549$_{(0.0087)}$ -> 0.7064$_{(0.0109)}$ | Helped |
| Add dynamic low-fidelity expert ensembling to predictor training | LSTM predictor on NDS-ResNeXt-A · NDS-ResNeXt-A, 1% training samples | Kendall's Tau: 0.3568$_{(0.0327)}$ -> 0.7753$_{(0.0010)}$ | Helped |
| Add dynamic low-fidelity expert ensembling to predictor training | LSTM predictor on MobileNet-V3 · MobileNet-V3, 1% training samples | Kendall's Tau: 0.7373$_{(0.0041)}$ -> 0.7698$_{(0.0018)}$ | Helped |
| Average expert outputs instead of dynamically weighting them (ablation) | DELE on NAS-Bench-201 · NAS-Bench-201, 1% training samples | Kendall's Tau: 0.6936$_{0.0038}$ vs 0.7835$_{(0.0062)}$ | Hurt |
| Use one learned set of weights for all architectures instead of dynamic weights (ablation) | DELE on NAS-Bench-201 · NAS-Bench-201, 1% training samples | Kendall's Tau: 0.7442$_{(0.0031)}$ vs 0.7835$_{(0.0062)}$ | Hurt |
| Average expert outputs instead of dynamically weighting them (ablation) | DELE on NDS-ResNet · NDS-ResNet, 1% training samples | Kendall's Tau: 0.5789$_{(0.0145)}$ vs 0.7064$_{(0.0109)}$ | Hurt |
| Use one learned set of weights for all architectures instead of dynamic weights (ablation) | DELE on NDS-ResNet · NDS-ResNet, 1% training samples | Kendall's Tau: 0.6794$_{(0.0174)}$ vs 0.7064$_{(0.0109)}$ | Hurt |
| Use one learned set of weights for all architectures instead of dynamic weights (ablation) | DELE on NDS-ResNeXt-A · NDS-ResNeXt-A, 1% training samples | Kendall's Tau: 0.7694$_{(0.0062)}$ vs 0.7753$_{(0.0010)}$ | Hurt |
Closest papers by meaning
- How Powerful are Performance Predictors in Neural Architecture Search? (NeurIPS 2021 · similarity 0.92)
- Stronger NAS with Weaker Predictors (NeurIPS 2021 · similarity 0.91)
- NAS-Bench-201: Extending the Scope of Reproducible Neural Architecture Search (ICLR 2020 · similarity 0.90)
- NAS-Bench-x11 and the Power of Learning Curves (NeurIPS 2021 · similarity 0.89)
- Speedy Performance Estimation for Neural Architecture Search (NeurIPS 2021 · similarity 0.89)
- Encodings for Prediction-based Neural Architecture Search (ICML 2024 · similarity 0.88)
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.