Research map / Architectures and efficiency / Neural architecture search
BLOX: Macro Neural Architecture Search Benchmark and Algorithms
Thomas Chun Pong Chau, Łukasz Dudziak, Hongkai Wen, Nicholas Donald Lane, Mohamed S Abdelfattah
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 |
|---|---|---|---|
| Fine-tune 10 epochs then retrain top models 200 epochs (FT10+FT200) instead of FT200 only, good teacher | Blockwise NAS · Blox | accuracy @ cost=40: 76.90 -> 77.66 | Improved |
| Search with only 10 fine-tuning epochs (FT10) instead of 200 (ablation) | Blockwise NAS · Blox | accuracy @ cost=40: 76.90 -> 73.47 | Hurt |
| Use bad teacher M5 instead of good teacher M1 with FT10 (ablation) | Blockwise NAS · Blox | accuracy @ cost=40: 73.47 -> 70.67 | Hurt |
| Retrain after bad-teacher search using iteratively chosen teacher | Blockwise NAS · Blox | accuracy @ cost=40: 70.67 -> 76.88 | Improved |
| Fine-tune with teacher chosen by training from scratch (iter 1) | Student models m1-m5 · Blox | accuracy (%): 74.89 -> 76.35 | Improved |
| Second iteration of teacher selection (iter 2, m4 as teacher) | Student model m4 · Blox | accuracy (%): 76.36 -> 76.67 | Improved |
| Second iteration of teacher selection | Student model m1 · Blox | accuracy (%): 76.35 -> 75.61 | Hurt |
| Blockwise NAS with fine-tuning versus conventional BRP-NAS | DONNA/blockwise · Blox | accuracy @ cost=40: 76.40 -> 77.66 | Improved |
| Regularized evolution versus BRP-NAS | conventional NAS · Blox | accuracy @ cost=40: 76.40 -> 76.10 | Hurt |
| Remove fine-tuning from DONNA (only block signatures) (ablation) | DONNA · Blox | search accuracy | No clear effect |
| Remove fine-tuning from LANA (ablation) | LANA · Blox | search accuracy | Hurt |
| Use bad teacher M5 rather than good teacher M1 for ranking models | Fine-tuning · Blox | rank correlation | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BRP-NAS — BRP-NAS: Prediction-based NAS using GCNs (NeurIPS 2020) · 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
- NAS-Bench-201: Extending the Scope of Reproducible Neural Architecture Search (ICLR 2020 · similarity 0.89)
- NAS-Bench-Suite: NAS Evaluation is (Now) Surprisingly Easy (ICLR 2022 · similarity 0.88)
- NATS-Bench: Benchmarking NAS Algorithms for Architecture Topology and Size (TPAMI 2021 · similarity 0.88)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.86)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.86)
- Towards Neural Architecture Search through Hierarchical Generative Modeling (ICML 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.
