Research map / Architectures and efficiency / Neural architecture search
CATE: Computation-aware Neural Architecture Encoding with Transformers
Shen Yan, Kaiqiang Song, Fei Liu, Mi Zhang
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 |
|---|---|---|---|
| Combine CATE-DNGO with local search | CATE-DNGO · NAS-Bench-101 and NAS-Bench-301 | test error (%): Prev. SOTA: 5.92 -> 5.88; 5.35 -> 5.28 | Improved |
| Use a larger query budget | CATE-DNGO-LS · DARTS search space (CIFAR-10) | average test error (%): 2.55 ± 0.08 -> 2.46 ± 0.05 | Improved |
| Use an indirect dependency mask to capture long-range dependencies (ablation) | CATE-DNGO · NAS-Bench-101 and NAS-Bench-301 | test error (%): 6.03 -> 5.94; 5.35 -> 5.30 | Improved |
| Use unidirectional rather than bidirectional encoding (ablation) | CATE-DNGO-LS · NAS-Bench-101 and NAS-Bench-301 | test error (%): 5.89 -> 5.88; 5.30 -> 5.28 | Improved |
| Use a smaller computational-neighborhood threshold δ (ablation) | CATE-DNGO · NAS-Bench-101 | test error (%): δ=2×10^6, K=2: 5.94; δ=8×10^6, K=4: 6.11 | Mixed |
| Vary the neighborhood size K (ablation) | CATE-DNGO · NAS-Bench-101 | test error (%): 5.94 to 6.11 | No clear effect |
| Increase the number of cross-attention Transformer blocks and feed-forward dimension (ablation) | CATE-DNGO · NAS-Bench-101 | test error (%): 6.07 -> 5.94 | Improved |
| Increase the corruption rate from 20% to 30% (ablation) | CATE · NAS-Bench-101 and NAS-Bench-301 | test error (%): 5.88 -> 5.93; 5.28 -> 5.29 | Hurt |
| Increase the corruption rate from 15% to 20% (ablation) | CATE · NAS-Bench-101 | test error (%): 5.89 -> 5.88 | Improved |
| Use a computation-aware encoding in the outside search space | CATE predictor · NAS-Bench-101 outside search space | validation error | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- BANANAS — BANANAS: Bayesian Optimization with Neural Architectures for Neural Architecture Search (AAAI 2021) · better on 2 of 2 shared results
- DARTS — DARTS-: Robustly Stepping out of Performance Collapse Without Indicators (ICLR 2021) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Generic Graph-Based Neural Architecture Encoding Scheme With Multifaceted Information (TPAMI 2022 · similarity 0.88)
- Towards modular and programmable architecture search (NeurIPS 2019 · similarity 0.87)
- Training-Free Transformer Architecture Search With Zero-Cost Proxy Guided Evolution (TPAMI 2024 · similarity 0.87)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.87)
- Encodings for Prediction-based Neural Architecture Search (ICML 2024 · similarity 0.86)
- Towards Neural Architecture Search through Hierarchical Generative Modeling (ICML 2024 · similarity 0.85)
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.
