Research map / Architectures and efficiency / Neural architecture search
Stronger NAS with Weaker Predictors
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 |
|---|---|---|---|
| progressively add weak predictors and guided sampling iterations (ablation) | WeakNAS · NAS-Bench-101 CIFAR10 | Test Acc.(%): 93.42 -> 94.26 | Helped |
| sample uniformly from Top 100 instead of using exponential-decay probabilities (ablation) | WeakNAS · NAS-Bench-101 CIFAR10 | Test Acc.(%): 93.96 -> 94.25 | Helped |
| sample uniformly from Top 100 instead of using linear-decay probabilities (ablation) | WeakNAS · NAS-Bench-101 CIFAR10 | Test Acc.(%): 94.06 -> 94.25 | Helped |
| expand the sampling pool from Top 100 to Top 1000 (ablation) | WeakNAS · NAS-Bench-101 CIFAR10 | Test Acc.(%): 94.25 -> 94.10 | Hurt |
| narrow the sampling pool from Top 100 to Top 10 (ablation) | WeakNAS · NAS-Bench-101 CIFAR10 | Test Acc.(%): 94.25 -> 94.24 | No clear effect |
| replace the simple sampling rule with ensemble uncertainty and Expected Improvement (ablation) | WeakNAS · NAS-Bench-101 CIFAR10 | Test Acc.(%): 94.25 -> 94.12 | Hurt |
| add CATE architecture embedding to adjacency-matrix encoding (ablation) | WeakNAS · NAS-Bench-101 CIFAR10 | Test Acc.(%): 94.10 -> 94.19 | Helped |
| reduce the initial sample count from 100 to 10 (ablation) | WeakNAS · NAS-Bench-101 CIFAR10, 1000 queries | Test Acc.(%): 94.25 -> 94.14 | Hurt |
| increase the initial sample count from 100 to 200 (ablation) | WeakNAS · NAS-Bench-101 CIFAR10, 1000 queries | Test Acc.(%): 94.25 -> 94.19 | Hurt |
| uniformly sample from Top 50 instead of Top 10 (ablation) | WeakNAS · ImageNet MobileNet Search Space, supernet proxy | SuperNet Test Acc.(%): 79.1552 -> 79.2005 | Helped |
| expand the sampling pool from Top 50 to Top 2500 (ablation) | WeakNAS · ImageNet MobileNet Search Space, supernet proxy | SuperNet Test Acc.(%): 79.2005 -> 79.1274 | Hurt |
Closest papers by meaning
- How Powerful are Performance Predictors in Neural Architecture Search? (NeurIPS 2021 · similarity 0.91)
- Semi-Supervised Neural Architecture Search (NeurIPS 2020 · similarity 0.91)
- Dynamic Ensemble of Low-Fidelity Experts: Mitigating NAS “Cold-Start” (AAAI 2023 · similarity 0.91)
- NAS-Bench-360: Benchmarking Neural Architecture Search on Diverse Tasks (NeurIPS 2022 · similarity 0.90)
- Surrogate NAS Benchmarks: Going Beyond the Limited Search Spaces of Tabular NAS Benchmarks (ICLR 2022 · similarity 0.90)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.89)
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.