Research map / Architectures and efficiency / Neural architecture search
NSGANetV2: Evolutionary Multi-Objective Surrogate-Assisted Neural Architecture Search
Zhichao Lu, Kalyanmoy Deb, Erik Goodman, Wolfgang Banzhaf, Vishnu Naresh Boddeti
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 |
|---|---|---|---|
| use online surrogate on near-front architectures instead of offline surrogates | MSuNAS · ImageNet | architectures sampled / rank correlation: 350 vs 16,000 (OnceForAll) architectures; 0.9 vs 0.476 rank correlation | Improved |
| use Adaptive Switching over MLP, CART, RBF, GP surrogates instead of a single surrogate (ablation) | MSuNAS accuracy predictor · ImageNet, CIFAR-10, CIFAR-100 | Spearman rank-order correlation | Improved |
| use RBF surrogate rather than MLP, CART or GP (ablation) | MSuNAS · ImageNet | hypervolume | Improved |
| higher surrogate rank-order correlation (ablation) | MSuNAS surrogate · ImageNet | hypervolume | Improved |
| initialize search from distribution of architectures found for related objectives (MAdds) instead of random (ablation) | MSuNAS targeting accuracy and CPU latency · ImageNet | hypervolume | Improved |
| use MSuNAS instead of RL or EA methods | NAS search · CIFAR-10 / ImageNet | search speedup: 20x faster; at least 3x faster than PNAS | Improved |
| use MSuNAS instead of NSGANet and random search | bi-objective NAS · ImageNet and CIFAR-10 | hypervolume per architecture evaluated: 2 - 5x faster | Improved |
| use NSGANetV2-s instead of MobileNetV3 | mobile ImageNet models · ImageNet val | top-1 accuracy (%): 75.2 -> 77.4 | Improved |
| use NSGANetV2-xl instead of EfficientNet-B1 | ImageNet models · ImageNet val | top-1 accuracy (%): 78.8 -> 80.4 | Improved |
| search directly on each target dataset instead of transferring architectures | non-standard datasets · CINIC-10, STL-10, Flowers102 | accuracy | Improved |
| optimize a scalarized single objective with 600M MAdds target | MSuNAS · ImageNet val | top-1 accuracy (%): 78.8 (EfficientNet-B1) -> 79.56 with 596M MAdds | Improved |
Closest papers by meaning
- NetAdaptV2: Efficient Neural Architecture Search With Fast Super-Network Training and Architecture Optimization (CVPR 2021 · similarity 0.89)
- Learning Where to Look – Generative NAS Is Surprisingly Efficient (ECCV 2022 · similarity 0.89)
- Neural Architecture Transfer (TPAMI 2021 · similarity 0.88)
- MOTE-NAS: Multi-Objective Training-based Estimate for Efficient Neural Architecture Search (NeurIPS 2024 · similarity 0.87)
- EG-NAS: Neural Architecture Search with Fast Evolutionary Exploration (AAAI 2024 · similarity 0.87)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.87)
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.
