Research map / Architectures and efficiency / Neural architecture search
A Generic Graph-based Neural Architecture Encoding Scheme for Predictor-based NAS
Xuefei Ning, Yin Zheng, Tianchen Zhao, Yu Wang, Huazhong Yang
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 MSE regression loss with hinge pairwise ranking loss (ablation) | GATES predictor · NAS-Bench-101 (0.1% train) | Kendall's Tau: 0.5425 -> 0.7789 | Improved |
| replace MSE regression loss with hinge pairwise ranking loss (ablation) | GATES predictor · NAS-Bench-201 (10% train) | Kendall's Tau: 0.8042 -> 0.8802 | Improved |
| replace MSE regression loss with hinge pairwise ranking loss (ablation) | LSTM predictor · NAS-Bench-201 (1% train) | Kendall's Tau: 0.4405 -> 0.5550 | Improved |
| replace GCN encoder with GATES encoder | predictor · NAS-Bench-101 (0.05% train) | Kendall's Tau: 0.5343 -> 0.7634 | Improved |
| replace GCN encoder with GATES encoder | predictor · NAS-Bench-101 (100% train) | Kendall's Tau: 0.8952 -> 0.9030 | Improved |
| replace LSTM encoder with GATES encoder | predictor · NAS-Bench-201 (1% train) | Kendall's Tau: 0.5550 -> 0.7401 | Improved |
| replace hinge pairwise loss with ListMLE listwise loss (ablation) | GATES predictor · NAS-Bench-101 (0.1% train) | Kendall's Tau: 0.7789 -> 0.7604 | No clear effect |
| replace hinge pairwise loss with pairwise comparator loss (ablation) | GATES predictor · NAS-Bench-101 (0.1% train) | Kendall's Tau: 0.7789 -> 0.7622 | No clear effect |
| replace GATES predictor with GCN in NAS flow, using GATES-powered predictor NAS versus random search | predictor-based NAS · NAS-Bench-101 search | median architectures evaluated: 220400 -> 400 | Improved |
| replace regularized evolution with GATES-powered predictor-based NAS | predictor-based NAS · NAS-Bench-101 search | median architectures evaluated: 23700 -> 400 | Improved |
| increase sample ratio r (n too small or too large) (ablation) | GATES-powered predictor NAS with EA · NAS-Bench-101 search | architectures evaluated to best accuracy | Mixed |
| increase number of encoder layers with ranking loss (ablation) | GCN encoder · NAS-Bench-101 / NAS-Bench-201 | Kendall's Tau | Hurt |
Closest papers by meaning
- A Generic Graph-Based Neural Architecture Encoding Scheme With Multifaceted Information (TPAMI 2022 · similarity 0.93)
- Encodings for Prediction-based Neural Architecture Search (ICML 2024 · similarity 0.90)
- Do Not Train It: A Linear Neural Architecture Search of Graph Neural Networks (ICML 2023 · similarity 0.90)
- Understanding Architectures Learnt by Cell-based Neural Architecture Search (ICLR 2020 · similarity 0.89)
- TNASP: A Transformer-based NAS Predictor with a Self-evolution Framework (NeurIPS 2021 · similarity 0.89)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · 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.
